Surface defect visual detection method and system for stainless steel EP tube for semiconductor processing

Abstract

The embodiment of the application provides a surface defect visual detection method and system for a stainless steel EP pipe for semiconductor processing, relates to the technical field of surface defect visual detection of the stainless steel EP pipe for semiconductor processing, and comprises the following steps: driving a detection probe to move along the axis direction of a pipe material to be detected, and acquiring image data of the inner wall of the pipe material to be detected to form an image sequence; for each frame of image in the image sequence, identifying a potential abnormal area where the image is different from a preset normal surface; acquiring a plurality of characteristic information of the potential abnormal area, and preliminarily classifying the potential abnormal area to obtain a defect candidate area; checking the spatial position continuity and characteristic stability of the defect candidate area in continuous images to obtain a behavior mode discrimination result; and confirming the defect of the inner wall of the pipe material to be detected based on the behavior mode discrimination result and outputting the type, size and position information of the defect. The application can improve the accuracy and reliability of the inner surface defect detection of the stainless steel EP pipe for semiconductor processing.

Description

Technical Field

[0001] This application relates to the field of visual inspection technology for surface defects of stainless steel EP tubes used in semiconductor processing, and more specifically, to a method and system for visual inspection of surface defects of stainless steel EP tubes used in semiconductor processing. Background Technology

[0002] In related technologies, the inner surface quality of electropolished (EP) stainless steel tubes in semiconductor clean manufacturing directly determines the reliability of ultra-high purity fluid transport systems. Micron-level defects can become sources of contamination, affecting chip yield. Therefore, automated visual inspection is commonly used in production lines to replace manual inspection. However, in actual production, EP tubes often exhibit slight axial curvature or cross-sectional ellipticity, causing the detection probe to be unable to maintain absolute coaxiality with the inner wall as it moves within the tube. This results in uneven distribution of the illumination spot and camera field of view, leading to dynamic lighting changes such as localized overexposure and shadows. Under these non-ideal conditions, the characteristic signals of low-contrast, minute defects such as micron-level scratches and shallow pits are easily drowned out by background noise, leading to missed detections. Furthermore, the uneven illumination pattern caused by probe eccentricity changes with probe movement, making it difficult to establish a stable benchmark to distinguish defects from artifacts, resulting in numerous false alarms and missed detections. In addition, ultra-cleanliness requirements necessitate the detection of transparent / semi-transparent trace residues such as polishing fluid residue and extremely thin oxide layers. These residues have weak light scattering and reflection characteristics and are almost completely masked by noise against an uneven illumination background, making them difficult to identify using traditional methods. Although these residues are not macroscopic defects, they may subsequently release particles that affect the performance of semiconductor devices. Identifying such extremely low-contrast residues under these complex imaging conditions presents a significant challenge for current visual inspection systems, and existing technologies urgently need improvement. Summary of the Invention

[0003] This application aims to address at least one of the technical problems existing in the prior art. To this end, this application proposes a visual inspection method and system for surface defects of stainless steel EP tubes used in semiconductor processing, aiming to solve the technical problem that, during the visual inspection of stainless steel EP tubes used in semiconductor processing, uneven illumination caused by probe eccentricity makes it difficult to effectively identify and distinguish physical defects with low contrast, shallow depth, or small size, as well as transparent or semi-transparent trace surface residues. In a first aspect, embodiments of this application provide a method for visually inspecting surface defects in stainless steel EP tubes used in semiconductor processing, including: The detection probe is driven to move along the axial direction of the pipe to be inspected, and image data of the inner wall of the pipe to be inspected is acquired by the imaging component at the front end of the detection probe to form an image sequence; For each frame of the image sequence, identify potential abnormal regions where the image differs from a preset normal surface; The potential abnormal region is obtained by acquiring multiple feature information, including at least one of local contrast, texture detail, edge intensity, and optical response features; Based on the image data and the various feature information, the potential abnormal regions are preliminarily classified to obtain defect candidate regions; The spatial continuity and feature stability of the defect candidate regions in continuous images are checked, the behavior patterns of the defect candidate regions are determined, and the behavior pattern determination results are obtained. Based on the behavior pattern discrimination result, the actual defect area of ​​the inner wall of the pipe to be inspected is obtained to confirm the defect of the inner wall of the pipe to be inspected, and the type, size and location information of the defect are output. According to some embodiments of this application, obtaining various feature information of the potential abnormal region includes: Analyze the brightness difference between the potential anomalous region and the surrounding background to obtain the local contrast of the potential anomalous region; Evaluate the uniformity and complexity of the surface texture of the potential anomaly region to obtain the texture details of the potential anomaly region; Identify the boundary clarity, shape contour, and size information of the potential anomaly region to obtain the edge strength of the potential anomaly region; The polarization information of the potential anomalous region in different polarization states is extracted to obtain the optical response characteristics of the potential anomalous region, wherein the polarization information includes the degree of polarization and the polarization angle. According to some embodiments of this application, analyzing the brightness difference between the potential anomalous region and the surrounding background to obtain the local contrast of the potential anomalous region includes: Obtain the brightness values ​​of the pixels within the potential abnormal region to obtain the maximum and minimum brightness values; Calculate the average brightness value and standard deviation of the brightness values ​​of the pixels within the potential abnormal region; Obtain the brightness value of the surrounding background of the potential abnormal region; The brightness difference between the potential anomaly region and its surrounding background is analyzed based on the average brightness value, the standard deviation, and the brightness value of the background surrounding the potential anomaly region to obtain the local contrast of the potential anomaly region. According to some embodiments of this application, evaluating the uniformity and complexity of the surface texture of the potential anomalous region to obtain the texture details of the potential anomalous region includes: The eigenvalues ​​of the gray-level co-occurrence matrix of the potential abnormal region are obtained to obtain the energy value and entropy value; The uniformity and complexity of the surface texture of the potential anomaly region are evaluated based on the energy value and the entropy value to obtain the texture details of the potential anomaly region, wherein the energy value and the uniformity are positively correlated, and the entropy value and the complexity are positively correlated. According to some embodiments of this application, the step of identifying the boundary sharpness, shape contour, and size information of the potential anomalous region to obtain the edge strength of the potential anomalous region includes: The boundaries of the potential abnormal regions are identified based on a preset edge detection algorithm; Based on connected component analysis, the area, perimeter, aspect ratio, roundness, and convexity of each potential anomaly region are calculated. The edge strength of the potential anomaly region is obtained by identifying the boundary clarity, shape contour, and size information of the potential anomaly region based on the boundary, area, perimeter, aspect ratio, roundness, and convexity. According to some embodiments of this application, extracting polarization information of the potential anomalous region in different polarization states to obtain the optical response characteristics of the potential anomalous region includes: Acquire images of the potential anomaly region in different polarization directions; The degree of polarization and the polarization angle are calculated from images of the potential anomaly region in different polarization directions. Based on the degree of polarization and the angle of polarization, information about the potential anomalous region in different polarization states is extracted to obtain the optical response characteristics of the potential anomalous region. According to some embodiments of this application, the preliminary classification of the potential abnormal region based on the image data and the various feature information to obtain defect candidate regions includes: Based on the preset threshold rules and the image data, an initial candidate region is obtained, wherein the preset threshold rules represent excluding regions in the potential abnormal region that are less than a preset local contrast and excluding regions in the potential abnormal region that are greater than a preset area. The initial candidate region is classified according to the various feature information to obtain the classification result, which includes suspected physical defects, suspected transparent residues, or suspected light artifacts. Based on the classification results, candidate defect regions are obtained. According to some embodiments of this application, checking the spatial continuity and feature stability of the defect candidate region in a continuous image includes: Calculate the average displacement vector of each of the defect candidate regions over N consecutive frames; Based on the average displacement vector and the preset displacement vector, check the spatial continuity of the defect candidate region in the continuous image; The multi-dimensional feature values ​​of each of the defect candidate regions are obtained in N consecutive frames to obtain the rate of change of the multi-dimensional feature values ​​within N frames. The multi-dimensional feature values ​​include local contrast, texture entropy, and polarization degree. Based on the rate of change and a preset threshold, the feature stability of the defect candidate region in continuous images is determined. According to some embodiments of this application, the driving detection probe moves along the axial direction of the pipe to be inspected, and acquires image data of the inner wall of the pipe to be inspected based on the imaging component at the front end of the detection probe to form an image sequence, including: During the movement of the detection probe, the distance between the detection probe and the inner wall of the pipe to be inspected, as well as the local curvature information of the inner wall of the pipe to be inspected, are measured in real time to obtain distance information and curvature information; Based on the distance information and the curvature information, the orientation of the detection probe is dynamically adjusted so that the imaging component at the front end of the detection probe is kept at a preset stable position relative to the inner wall of the pipe to be inspected. The preset stable position includes a vertical position and a centered position. While keeping the imaging component at the front end of the detection probe in a preset stable position relative to the inner wall of the pipe to be inspected, image data of the inner wall of the pipe to be inspected is acquired by the imaging component at the front end of the detection probe to form an image sequence. Secondly, this application also discloses a visual inspection system for surface defects of stainless steel EP tubes used in semiconductor processing, including: The first acquisition module is used to drive the detection probe to move along the axial direction of the pipe to be inspected, and to acquire image data of the inner wall of the pipe to be inspected according to the imaging component at the front end of the detection probe, so as to form an image sequence. The identification module is used to identify potential abnormal regions in each frame of the image sequence that differ from a preset normal surface. The second acquisition module is used to acquire multiple feature information of the potential abnormal region, the feature information including at least one of local contrast, texture details, edge intensity and optical response features; The classification module is used to perform preliminary classification of the potential abnormal regions based on the image data and the various feature information to obtain defect candidate regions; The discrimination module is used to check the spatial continuity and feature stability of the defect candidate region in continuous images, discriminate the behavior pattern of the defect candidate region, and obtain the behavior pattern discrimination result. The output module is used to determine the area of ​​the actual defect in the inner wall of the pipe under inspection based on the behavior pattern discrimination result, so as to confirm the defect in the inner wall of the pipe under inspection, and output the type, size and location information of the defect. The technical solution according to the embodiments of this application has at least the following beneficial effects: The visual inspection method for surface defects of stainless steel EP tubes for semiconductor processing disclosed in this application involves driving a detection probe to move along the axial direction of the tube to be inspected and acquiring inner wall image data to form an image sequence. For each frame of the image sequence, potential abnormal regions that differ from a preset normal surface are identified. Based on this, multiple feature information of the potential abnormal regions is acquired, including at least one of local contrast, texture detail, edge intensity, and optical response features. Subsequently, the potential abnormal regions are preliminarily classified based on the image data and multiple feature information to obtain defect candidate regions. Further, the spatial continuity and feature stability of the defect candidate regions in continuous images are checked, and the behavior pattern of the defect candidate regions is judged to obtain the behavior pattern discrimination result. Finally, based on the behavior pattern discrimination result, the actual defect region of the inner wall of the tube to be inspected is obtained, and the type, size, and location information of the defect are output. This method effectively solves the problems of probe eccentricity, uneven illumination, and changes in observation angle caused by the bending of the EP tube axis or the ellipticity of the cross-section in the prior art. By acquiring multi-dimensional feature information, particularly by introducing optical response features, this application can effectively identify and distinguish physical defects with low contrast, shallow depth, or small size that are difficult to detect using traditional methods, as well as transparent or translucent trace surface residues. Simultaneously, by judging the spatial continuity and feature stability of defect candidate regions in consecutive images, it can effectively distinguish between real defects and artifacts caused by illumination changes, significantly reducing false alarms and false negatives. Therefore, the method of this application can significantly improve the accuracy and reliability of detecting internal surface defects in stainless steel EP tubes used in semiconductor processing. Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0004] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application. Figure 1 This is a flowchart illustrating a method for visually inspecting surface defects in stainless steel EP tubes used in semiconductor processing, provided in one embodiment of this application. Figure 2 This is a schematic diagram of a visual inspection system for surface defects of stainless steel EP tubes used in semiconductor processing, provided in one embodiment of this application. Detailed Implementation

[0005] To make the objectives, technical methods, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. It should be noted that the meaning of "multiple" (or "more than") in the description of the embodiments of this application refers to two or more, and "greater than," "less than," "exceeding," etc. are understood to exclude the number itself, while "above," "below," "within," etc. are understood to include the number itself. If "first," "second," etc. are used in the description, they are only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated or the order of the technical features indicated. In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: the existence of a alone, the existence of b alone, the existence of c alone, the simultaneous existence of a and b, the simultaneous existence of a and c, the simultaneous existence of b and c, or the simultaneous existence of a, b, and c, where a, b, and c can be single or multiple. In the description of this application, unless otherwise expressly defined, terms such as "setup," "installation," and "connection" should be interpreted broadly, and those skilled in the art can reasonably determine the specific meaning of the above terms in this application in conjunction with the specific content of the technical solution. The visual inspection method for surface defects of stainless steel EP tubes used in semiconductor processing provided in this application embodiment can be applied to a terminal, a server, or software running on either a terminal or a server. In some embodiments, the terminal can be a smartphone, tablet, laptop, desktop computer, etc.; the server can be configured as an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms; the software can be an application that implements the visual inspection method for surface defects of stainless steel EP tubes used in semiconductor processing, etc., but is not limited to the above forms. This application can be applied to numerous general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics devices, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This application can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via communication networks. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices. It should be noted that in various specific embodiments of this invention, when processing is required based on data related to the characteristics of an object (e.g., user attributes or sets of attribute information), permission or consent from the corresponding object is obtained first, and the collection, use, and processing of this data comply with relevant laws and standards. Furthermore, when the embodiments of the present invention need to obtain the attribute information of an object, they will obtain the separate permission or separate consent of the corresponding object through pop-up windows or redirection to a confirmation page. After obtaining the separate permission or separate consent of the corresponding object, they will then obtain the relevant data of the object necessary for the embodiments of the present invention to operate normally. See Figure 1 , Figure 1 This is a schematic flowchart illustrating a method for visually inspecting surface defects in stainless steel EP tubes used in semiconductor processing, according to an embodiment of this application. The method for visually inspecting surface defects in stainless steel EP tubes used in semiconductor processing provided in this embodiment includes, but is not limited to, steps S110 to S160, which will be described in detail below. Step S110: Drive the detection probe to move along the axial direction of the pipe to be inspected, and acquire image data of the inner wall of the pipe to be inspected based on the imaging component at the front end of the detection probe to form an image sequence; Step S120: For each frame of the image sequence, identify potential abnormal regions where the image differs from a preset normal surface; Step S130: Obtain multiple feature information of potential abnormal regions, including at least one of local contrast, texture details, edge intensity, and optical response features; Step S140: Based on image data and various feature information, perform preliminary classification of potential abnormal regions to obtain defect candidate regions; Step S150: Check the spatial continuity and feature stability of the defect candidate region in the continuous image, determine the behavior pattern of the defect candidate region, and obtain the behavior pattern discrimination result; Step S160: Based on the behavior pattern discrimination result, obtain the area of ​​the real defect on the inner wall of the pipe to be inspected to confirm the defect on the inner wall of the pipe to be inspected, and output the type, size and location information of the defect. It should be noted that stainless steel EP tubing for semiconductor processing refers to stainless steel tubing that has undergone electrolytic polishing to achieve ultra-high cleanliness requirements on its inner surface. It is widely used in semiconductor manufacturing to transport ultrapure fluids. Minor defects on its inner wall are crucial to the yield of semiconductor products. The inspection probe is typically a precision device integrating imaging components (such as a camera and light source) and a motion mechanism, used to penetrate deep into the tubing for inspection. The imaging components are responsible for capturing visual information of the tubing's inner wall and generating image data. An image sequence refers to a series of image frames continuously acquired by the inspection probe during its movement; these image frames together constitute a complete visual record of the tubing's inner wall. Potentially abnormal areas refer to any area in the image that visually differs from the normal tubing surface. These areas may be defects or artifacts caused by uneven lighting. Defect candidate areas are areas considered to have a high probability of being real defects after initial screening. Behavioral pattern discrimination results refer to the conclusions drawn by analyzing the dynamic characteristics of defect candidate areas in continuous images to determine whether they are real defects or artifacts. In one embodiment, firstly, the detection probe is moved along the axis of the pipe to be inspected, and image data of the inner wall of the pipe is acquired by an imaging component at the front end of the detection probe. The imaging component at the front end of the detection probe can be a miniature industrial camera with a ring LED light source, responsible for capturing images of the inner wall of the pipe in real time. During the movement of the probe, the imaging component continuously captures images at a preset frame rate, thereby forming a continuous image sequence. The probe can move at a speed of 10 millimeters per second, while the camera captures images at a speed of 30 frames per second, ensuring sufficient overlap between the images for subsequent analysis of spatial continuity. Secondly, for each frame of the image sequence, potential abnormal areas that differ from a preset normal surface are identified. This step aims to preliminarily screen out all potentially abnormal areas. A background subtraction-based method can be used to compare the current frame image with a pre-established reference image of the normal inner wall of the pipe, or a method based on image statistical characteristics, such as local variance analysis, can be used to identify areas whose brightness, color, or texture is significantly different from the surrounding background. Thirdly, various feature information of the potential abnormal areas is acquired, including at least one of local contrast, texture detail, edge intensity, and optical response characteristics. Local contrast can be obtained by calculating the brightness difference of pixels within a potential anomalous region to distinguish between bright and dark areas. Texture details can be extracted using methods such as Gray-Level Co-occurrence Matrix (GLCM) to describe the roughness and uniformity of the region. Edge intensity can be calculated using edge detection operators such as Sobel and Canny to reflect the sharpness of the region boundary. Optical response features, such as polarization information, can be obtained by acquiring images in different polarization directions and calculating the degree of polarization and polarization angle. For a region suspected of being a scratch, its local contrast may be high, the edge intensity obvious, and the texture details may exhibit linear discontinuities. Next, based on image data and various feature information, potential anomalous regions are initially classified to obtain defect candidate regions. After obtaining various feature information of potential anomalous regions, this information can be used to perform preliminary screening and classification of potential anomalous regions. A series of threshold rules based on feature information are set; if the local contrast is below a certain threshold and the area is smaller than a certain threshold, it is excluded as a lighting artifact. Alternatively, machine learning classifiers, such as Support Vector Machines (SVM) or Random Forests, can be used to classify potentially anomalous regions into categories such as suspected physical defects, suspected transparent residues, or suspected lighting artifacts. A region with high local contrast, strong edge intensity, and specific texture features may be initially classified as a suspected physical defect, while a region with low local contrast but abnormal polarization may be classified as a suspected transparent residue. In one embodiment, time-series information is used to further distinguish between real defects and dynamic artifacts. For example, the positional changes of each defect candidate region in consecutive frames of images can be tracked. If a region exhibits a stable spatial position and similar features (such as local contrast, texture, etc.) in consecutive frames, it is likely to be a real defect. Conversely, if a region's position drifts rapidly, its shape changes drastically, or its features fluctuate greatly in consecutive frames, it is likely an artifact caused by probe shaking or uneven lighting. Finally, based on the behavioral pattern discrimination results, the regions of real defects on the inner wall of the pipe under inspection are obtained to confirm the defects on the inner wall of the pipe under inspection, and the type, size, and location information of the defects are output. After discerning the behavioral patterns of the defect candidate regions, the system can finally confirm the real defects based on these results. For example, only regions that are judged to be real defect behavioral patterns are confirmed as real defects. For these real defects, the system further extracts their precise type (such as scratches, pits, residues), size (such as length, width, depth), and precise location information on the inner wall of the pipe (such as axial position, circumferential angle), and outputs this information. It should be noted that the visual inspection method for surface defects of stainless steel EP tubes used in semiconductor processing proposed in this application organically combines image acquisition, potential abnormal region identification, multi-feature information extraction, preliminary classification, behavior pattern discrimination, and final defect confirmation and information output to form a complete and efficient inspection process. During the process of the probe moving along the axis of the tube under inspection and acquiring image sequences, the system can capture visual information of the inner wall of the tube in real time. Subsequently, each frame of image is processed to identify all potential abnormal regions that differ from normal surfaces. These potential abnormal regions are further analyzed to extract various feature information, including local contrast, texture details, edge intensity, and optical response characteristics. This multi-dimensional feature information allows the system to more comprehensively understand the nature of the abnormal regions. Based on this image data and various feature information, the system performs preliminary classification of potential abnormal regions and filters out candidate defect regions. This preliminary classification step can effectively eliminate most artifacts caused by uneven illumination or noise. Subsequently, the system checks the spatial continuity and feature stability of the candidate defect regions in continuous images to determine their behavior patterns. Through this behavior pattern discrimination, the system can greatly reduce the false alarm rate. Ultimately, based on the behavioral pattern discrimination results, the system can accurately identify the actual defect area on the inner wall of the pipe under inspection and confirm the existence of the defect. Simultaneously, the system will also output the type, size, and precise location information of the defect. Specifically, obtaining various feature information of potential anomaly regions as described above may include the following steps: Analyze the brightness difference between the potential anomalous region and its surrounding background to obtain the local contrast of the potential anomalous region; Evaluate the uniformity and complexity of the surface texture in potential anomaly regions to obtain texture details of the potential anomaly regions; Identify the boundary clarity, shape contour, and size information of potential anomalous regions to obtain the edge strength of these regions. The polarization information of the potential anomalous region in different polarization states is extracted to obtain the optical response characteristics of the potential anomalous region. The polarization information includes the degree of polarization and the polarization angle. The analysis of brightness differences between potential anomalous regions and their surrounding background aims to quantify the visual prominence of these regions. Local contrast is a crucial indicator of the brightness difference between an area and its environment; high contrast typically suggests the presence of an anomaly. The uniformity and complexity of the surface texture of potential anomalous regions are evaluated to capture their microstructural features. Texture details reflect the roughness, periodicity, or randomness of the material surface. Identifying the boundary sharpness, shape contour, and size of potential anomalous regions aims to characterize their geometric properties. Edge strength reflects the clarity of the anomalous region's boundaries, the regularity of its shape, and its physical size within the image—information crucial for defect localization and classification. Extracting polarization information from potential anomalous regions in different polarization states aims to obtain their optical response characteristics. Optical response characteristics, particularly polarization information, reveal the reflection, scattering, and absorption properties of light on a material surface, offering unique advantages for detecting defects that are difficult to capture using traditional brightness or color information, such as transparent residues, thin film defects, or surface stress. Polarization information typically includes the degree of polarization and the angle of polarization, which quantify the degree and direction of light polarization, respectively. It should be noted that this application can significantly improve the accuracy and robustness of visual inspection of surface defects in stainless steel EP tubes used in semiconductor processing. By quantifying local contrast, false alarms caused by uneven illumination or background texture can be effectively eliminated; by evaluating texture details, different types of surface damage, such as scratches and corrosion spots, can be distinguished more finely; by identifying edge intensity, the geometric contours of defects can be accurately delineated, providing a reliable basis for dimensional measurement; and by extracting optical response features, especially polarization information, transparent or translucent defects that are difficult to detect by traditional visual inspection methods, such as oil stains, water stains, or film residues, can be effectively detected. These detailed feature acquisition methods work together to enable the system to understand the nature of potential abnormal areas more comprehensively and deeply, thereby significantly reducing the false alarm and false negative rates and improving the reliability and accuracy of defect detection. The above analysis of the brightness difference between the potential anomalous region and its surrounding background to obtain the local contrast of the potential anomalous region specifically includes: Obtain the brightness values ​​of the pixels within the potential abnormal region to obtain the maximum and minimum brightness values; Calculate the average brightness value and standard deviation of the brightness values ​​of the pixels within the potential anomaly region; Obtain the brightness value of the surrounding background of the potential abnormal area; Based on the average luminance value, standard deviation, and luminance value of the surrounding background of the potential anomaly area, the luminance difference between the potential anomaly area and the surrounding background is analyzed to obtain the local contrast of the potential anomaly area. In this method, inner pixels refer to pixels located within the potential anomaly region. By acquiring the brightness values ​​of these pixels, the brightness range of the region can be initially understood. The average brightness value reflects the overall brightness level of the region, while the standard deviation characterizes the dispersion or range of variation of pixel brightness values ​​within the region. The surrounding background refers to the normal surface area adjacent to the potential anomaly region, and its brightness value can be used as a reference benchmark. Local contrast is an important indicator for measuring the visual difference between a region and its surrounding environment, playing a crucial role in defect identification. By acquiring the maximum and minimum brightness values ​​within the region, extreme brightness information within the region can be captured; by calculating the average brightness value and standard deviation, the overall brightness and brightness distribution characteristics of the region can be comprehensively reflected; and by introducing the brightness value of the surrounding background as a reference, the calculation of local contrast becomes more relative and objective. Therefore, this method can effectively distinguish between real defects and artifacts caused by uneven lighting or material changes, because real defects usually have significant and stable differences from the surrounding background in terms of brightness, texture, etc. In some embodiments of this application described above, obtaining various feature information of a potential anomalous region includes evaluating the uniformity and complexity of the surface texture of the potential anomalous region to obtain texture details. Specifically, the step of evaluating the uniformity and complexity of the surface texture of the potential anomalous region to obtain texture details may include the following: The eigenvalues ​​of the gray-level co-occurrence matrix of potential abnormal regions are obtained to get the energy and entropy values; The uniformity and complexity of the surface texture of potential anomaly regions are evaluated based on energy and entropy values ​​to obtain texture details of the potential anomaly regions. Energy and uniformity are positively correlated, and entropy and complexity are positively correlated. Gray-Level Co-occurrence Matrix (GLCM) is a commonly used texture analysis method. It describes the texture features of an image by statistically analyzing the distribution of gray values ​​of pixel pairs with certain spatial relationships (e.g., adjacent, separated by a certain distance). The eigenvalues ​​of the GLCM specifically refer to the numerical values ​​of quantified texture attributes extracted from the matrix, such as energy and entropy. Energy reflects the uniformity of the image's gray-level distribution and the coarseness of the texture. A higher energy value indicates a uniform texture with minimal gray-level variation; conversely, a lower energy value indicates a complex texture with drastic gray-level changes. Entropy reflects the complexity and randomness of the image texture. A higher entropy value indicates a more complex texture with higher randomness; conversely, a lower entropy value indicates a simpler texture with stronger regularity. Therefore, energy is positively correlated with uniformity, meaning a higher energy value indicates a more uniform texture; entropy is positively correlated with complexity, meaning a higher entropy value indicates a more complex texture. The proposed solution quantifies the texture detail of potential anomalous regions by introducing a gray-level co-occurrence matrix and its eigenvalues ​​(energy and entropy). Specifically, by calculating the gray-level co-occurrence matrix of potential anomalous regions and extracting energy and entropy values ​​from it, the texture uniformity and complexity of the region can be objectively reflected. The energy value directly indicates the smoothness of the texture, while the entropy value reveals the disorder or fineness of the texture. In some embodiments of this application, in order to more accurately obtain various feature information of potential anomaly regions, it is necessary to identify the edge intensity of the potential anomaly regions. To this end, this application further proposes to identify the boundary sharpness, shape contour, and size information of potential anomaly regions to obtain their edge intensity, specifically including: Identify the boundaries of potential abnormal regions based on a preset edge detection algorithm; Based on connected component analysis, the area, perimeter, aspect ratio, roundness, and convexity of each potential anomaly region are calculated. Identify the edge strength of potential anomalous regions by assessing their boundary clarity, shape profile, and size information based on their boundaries, area, perimeter, aspect ratio, roundness, and convexity. Specifically, a pre-defined edge detection algorithm can be understood as a mathematical method for identifying regions in an image where brightness or color changes abruptly. For example, operators such as Canny, Sobel, Prewitt, or Laplacian can be used, with the aim of accurately outlining the contours of potential anomalous regions. Connected component analysis refers to the process of identifying and labeling connected regions in an image with the same or similar pixel values. Through connected component analysis, the geometric properties of each potential anomalous region can be calculated, such as area, perimeter, aspect ratio, roundness, and convexity. Area represents the number of pixels occupied by the region, perimeter represents the length of the region's boundary, aspect ratio reflects the region's direction of extension, roundness measures how closely the region's shape resembles a circle, and convexity describes the degree of protrusion of the region. The purpose of acquiring these geometric features is to provide rich morphological evidence for subsequent defect classification. Based on the identified boundaries, calculated area, perimeter, aspect ratio, roundness, and convexity, the boundary clarity, shape contour, and size information of potential anomalous regions can be comprehensively judged. For example, clear boundaries, regular shape contours, and specific size ranges are often associated with a certain type of defect, while unclear boundaries or irregular shapes may indicate artifacts or noise. The solution in this application first uses a pre-defined edge detection algorithm to accurately identify the boundaries of potential anomaly regions, and then calculates a series of geometric features for each potential anomaly region, such as area, perimeter, aspect ratio, roundness, and convexity, through connected component analysis. It is precisely because these detailed geometric features can comprehensively describe the morphological characteristics of potential anomaly regions that the system can comprehensively identify the boundary clarity, shape contour, and size information of potential anomaly regions based on these boundaries and geometric features, thereby accurately obtaining their edge strength. In some embodiments of this application, when acquiring various feature information of a potential anomalous region, it is necessary to extract the polarization information of the potential anomalous region in different polarization states to obtain the optical response characteristics of the potential anomalous region. Specifically, the extraction of the polarization information of the potential anomalous region in different polarization states to obtain the optical response characteristics of the potential anomalous region includes: Acquire images of potential anomalous regions in different polarization directions; The degree of polarization and the polarization angle are calculated from images of potential anomaly regions in different polarization directions. Information about potential anomalous regions in different polarization states is extracted based on the degree of polarization and polarization angle to obtain the optical response characteristics of the potential anomalous regions. Specifically, acquiring images of potential anomalous regions in different polarization directions involves using an imaging component with polarization imaging capabilities to capture images of the potential anomalous region in multiple preset polarization directions (e.g., 0°, 45°, 90°, 135°, etc.). These images contain light intensity information under different polarization states, forming the basis for subsequent calculations of polarization information. The polarization imaging component can be understood as an optical sensor capable of capturing the polarization characteristics of light, aiming to obtain the reflection or scattering response of an object's surface to light of different polarizations. Further, calculating the degree of polarization and polarization angle of the images of the potential anomalous region in different polarization directions means calculating the degree of polarization and polarization angle of each pixel based on the acquired image data in different polarization directions using a specific polarization calculation model or algorithm. The degree of polarization characterizes the degree of light polarization, while the polarization angle indicates the polarization direction. In practical applications, methods such as the Stokes vector or Mueller matrix can be used for calculation, aiming to quantify the optical polarization characteristics of the potential anomalous region. Therefore, information about the potential anomalous region in different polarization states is extracted based on the degree of polarization and polarization angle to obtain the optical response characteristics of the potential anomalous region. This means that the calculated degree of polarization and polarization angle are used as optical response characteristics of potential anomalous regions. By acquiring images of potential anomalous regions in different polarization directions and calculating the degree of polarization and polarization angle based on these images, polarization information of potential anomalous regions in different polarization states can be extracted. This method allows the system to perform in-depth analysis of potential anomalous regions from the perspective of light polarization characteristics, rather than relying solely on traditional brightness or color information. By quantifying the degree of polarization and polarization angle, the surface characteristics of potential anomalous regions can be characterized more accurately, such as surface roughness, material type, and the presence of transparent films or residues. In this regard, this application further proposes the following steps for preliminary classification of potential anomaly regions based on image data and multiple feature information to obtain defect candidate regions: Based on preset threshold rules and image data, initial candidate regions are obtained. The preset threshold rules represent the exclusion of regions in the potential abnormal region that are less than the preset local contrast and regions in the potential abnormal region that are greater than the preset area. The initial candidate regions are classified based on various feature information to obtain classification results, which include suspected physical defects, suspected transparent residues, or suspected light artifacts. Defect candidate regions are obtained based on the classification results. Specifically, in the initial classification process, the identified potential abnormal regions are first screened based on preset threshold rules and image data to obtain initial candidate regions. The preset threshold rules aim to exclude regions that are clearly not defects. For example, regions with excessively low local contrast are excluded, as these regions are usually image noise or background brightness fluctuations rather than genuine surface defects. Similarly, regions with excessively large areas are excluded, as these abnormal regions are often artifacts caused by uneven lighting or large-scale background changes, rather than typical local defects. This approach effectively reduces the number of regions to be processed subsequently, improving processing efficiency. Furthermore, the obtained initial candidate regions are classified based on various feature information. These features can include local contrast, texture details, edge intensity, and optical response characteristics. By comprehensively analyzing these features, the initial candidate regions can be subdivided into different types, such as suspected physical defects, suspected transparent residues, or suspected lighting artifacts. Suspected physical defects typically refer to surface structural damage such as scratches, dents, and protrusions; suspected transparent residues may refer to substances adhering to the surface such as oil stains, water stains, and chemical residues; while suspected lighting artifacts refer to image representations of non-real defects caused by light reflection, shadows, or the characteristics of the imaging system itself. This detailed classification helps to more accurately understand the nature of abnormal regions. Finally, based on the above classification results, the final defect candidate regions can be obtained. This means that only regions classified as suspected physical defects or suspected transparent residues (or other real defect types) will be retained as defect candidate regions, while non-defect regions such as suspected lighting artifacts will be excluded. By introducing a phased preliminary classification strategy, the problems of false positives and false negatives that may exist in the preliminary classification of traditional methods are effectively solved. First, potential abnormal regions are initially screened through preset threshold rules. By utilizing macroscopic features such as local contrast and area, a large amount of non-defect interference information, such as tiny noise points or large areas of uneven lighting, can be quickly and effectively eliminated, thus avoiding complex subsequent analysis of these regions and significantly improving processing efficiency. Secondly, based on the initial candidate regions, the system further utilizes various feature information for detailed classification, distinguishing abnormal regions into suspected physical defects, suspected transparent residues, or suspected lighting artifacts. This multi-dimensional and multi-category classification mechanism enables the system to more accurately identify the nature of anomalies. For example, texture details and optical response features can effectively distinguish between physical damage and transparent residues, while local contrast and edge intensity can distinguish between real defects and lighting artifacts. In one embodiment, during the detection process, the imaging component acquires image data of the inner wall of the pipe to be inspected and identifies multiple potential abnormal areas. One area, A, exhibits extremely low local contrast and a very small area; another area, B, exhibits a very large area and gradual brightness variation; and yet another area, C, exhibits high local contrast, specific texture, and sharp edges. First, in the preliminary classification stage, based on a preset threshold rule, area A is excluded because its local contrast is below the preset threshold; area B is also excluded because its area is greater than the preset threshold. Thus, only area C is retained as an initial candidate area. Next, for the initial candidate area C, the system acquires various feature information such as its local contrast, texture details, edge intensity, and optical response characteristics. For example, by analyzing its high local contrast, irregular texture details, and sharp edge intensity, and combining this with its optical response characteristics under different polarization states, the system classifies it as a suspected physical defect. Ultimately, based on the classification results, region C was identified as a candidate defect region and entered the subsequent behavior pattern discrimination stage. In this way, the proposed solution effectively avoids misclassifying noise or lighting artifacts as defects, while ensuring that genuine defects are accurately identified and classified. This application further proposes a more refined and robust inspection method to ensure the accuracy of defect identification. To this end, this application further proposes methods to check the spatial continuity and feature stability of the aforementioned defect candidate regions in continuous images, including: Calculate the average displacement vector of each defect candidate region over N consecutive frames; Based on the average displacement vector and the preset displacement vector, check the spatial continuity of the defect candidate region in continuous images; The multi-dimensional feature values ​​of each defect candidate region are obtained in N consecutive frames to obtain the rate of change of the multi-dimensional feature values ​​within N frames. The multi-dimensional feature values ​​include local contrast, texture entropy, and polarization degree. Based on the rate of change and a preset threshold, the feature stability of the defect candidate region in continuous images is determined. Specifically, calculating the average displacement vector of each defect candidate region over N consecutive frames involves using image processing techniques, such as region-based feature point matching or centroid tracking algorithms, to track the spatial position changes of the same defect candidate region across N consecutive frames. Averaging these position changes yields an average displacement vector representing the overall movement trend of the defect candidate region within those N frames. This average displacement vector effectively filters out positioning errors caused by instantaneous noise or slight jitter in a single frame, thus more accurately reflecting the actual trajectory of the defect candidate region. Checking the spatial continuity of the defect candidate region in consecutive images, based on the average displacement vector and a preset displacement vector, can be understood as comparing the calculated average displacement vector with a preset displacement vector. The preset displacement vector can be set based on the expected movement speed and direction of the detection probe, or set to a threshold close to zero, to determine whether the defect candidate region maintains the expected spatial positional relationship across consecutive frames. For example, if the average displacement vector of the defect candidate region is consistent with the movement direction and speed of the detection probe, or if its displacement is much smaller than the preset static threshold, then its spatial position can be considered continuous. Obtaining the multi-dimensional feature values ​​of each defect candidate region across N consecutive frames to obtain the rate of change of these multi-dimensional feature values ​​within N frames involves extracting various features such as local contrast, texture entropy, and polarization degree for the same defect candidate region across N consecutive image frames. Local contrast reflects the brightness difference between the region and the surrounding background, texture entropy characterizes the complexity and randomness of the texture within the region, and polarization degree reveals the surface's response to polarized light. These multi-dimensional features can describe the visual characteristics of the defect candidate region from different perspectives. By calculating the rate of change of these features within N frames (e.g., standard deviation, the difference between the maximum and minimum values, or the slope of a linear regression), the stability of these features over time can be quantified. Determining the feature stability of the defect candidate region in consecutive images based on the rate of change and a preset threshold involves comparing the rate of change of the multi-dimensional feature values ​​obtained above with a preset stability threshold. If the rate of change of all or most features is lower than the corresponding preset threshold, it indicates that the visual features of the defect candidate region maintain a high degree of consistency across consecutive frames, thus being judged as having feature stability. Conversely, a high rate of feature change may indicate that the region is caused by transient artifacts or changes in illumination, lacking stability. This application can effectively reduce the false alarm rate and false negative rate in the visual inspection process of surface defects in stainless steel EP tubes used in semiconductor processing. By rigorously checking the spatial continuity and multi-dimensional feature stability of the defect candidate region, artifacts caused by environmental noise, changes in illumination, slight vibration of the detection probe, or transient surface reflections can be effectively filtered out, ensuring that only real and persistent surface defects are ultimately confirmed.This not only improves the accuracy of test results and reduces unnecessary subsequent manual re-inspection work, but also enhances the robustness and reliability of the entire testing system. The aforementioned driving detection probe moves along the axial direction of the pipe to be inspected and acquires image data of the inner wall of the pipe to be inspected based on the imaging component at the front end of the detection probe to form an image sequence, including: During the movement of the detection probe, the distance between the detection probe and the inner wall of the pipe under inspection, as well as the local curvature information of the inner wall of the pipe under inspection, are measured in real time to obtain distance information and curvature information; Based on distance and curvature information, the orientation of the detection probe is dynamically adjusted so that the imaging component at the front end of the detection probe is kept in a preset stable position relative to the inner wall of the pipe to be inspected. The preset stable position includes a vertical position and a centered position. While maintaining the imaging component at the front end of the detection probe in a preset stable position relative to the inner wall of the pipe under inspection, image data of the inner wall of the pipe under inspection is acquired by the imaging component at the front end of the detection probe to form an image sequence. Specifically, during the movement of the detection probe, the distance between the detection probe and the inner wall of the pipe under inspection, as well as the local curvature information of the inner wall of the pipe under inspection, are measured in real time. The purpose is to obtain the relative position and attitude data of the probe and the inner wall of the pipe under inspection, providing accurate input for subsequent dynamic adjustments. The distance information can be understood as the normal distance between the imaging component of the probe and the surface of the inner wall of the pipe, while the curvature information characterizes the degree of bending of the inner wall of the pipe at the current position of the probe. This information can be acquired in real time by various sensors integrated on the detection probe, such as laser rangefinders, ultrasonic sensors, or vision-based depth sensing modules. Based on the distance and curvature information, the attitude of the detection probe is dynamically adjusted to ensure that the imaging component is always in the optimal imaging state. The attitude adjustment of the probe can include translation and rotation to maintain the imaging component at the front end of the detection probe in a preset stable position relative to the inner wall of the pipe under inspection. A preset stable position refers to the relative positional relationship that yields the best image quality, such as a vertical position or a centered position. A vertical position means the optical axis of the imaging component is perpendicular to the inner wall surface of the pipe under inspection, avoiding image distortion and uneven illumination. A centered position means the imaging component maintains central symmetry within the pipe's cross-section, ensuring uniform coverage and imaging of the entire circumferential inner wall of the pipe under inspection. In practical applications, dynamic adjustment of the probe can be achieved through miniature actuators or robotic arms integrated on the probe. These actuators are precisely controlled by the control system based on real-time measured distance and curvature information. By monitoring the relative position and orientation of the detection probe and the inner wall of the pipe in real time and making dynamic adjustments accordingly, the problem of image quality degradation caused by unstable probe orientation in traditional inspection is solved. Specifically, real-time measurement of distance and curvature information allows the system to accurately determine the relative geometric relationship between the probe and the inner wall of the pipe under inspection. Based on this information, the system can precisely control the probe's orientation, ensuring that its imaging component always remains in a preset stable position, such as perpendicular to and centered on the inner wall of the pipe under inspection, ensuring the consistency of the imaging field of view and the uniformity of illumination conditions. This application significantly improves the stability and quality of image acquisition of the inner wall of stainless steel EP tubes used in semiconductor processing. The imaging component maintains a preset stable position relative to the inner wall of the tube under inspection, effectively avoiding problems such as image blurring, distortion, or uneven illumination caused by probe movement, distance changes, or angular shifts. This not only ensures the continuity and consistency of the image sequence but also provides high-quality raw data for subsequent identification of potential abnormal areas, acquisition of feature information, and defect classification, thereby greatly improving the accuracy and reliability of defect detection and reducing the risk of missed detections and false alarms. In one embodiment, the detection probe can integrate multiple laser rangefinders or ultrasonic sensors to acquire real-time distance information between the probe and the inner wall of the pipe under inspection. Simultaneously, the probe can also be equipped with a miniature inertial measurement unit (IMU) or a vision tracking system to acquire the probe's attitude and the local curvature information of the inner wall of the pipe under inspection. This sensor data is transmitted to the control unit in real time. The control unit calculates the adjustment amount of the probe's attitude according to a preset algorithm and dynamically adjusts the probe's pitch angle, yaw angle, and roll angle, as well as its radial position within the pipe's cross-section, by driving a miniature servo motor or piezoelectric actuator on the probe. This ensures that the imaging component at the probe's tip is always perpendicular to the inner wall surface of the pipe under inspection and located on the pipe's central axis. In this stable state, the imaging component (e.g., a high-resolution CMOS camera) acquires images of the inner wall of the pipe at a constant frame rate, forming a high-quality image sequence. See Figure 2 , Figure 2 This is a schematic diagram of a visual inspection system for surface defects of stainless steel EP tubes used in semiconductor processing, provided in one embodiment of this application. The visual inspection system 200 for surface defects of stainless steel EP tubes used in semiconductor processing includes: The first acquisition module 210 is used to drive the detection probe to move along the axial direction of the pipe to be inspected, and to acquire image data of the inner wall of the pipe to be inspected based on the imaging component at the front end of the detection probe, so as to form an image sequence. The recognition module 220 is used to identify potential abnormal regions in each frame of the image sequence that differ from a preset normal surface. The second acquisition module 230 is used to acquire multiple feature information of potential abnormal regions, including at least one of local contrast, texture details, edge intensity, and optical response features; The classification module 240 is used to perform preliminary classification of potential abnormal regions based on image data and multiple feature information to obtain defect candidate regions. The discrimination module 250 is used to check the spatial continuity and feature stability of the defect candidate region in the continuous image, discriminate the behavior pattern of the defect candidate region, and obtain the behavior pattern discrimination result. The output module 260 is used to obtain the area of ​​the actual defect on the inner wall of the pipe under inspection based on the behavior pattern discrimination result, so as to confirm the defect on the inner wall of the pipe under inspection, and output the type, size and location information of the defect. It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, and they will not be repeated here. It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium. The foregoing has provided a detailed description of the preferred embodiments of this application. However, this application is not limited to the above-described embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application. All such equivalent modifications or substitutions are included within the scope defined in this application.

Claims

1. A visual inspection method for surface defects in stainless steel EP tubes used in semiconductor processing, characterized in that, include: The detection probe is driven to move along the axial direction of the pipe to be inspected, and image data of the inner wall of the pipe to be inspected is acquired by the imaging component at the front end of the detection probe to form an image sequence; For each frame of the image sequence, identify potential abnormal regions where the image differs from a preset normal surface; The potential abnormal region is obtained by acquiring multiple feature information, including at least one of local contrast, texture detail, edge intensity, and optical response features; Based on the image data and the various feature information, the potential abnormal regions are preliminarily classified to obtain defect candidate regions; The spatial continuity and feature stability of the defect candidate regions in continuous images are checked, the behavior patterns of the defect candidate regions are determined, and the behavior pattern determination results are obtained. Based on the behavior pattern discrimination result, the actual defect area of ​​the inner wall of the pipe to be inspected is obtained to confirm the defect of the inner wall of the pipe to be inspected, and the type, size and location information of the defect are output.

2. The method according to claim 1, characterized in that, The acquisition of various feature information of the potential abnormal region includes: Analyze the brightness difference between the potential anomalous region and the surrounding background to obtain the local contrast of the potential anomalous region; Evaluate the uniformity and complexity of the surface texture of the potential anomaly region to obtain the texture details of the potential anomaly region; Identify the boundary clarity, shape contour, and size information of the potential anomaly region to obtain the edge strength of the potential anomaly region; The polarization information of the potential anomalous region in different polarization states is extracted to obtain the optical response characteristics of the potential anomalous region, wherein the polarization information includes the degree of polarization and the polarization angle.

3. The method according to claim 2, characterized in that, The step of analyzing the brightness difference between the potential anomalous region and its surrounding background to obtain the local contrast of the potential anomalous region includes: Obtain the brightness values ​​of the pixels within the potential abnormal region to obtain the maximum and minimum brightness values; Calculate the average brightness value and standard deviation of the brightness values ​​of the pixels within the potential abnormal region; Obtain the brightness value of the surrounding background of the potential abnormal region; The brightness difference between the potential anomaly region and its surrounding background is analyzed based on the average brightness value, the standard deviation, and the brightness value of the background surrounding the potential anomaly region to obtain the local contrast of the potential anomaly region.

4. The method according to claim 2, characterized in that, The evaluation of the uniformity and complexity of the surface texture of the potential anomaly region to obtain the texture details of the potential anomaly region includes: The eigenvalues ​​of the gray-level co-occurrence matrix of the potential abnormal region are obtained to obtain the energy value and entropy value; The uniformity and complexity of the surface texture of the potential anomaly region are evaluated based on the energy value and the entropy value to obtain the texture details of the potential anomaly region, wherein the energy value and the uniformity are positively correlated, and the entropy value and the complexity are positively correlated.

5. The method according to claim 2, characterized in that, The process of identifying the boundary clarity, shape contour, and size information of the potential anomaly region to obtain the edge strength of the potential anomaly region includes: The boundaries of the potential abnormal regions are identified based on a preset edge detection algorithm; Based on connected component analysis, the area, perimeter, aspect ratio, roundness, and convexity of each potential anomaly region are calculated. The edge strength of the potential anomaly region is obtained by identifying the boundary clarity, shape contour, and size information of the potential anomaly region based on the boundary, area, perimeter, aspect ratio, roundness, and convexity.

6. The method according to claim 2, characterized in that, The step of extracting polarization information of the potential anomalous region in different polarization states to obtain the optical response characteristics of the potential anomalous region includes: Acquire images of the potential anomaly region in different polarization directions; The degree of polarization and the polarization angle are calculated from images of the potential anomaly region in different polarization directions. Based on the degree of polarization and the angle of polarization, information about the potential anomalous region in different polarization states is extracted to obtain the optical response characteristics of the potential anomalous region.

7. The method according to claim 1, characterized in that, The preliminary classification of the potential abnormal regions based on the image data and the various feature information yields defect candidate regions, including: Based on the preset threshold rules and the image data, an initial candidate region is obtained, wherein the preset threshold rules represent excluding regions in the potential abnormal region that are less than a preset local contrast and excluding regions in the potential abnormal region that are greater than a preset area. The initial candidate region is classified according to the various feature information to obtain the classification result, which includes suspected physical defects, suspected transparent residues, or suspected light artifacts. Based on the classification results, candidate defect regions are obtained.

8. The method according to claim 1, characterized in that, The process of checking the spatial continuity and feature stability of the defect candidate region in consecutive images includes: Calculate the average displacement vector of each of the defect candidate regions over N consecutive frames; Based on the average displacement vector and the preset displacement vector, check the spatial continuity of the defect candidate region in the continuous image; The multi-dimensional feature values ​​of each of the defect candidate regions are obtained in N consecutive frames to obtain the rate of change of the multi-dimensional feature values ​​within N frames. The multi-dimensional feature values ​​include local contrast, texture entropy, and polarization degree. Based on the rate of change and a preset threshold, the feature stability of the defect candidate region in continuous images is determined.

9. The method according to claim 1, characterized in that, The driving detection probe moves along the axial direction of the pipe to be inspected, and acquires image data of the inner wall of the pipe to be inspected based on the imaging component at the front end of the detection probe to form an image sequence, including: During the movement of the detection probe, the distance between the detection probe and the inner wall of the pipe to be inspected, as well as the local curvature information of the inner wall of the pipe to be inspected, are measured in real time to obtain distance information and curvature information; Based on the distance information and the curvature information, the orientation of the detection probe is dynamically adjusted so that the imaging component at the front end of the detection probe is kept at a preset stable position relative to the inner wall of the pipe to be inspected. The preset stable position includes a vertical position and a centered position. While keeping the imaging component at the front end of the detection probe in a preset stable position relative to the inner wall of the pipe to be inspected, image data of the inner wall of the pipe to be inspected is acquired by the imaging component at the front end of the detection probe to form an image sequence.

10. A visual inspection system for surface defects of stainless steel EP tubes used in semiconductor processing, characterized in that, include: The first acquisition module is used to drive the detection probe to move along the axial direction of the pipe to be inspected, and to acquire image data of the inner wall of the pipe to be inspected according to the imaging component at the front end of the detection probe, so as to form an image sequence. The identification module is used to identify potential abnormal regions in each frame of the image sequence that differ from a preset normal surface. The second acquisition module is used to acquire multiple feature information of the potential abnormal region, the feature information including at least one of local contrast, texture details, edge intensity and optical response features; The classification module is used to perform preliminary classification of the potential abnormal regions based on the image data and the various feature information to obtain defect candidate regions; The discrimination module is used to check the spatial continuity and feature stability of the defect candidate region in continuous images, discriminate the behavior pattern of the defect candidate region, and obtain the behavior pattern discrimination result. The output module is used to determine the area of ​​the actual defect in the inner wall of the pipe under inspection based on the behavior pattern discrimination result, so as to confirm the defect in the inner wall of the pipe under inspection, and output the type, size and location information of the defect.

Images