Real-time screening method and system for inner tube surface defects based on machine vision

Abstract

The application relates to the technical field of machine vision image processing, and discloses a real-time screening method and system for inner tube surface flaws based on machine vision, which comprises the following steps: carrying out circumferential expansion scanning on the inner tube surface to obtain a multi-angle image sequence, carrying out contrast difference synthesis on the light intensity of multi-light paths at the same position to generate a differential contrast image; extracting an abnormal candidate area connected domain from the image as a candidate flaw connected domain, and quantifying morphological characteristic parameters thereof; comparing the characteristic parameters with preset flaw judgment conditions item by item, screening out a real flaw area, and synchronously recording flaw coordinates and type marks; generating a flaw distribution map based on the coordinates and type marks, jointly deciding the distribution map in combination with the number and types of flaws, and judging the quality grade of the inner tube; converting the quality grade into a screening control instruction, outputting the screening control instruction to a sorting execution mechanism, and completing real-time flaw sorting of the inner tube; and the application can improve the efficiency of real-time screening of inner tube surface flaws based on machine vision.

Description

Technical Field

[0001] This invention relates to the field of machine vision image processing technology, and in particular to a machine vision-based method and system for real-time screening of defects on the surface of inner tubes. Background Technology

[0002] Currently, rubber inner tube production lines generally use machine vision to complete surface defect detection and sorting. Existing detection solutions mostly rely on single-angle image acquisition and fixed threshold grayscale segmentation to complete defect identification. When faced with low-contrast defects such as inner tube curved surfaces, uneven rubber material reflection, small scratches, bubbles, and missing rubber, the identification stability is poor. The overall screening accuracy and real-time processing speed are difficult to match the needs of high-speed continuous production lines.

[0003] For example, conventional single-camera frontal imaging can only capture the reflection of a single light path. The circumferential curved surface of the inner tube has differences in light and shadow, and small pits and linear cracks can be masked by the background grayscale due to reflection, directly causing missed detections. At the same time, traditional algorithms rely on the single indicator of defect pixel area to judge defects. They cannot effectively distinguish interference noise such as demolding patterns and slight rubber material protrusions on the rubber surface, and are prone to misclassifying false noise as defects, resulting in a large number of false detections. Furthermore, they lack a comprehensive assessment of the spatial distribution of defects and the degree of harm caused by clusters. They rely on a single defect judgment standard to classify product grades, and the grading results deviate significantly from the actual scrapping standards for inner tubes. In addition, the existing detection process does not achieve full-link linkage of image acquisition, feature extraction, defect judgment, and sorting execution. The lag in image processing will cause the inner tube flow and sorting actions on the production line to be out of sync, resulting in missorted and missed workpieces.

[0004] Therefore, existing machine vision-based inner tube defect screening technologies suffer from multiple technical shortcomings, including limited image imaging information, insufficient quantification of defect features, weak ability to distinguish between genuine and fake defects, one-sided product quality grading logic, and poor real-time linkage with production lines. These technologies cannot simultaneously ensure detection accuracy and real-time sorting in high-speed production scenarios, severely restricting the efficiency of quality inspection of finished inner tubes. There is an urgent need for a real-time defect screening solution that integrates multi-optical path imaging, multi-dimensional feature quantification, and cluster joint rating to address these issues. Summary of the Invention

[0005] This invention provides a machine vision-based method and system for real-time screening of defects on the surface of inner tubes, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a machine vision-based real-time screening method for inner tube surface defects, comprising: S1. Perform a circumferential unfolding scan on the surface of the inner tube to obtain a multi-angle image sequence of the inner tube; S2. Perform contrast difference synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate the differential contrast image of the inner tube; S3. Connected component extraction is performed on the candidate regions of surface anomalies in the differential contrast image to obtain the candidate defect connected components of the inner tube, and morphological quantization is performed on the candidate defect connected components to obtain the morphological feature parameters of the inner tube. S4. Compare the morphological feature parameters with the preset defect judgment conditions item by item to determine the actual defect area of ​​the inner tube, and record the position coordinates and defect type mark of the actual defect area in the inner tube. S5. Perform topological reconstruction on the location coordinates and the defect type markers to obtain the defect distribution map of the inner tube, and make a joint decision on the defect distribution map based on the number and type of the actual defect areas to obtain the quality grade of the inner tube; S6. Map the quality grade to a screening control command to drive the sorting execution mechanism to perform real-time screening of the inner tube.

[0007] Preferably, the step of performing a circumferential unfolding scan of the inner tube surface to obtain a multi-angle image sequence of the inner tube includes: The inner tube is mounted on the rotary drive assembly, and the inner tube is driven to rotate at a constant speed around the axis of the rotary drive assembly. During the rotation process, the surface images of the inner tube are acquired at equal angular intervals along the circumference of the inner tube by the image acquisition component, resulting in a circumferential image sequence of the inner tube; A circular unfolding mapping is performed on each frame of the circumferential image sequence, and the mapping results are spliced ​​in time to obtain the panoramic unfolded image of the inner tube. A sliding window extraction is performed on the panoramic unfolded image to obtain a multi-angle image sequence of the inner tube.

[0008] Preferably, the step of performing contrast differential synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate a differential contrast image of the inner tube includes: Extract the pixel values ​​of each frame in the multi-angle image sequence at the same surface position to obtain the multi-angle pixel value sequence of the inner tube; Extreme value detection is performed on the multi-angle pixel value sequence to determine the maximum and minimum pixel values ​​of the inner tube; The initial difference value of the inner tube is obtained by performing a difference operation between the maximum pixel value and the minimum pixel value; The initial difference value is mapped to a preset gray value range to obtain the normalized difference value of the inner tube; By reorganizing the spatial positions of the normalized difference values ​​according to the actual spatial arrangement order of the inner tube, a difference contrast image of the inner tube is generated.

[0009] Preferably, the step of extracting connected components from the candidate surface anomaly regions in the differential contrast image to obtain the candidate defect connected components of the inner tube includes: Based on the circumferential diameter and axial length of the inner tube, the differential contrast image is divided into equidistant grids to obtain local detection sub-regions of the inner tube; The grayscale values ​​of pixels within the local detection sub-region are statistically analyzed and the outlier degree is measured to obtain the local mean and local standard deviation of the local detection sub-region. The local detection sub-region is binarized and segmented using the local mean and the local standard deviation to obtain a local binarized sub-map of the local detection sub-region. The local binarized sub-images are stitched together and fused, and morphological opening is performed on the fused image to obtain the global binarized image of the inner tube. Connected component labeling is performed on the global binarized image to obtain the edge pixel set of the inner tube, and the pixel area of ​​the region enclosed by the edge pixel set is calculated to obtain the actual pixel area of ​​the inner tube. Connected regions whose actual pixel area is less than a preset area threshold are discarded as pseudo-noise, and the remaining connected regions are retained as candidate defect connected regions of the inner tube.

[0010] Preferably, the step of performing morphological quantization on the connected components of the candidate defects to obtain the morphological feature parameters of the inner tube includes: Extract the minimum bounding ellipse of the candidate defect connected domain, and obtain the length of the major axis of the minimum bounding ellipse; The directional deflection parameters of the candidate defect connected domain are obtained by directional calculation of the angle between the length of the major axis and the axial direction of the inner tube. The edge pixel set of the candidate defective connected region is fitted with a second moment to obtain the dispersion parameter and elongation parameter of the candidate defective connected region. Extend a predetermined number of pixels outward along the edge normal direction of the candidate defect connected region, and collect the grayscale values ​​of the pixels in the extended region; Based on the edge normal direction, the gradient magnitude of the gray value is analyzed by gradient decay to obtain the edge sharpness parameter of the candidate defect connected domain; The morphological feature parameters of the inner tube are obtained by combining the actual pixel area, the orientation deflection parameter, the dispersion parameter, the elongation parameter, and the edge sharpness parameter.

[0011] Preferably, the step of comparing the morphological feature parameters with preset defect judgment conditions item by item to determine the actual defect area of ​​the inner tube, and recording the position coordinates and defect type label of the actual defect area in the inner tube, includes: The morphological feature parameters are compared with the preset defect judgment conditions one by one to obtain the independent comparison results of the inner tube, and the independent comparison results are weighted and evaluated to obtain the comprehensive score value of the inner tube. The comprehensive score is compared with a preset comprehensive judgment threshold. When the comprehensive score is not less than the comprehensive judgment threshold, the candidate defect connected region is determined to be the real defect region of the inner tube; otherwise, it is determined to be a false defect region and is removed. The actual defect area is matched with a preset defect type determination table to determine the defect type label of the actual defect area; Extract the centroid coordinates and the vertex coordinates of the smallest bounding rectangle of the actual defect area, and map the centroid coordinates and the vertex coordinates of the smallest bounding rectangle to the spatial coordinate system of the inner tube to obtain the position coordinates of the actual defect area.

[0012] Preferably, the step of performing topological reconstruction on the position coordinates and the defect type markers to obtain the defect distribution map of the inner tube includes: The position coordinates are mapped to the two-dimensional unfolded plane coordinate system of the inner tube to obtain the plane coordinates of the actual defect area; Based on the planar coordinates, the spatial distance between the actual defective areas is calculated to obtain the spatial distance value of the actual defective areas; By using the spatial distance value, adjacency determination and connectivity analysis are performed on the actual defect area to obtain the spatial adjacency relationship diagram of the actual defect area; Extract the defect clusters from the spatial adjacency graph, and count the types and number of defects in the defect clusters to obtain the cluster type composition information of the defect clusters; The cluster type composition information is associated with the spatial distribution location of the defect cluster to generate a defect distribution map of the inner tube.

[0013] Preferably, the step of jointly deciding on the defect distribution map based on the quantity and type of the actual defective areas to obtain the quality grade of the inner tube includes: The total number of real defect areas in the defect cluster is counted to obtain the defect density value of the real defect area; The cluster type composition information is subjected to key type identification to obtain the type identification result of the real defect area; Based on the defect density value and the type identification result, the cluster hazard quantification of the defect distribution map is performed to obtain the cluster hazard quantification value of the defect distribution map; The quality level of the inner tube is obtained by globally weighting and aggregating the cluster hazard quantification values ​​and mapping them to a level.

[0014] Preferably, the step of mapping the quality grade to screening control instructions to drive the sorting execution mechanism to perform real-time screening of the inner tubes includes: According to the preset quality grade classification rules, the quality grades are classified to determine the grade category corresponding to the inner tube; Based on the grade category, a preset filtering strategy mapping table is matched to determine the target filtering strategy corresponding to the grade category, and a corresponding filtering control instruction is generated based on the target filtering strategy. The screening control command is output to the sorting execution mechanism, which drives the sorting execution mechanism to perform a screening action on the inner tube that matches the quality grade.

[0015] To address the above problems, the present invention also provides a machine vision-based real-time screening system for inner tube surface defects, the system comprising: The image acquisition module is used to perform a circumferential unfolding scan of the surface of the inner tube to obtain a multi-angle image sequence of the inner tube. The contrast difference module is used to perform contrast difference synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate the differential contrast image of the inner tube. The parameter quantization module is used to extract the connected components of the surface anomaly candidate regions in the differential contrast image to obtain the candidate defect connected components of the inner tube, and to perform morphological quantization on the candidate defect connected components to obtain the morphological feature parameters of the inner tube. The defect marking module is used to compare the morphological feature parameters with the preset defect judgment conditions one by one to determine the actual defect area of ​​the inner tube, and record the position coordinates of the actual defect area in the inner tube and the defect type mark. The grade determination module is used to perform topological reconstruction on the location coordinates and the defect type markers to obtain the defect distribution map of the inner tube, and to make a joint decision on the defect distribution map based on the number and type of the actual defect areas to obtain the quality grade of the inner tube. The real-time screening module is used to map the quality grade into screening control instructions, driving the sorting execution mechanism to perform real-time screening of the inner tube.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention performs a circumferential panoramic scan of the inner tube and generates a differential contrast image, which can fully extract the reflective information from different angles on the curved surface, completely highlighting various minute defects on the rubber surface. Combined with regional grayscale statistics and multi-dimensional morphological parameter quantification, it can accurately remove image noise, stably identify various real defects, and simultaneously record the precise coordinates and defect types of defects, greatly improving the accuracy of inner tube surface defect identification.

[0017] 2. This invention generates a defect distribution map by completing topological reconstruction based on the spatial distribution of defects. It comprehensively determines the product quality level by combining the number, type, and cluster hazard of defects, and then directly outputs the corresponding control signal to drive the sorting equipment to perform sorting actions synchronously. The entire process of collection, detection, rating, and sorting runs in a continuous manner, which can adapt to the continuous operation requirements of the production line and effectively improve the overall processing efficiency of inner tube defect screening. Attached Figure Description

[0018] Figure 1 This is a flowchart illustrating a machine vision-based real-time screening method for inner tube surface defects according to an embodiment of the present invention. Figure 2 A functional block diagram of a machine vision-based real-time screening system for inner tube surface defects provided in an embodiment of the present invention; The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0019] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0020] This application provides a machine vision-based real-time screening method for inner tube surface defects. The executing entity of this machine vision-based real-time screening method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the machine vision-based real-time screening method for inner tube surface defects can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides 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 (CDNs), and big data and artificial intelligence platforms.

[0021] Reference Figure 1The diagram shown is a flowchart illustrating a machine vision-based real-time screening method for inner tube surface defects according to an embodiment of the present invention. In this embodiment, the machine vision-based real-time screening method for inner tube surface defects includes: S1. Perform a circumferential unfolding scan on the surface of the inner tube to obtain a multi-angle image sequence of the inner tube; In this embodiment of the invention, the step of performing a circumferential unfolding scan of the inner tube surface to obtain a multi-angle image sequence of the inner tube includes: The inner tube is mounted on the rotary drive assembly, and the inner tube is driven to rotate at a constant speed around the axis of the rotary drive assembly. During the rotation process, the surface images of the inner tube are acquired at equal angular intervals along the circumference of the inner tube by the image acquisition component, resulting in a circumferential image sequence of the inner tube; A circular unfolding mapping is performed on each frame of the circumferential image sequence, and the mapping results are spliced ​​in time to obtain the panoramic unfolded image of the inner tube. A sliding window extraction is performed on the panoramic unfolded image to obtain a multi-angle image sequence of the inner tube.

[0022] Specifically, the fully formed rubber inner tube is mounted on the inner support tooling structure of the rotary drive assembly. The outer support working surface of the tooling structure is completely in contact with the inner ring sidewall of the inner tube. The central axis of the tooling structure is completely coincident with the annular central axis of the inner tube itself. The power transmission structure inside the rotary drive assembly outputs a stable and uniform rotational torque. The torque is transmitted to the inner tube through the tooling structure, causing the inner tube to rotate continuously and uniformly around the central axis of the tooling. The outer support force of the tooling structure is continuous and uniform, ensuring that the inner tube does not experience axial deviation, radial runout, or circumferential slippage throughout the entire rotation process. This allows every area on the surface of the inner tube to pass through the field of view of the image acquisition assembly at a uniform linear velocity.

[0023] The image acquisition component is fixedly mounted on a fixed bracket on the outside of the inner tube's rotation path. The camera lens is vertically aimed at the outer surface area of ​​the inner tube. The rotary encoder of the rotary drive component maintains signal communication with the image acquisition component. During the rotation of the inner tube, the encoder outputs a acquisition trigger signal to the image acquisition component every time it rotates through a preset standard angle interval. After receiving the signal, the image acquisition component immediately captures a frame of the tire surface image within the current field of view. All tire surface images acquired sequentially at angle intervals are collected and stored in the order of shooting. All collected images form a unified circumferential image sequence of the inner tube. Each frame of the circumferential image sequence corresponds to a non-overlapping tire surface area on the circumference of the inner tube, and there is no situation where image content is repeatedly covered or the tire surface area is missed.

[0024] Each independent circumferential image in the stored circumferential image sequence is retrieved. For a single frame of circumferential image with an arc-shaped surface imaging form, a circular unfolding mapping operation is performed. During the operation, according to the standard circumferential curvature preset by the inner tube, the tire tread pixels distributed along the arc in the image are converted into horizontally arranged pixels row by row. The curved surface imaging image is corrected into a planar rectangular unfolded sub-image. After the single frame image mapping process is completed, a separate unfolded sub-image is generated. Then, according to the chronological order of the inner tube rotation and shooting, all unfolded sub-images obtained by mapping all single frame circumferential images are seamlessly stitched together end to end along the corresponding horizontal direction of the circumference. During the stitching process, the overlapping tire tread pixel content at the edges of adjacent sub-images is aligned to eliminate stitching gaps. After stitching and fusion, a panoramic unfolded image of the inner tube that can completely display the entire outer surface area of ​​the inner tube is generated. The panoramic unfolded image completely contains all tire tread pixel information contained in the circumferential image sequence, and there are no problems of missing tire tread area information or pixel image fragmentation.

[0025] It should be noted that the specific implementation of the circular unfolding mapping is as follows: Let the polar coordinates of any pixel in the original circumferential image be ( , ),in This represents the radial distance of the pixel from the image center (i.e., the center of the inner tube cross-section). Let be the polar angle of that pixel. In the unfolded rectangular image, the _____ Line number The column pixels correspond to the polar coordinate positions in the original image as follows: .

[0026] in, and These represent the minimum and maximum radial radii of the inner tube's annular region, respectively. and These are the height and width of the unfolded rectangular image, respectively. The original image coordinates obtained from the above mapping ( , The coordinates are usually not integers. Bilinear interpolation is used to estimate the pixel grayscale value at non-integer coordinates. This involves a weighted average of the grayscale values ​​of the four nearest integer pixels around that coordinate, with the weights inversely proportional to the distance. Bilinear interpolation has a moderate computational cost and provides better interpolation results than nearest-neighbor interpolation, effectively reducing the jagged edges of the unfolded image while maintaining processing speed.

[0027] The generated panoramic image of the inner tube is retrieved, and a rectangular sliding window of fixed size is used to perform a region-by-region extraction operation on the panoramic image. Starting from the beginning of the panoramic image, the sliding window continuously moves along the corresponding horizontal direction in the circumferential direction at a fixed step size to traverse the entire panoramic image. Due to the imaging characteristics of the original curved surface of the inner tube, the same surface position in the panoramic image corresponds to different observation reflection angles at different window positions. Each time the sliding window completes a translation and pauses, it captures all the pixels in the current coverage area of ​​the window to generate an independent block image. The sliding window continues to translate and capture until all pixel areas of the panoramic image are completely traversed. All the block images captured by the sliding window are uniformly organized and collected according to the order of the window translation and traversal. The collected block images form a multi-angle image sequence of the inner tube.

[0028] It is important to note that a geometric conversion relationship must be established beforehand between the horizontal coordinate position of the sliding window center point in the panoramic unfolded image and the equivalent observation azimuth angle of the inner tube surface. Specifically, the inner tube's annular radius and the total width of the panoramic unfolded image are obtained first; this total width corresponds to the inner tube's complete circumference. The horizontal coordinate value of the sliding window center is divided by the total width of the panoramic unfolded image, and the resulting ratio is multiplied by 360 degrees to obtain the equivalent observation azimuth angle corresponding to the current window. When the sliding window center is located at two different horizontal coordinate positions, the above conversion relationship yields two corresponding equivalent observation azimuth angles. Subtracting the two azimuth angles gives the difference in the incident angle of illumination at the same surface location in the two frames. This mapping relationship ensures that the "multiple angles" upon which subsequent differential synthesis operations are based have clear and quantifiable physical and geometric meaning.

[0029] The size and translation step of the sliding window are set according to the following rules: The size of the sliding window is a fixed pixel window pre-calibrated based on the physical size of the minimum effective defect on the inner tube surface and the imaging resolution. The window width corresponds to the imaging coverage area in the circumferential direction of the inner tube, and the window height corresponds to the imaging coverage area in the axial direction of the inner tube. The translation step of the window in the circumferential direction is set to one-quarter of the window width to ensure sufficient pixel overlap between adjacent windows, so that the same surface position can be recorded from multiple different observation angles. The window starts from the beginning of the panoramic unfolded image and is translated successively in the circumferential direction until it covers the entire pixel area of ​​the panoramic unfolded image.

[0030] It should be noted that, due to the curved surface of the inner tube, the normal directions originally located at different positions on the curved surface are mapped to different orientations within the plane during the generation of the panoramic image. When the sliding window traverses different regions of the panoramic image, the microscopic geometric orientation of the same physical location on the inner tube surface relative to the image acquisition optical axis changes. Therefore, the different segmented images captured by the sliding window essentially record the reflected light intensity information of that surface location at different equivalent observation angles. Each segmented image within the angle image sequence corresponds to a different angle imaging of a local area of ​​the tire tread, completely inheriting all the tire tread imaging information carried by the panoramic image.

[0031] In summary, by acquiring complete images of the entire inner tube surface, the circular unfolding and time-series stitching can regularize the surface imaging shape. The resulting image sequence divided by the sliding window can perfectly adapt to subsequent image processing procedures such as light intensity comparison and feature extraction, ensuring the smooth progress of the back-end defect identification and judgment process, and laying a solid foundation for the entire screening scheme.

[0032] S2. Perform contrast difference synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate the differential contrast image of the inner tube; In this embodiment of the invention, the step of performing contrast differential synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate a differential contrast image of the inner tube includes: Extract the pixel values ​​of each frame in the multi-angle image sequence at the same surface position to obtain the multi-angle pixel value sequence of the inner tube; Extreme value detection is performed on the multi-angle pixel value sequence to determine the maximum and minimum pixel values ​​of the inner tube; The initial difference value of the inner tube is obtained by performing a difference operation between the maximum pixel value and the minimum pixel value; The initial difference value is mapped to a preset gray value range to obtain the normalized difference value of the inner tube; By reorganizing the spatial positions of the normalized difference values ​​according to the actual spatial arrangement order of the inner tube, a difference contrast image of the inner tube is generated.

[0033] Specifically, all frame images in the multi-angle image sequence are generated by cropping the same panoramic unfolded image of the inner tube through a sliding window. Each frame image shares the same pixel coordinate system as the panoramic unfolded image. Each coordinate point corresponds to the same fixed physical position on the surface of the inner tube. All pixel coordinate points are traversed row by row and column by column. For each coordinate point, the pixel grayscale value of each frame image at that coordinate point is extracted in the order of the frames in the multi-angle image sequence. All grayscale values ​​extracted from the same coordinate point are collected into a set of corresponding data. The grayscale value sets corresponding to all coordinate points together constitute the multi-angle pixel value sequence of the inner tube.

[0034] For each group of grayscale values ​​corresponding to the same surface location in the multi-angle pixel value sequence, the comparison is performed sequentially from the first grayscale value in the group. During the comparison, the maximum and minimum values ​​encountered are continuously recorded. After all grayscale values ​​in the group have been compared, the maximum value recorded is the maximum pixel value at that surface location, and the minimum value recorded is the minimum pixel value at that surface location. After comparing all surface locations, the maximum and minimum pixel values ​​corresponding to each surface location of the inner tube are obtained.

[0035] For each surface position of the inner tube, the maximum pixel value corresponding to that position is subtracted from the minimum pixel value corresponding to the same position. The result of the subtraction is the initial difference value of that surface position. After the subtraction calculation is completed for all surface positions of the inner tube, the initial difference value of all positions is bound and stored with their respective surface position coordinates to form a complete set of initial difference values ​​for the inner tube.

[0036] The preset grayscale value range is a fixed range pre-defined based on the numerical range of a standard grayscale image. First, the largest and smallest initial difference values ​​are selected from all initial difference values. The smallest initial difference value corresponds to the lower limit of the grayscale value range, and the largest initial difference value corresponds to the upper limit of the grayscale value range. Each initial difference value is converted to the preset grayscale value range according to a linear proportional rule. During the conversion process, the relative size relationship between all initial difference values ​​remains unchanged. After the conversion, each initial difference value corresponds to a value within the preset grayscale value range. This value is the normalized difference value of the corresponding surface position. The normalized difference values ​​of all positions together constitute the set of normalized difference values ​​of the inner tube.

[0037] It is important to note that the preset grayscale value range specifically adopts the numerical range of a standard 8-bit grayscale image, i.e., a lower limit of zero and an upper limit of 255. During normalization mapping, the overall maximum and minimum values ​​are first selected from all initial difference values. Then, a linear proportional transformation is performed on each initial difference value: the current initial difference value is subtracted from the overall minimum value, the resulting difference is divided by the difference between the overall maximum and the overall minimum value, and then multiplied by 255. The product is the normalized difference value corresponding to that position. Through this transformation, the relative magnitudes of all initial difference values ​​remain unchanged, and all normalized difference values ​​are confined to the grayscale range of zero to 255.

[0038] Based on the row and column spatial arrangement order of the inner tube surface in the panoramic unfolded image, and maintaining the row and column arrangement rules completely consistent with the original panoramic unfolded image, the normalized difference value corresponding to each surface position is sequentially filled into the corresponding coordinate pixel position. The rows are arranged sequentially in the original order, and the columns are aligned sequentially in the original order. After all pixel positions are filled, a complete grayscale image is formed. This image is the differential contrast image of the inner tube. The grayscale value of each pixel in the image directly corresponds to the normalized difference value at the corresponding position. The overall size of the image is completely consistent with the panoramic unfolded image of the inner tube, fully presenting the contrast difference calculation results of the entire surface of the inner tube.

[0039] In summary, multi-view pixel differences highlight surface brightness variations, effectively amplifying the visibility of subtle imperfections and unifying imaging standards after numerical normalization. The well-organized contrast image can accurately distinguish between normal and abnormal areas, providing clear and reliable image data for subsequent core steps such as connected component extraction and morphological feature quantification, thus ensuring the overall accuracy of detection and judgment.

[0040] S3. Connected component extraction is performed on the candidate regions of surface anomalies in the differential contrast image to obtain the candidate defect connected components of the inner tube, and morphological quantization is performed on the candidate defect connected components to obtain the morphological feature parameters of the inner tube. In this embodiment of the invention, the step of extracting connected components from the candidate surface anomaly regions in the differential contrast image to obtain the candidate defect connected components of the inner tube includes: Based on the circumferential diameter and axial length of the inner tube, the differential contrast image is divided into equidistant grids to obtain local detection sub-regions of the inner tube; The grayscale values ​​of pixels within the local detection sub-region are statistically analyzed and the outlier degree is measured to obtain the local mean and local standard deviation of the local detection sub-region. The local detection sub-region is binarized and segmented using the local mean and the local standard deviation to obtain a local binarized sub-map of the local detection sub-region. The local binarized sub-images are stitched together and fused, and morphological opening is performed on the fused image to obtain the global binarized image of the inner tube. Connected component labeling is performed on the global binarized image to obtain the edge pixel set of the inner tube, and the pixel area of ​​the region enclosed by the edge pixel set is calculated to obtain the actual pixel area of ​​the inner tube. Connected regions whose actual pixel area is less than a preset area threshold are discarded as pseudo-noise, and the remaining connected regions are retained as candidate defect connected regions of the inner tube.

[0041] The step of performing morphological quantization on the connected components of the candidate defects to obtain the morphological feature parameters of the inner tube includes: Extract the minimum bounding ellipse of the candidate defect connected domain, and obtain the length of the major axis of the minimum bounding ellipse; The directional deflection parameters of the candidate defect connected domain are obtained by directional calculation of the angle between the length of the major axis and the axial direction of the inner tube. The edge pixel set of the candidate defective connected region is fitted with a second moment to obtain the dispersion parameter and elongation parameter of the candidate defective connected region. Extend a predetermined number of pixels outward along the edge normal direction of the candidate defect connected region, and collect the grayscale values ​​of the pixels in the extended region; Based on the edge normal direction, the gradient magnitude of the gray value is analyzed by gradient decay to obtain the edge sharpness parameter of the candidate defect connected domain; The morphological feature parameters of the inner tube are obtained by combining the actual pixel area, the orientation deflection parameter, the dispersion parameter, the elongation parameter, and the edge sharpness parameter.

[0042] Specifically, based on the actual circumferential diameter and axial length of the inner tube, and combined with the imaging resolution pre-calibrated in the image acquisition stage, the pixel conversion relationship corresponding to a unit physical length is determined. Then, according to the pre-set fixed physical side length standard, the differential contrast image is uniformly cut along the circumferential unfolding direction and the axial direction to divide it into several rectangular sub-regions with completely consistent physical areas. All the divided rectangular sub-regions are uniformly used as local detection sub-regions of the inner tube.

[0043] For each local detection sub-region, the grayscale values ​​of all pixels within the sub-region are collected one by one. The average grayscale value of the sub-region is calculated after summing all the grayscale values ​​as the local mean. Then, the grayscale dispersion value of the sub-region is calculated based on the degree of deviation of each pixel's grayscale value from the local mean as the local standard deviation. After all local detection sub-regions have been calculated, their respective local mean and local standard deviation are obtained.

[0044] For each local detection sub-region, the local mean of the sub-region is used as a benchmark, and 1.5 or 2 times the local standard deviation of the sub-region is superimposed. The superposition result is used as the gray-level segmentation threshold for the sub-region. Pixels with gray-level values ​​higher than the segmentation threshold are uniformly marked as foreground pixels, and pixels with gray-level values ​​lower than the segmentation threshold are uniformly marked as background pixels. After the segmentation is completed, a single sub-image containing only foreground and background pixels is generated. This sub-image is the local binarized sub-image of the corresponding local detection sub-region.

[0045] Following the original grid division and arrangement order of the differential contrast image, all local binarized sub-images are sequentially placed back into their original positions and seamlessly stitched together to obtain an initial binarized image with the same size as the differential contrast image. Then, a morphological opening operation is performed on the initial binarized image using a pre-set square structuring element. During the operation, the foreground pixel area is first eroded to remove small protrusions at the edges, and then the eroded foreground area is dilated to restore the main outline. After processing, the global binarized image of the inner tube is obtained.

[0046] A pre-defined eight-adjacency determination rule is used to perform connected component labeling on all foreground pixels in the globally binarized image. Adjacent foreground pixels are grouped into the same independent connected region. All pixels on the outermost side of each connected region are extracted to form the edge pixel set of that region. The total number of all foreground pixels contained in each connected region is then counted. This total number is the actual pixel area of ​​the corresponding connected region. Each connected region has its own edge pixel set and actual pixel area.

[0047] The preset area threshold is a fixed pixel area value calculated based on the physical size of the smallest effective defect on the inner tube surface and the imaging resolution. The actual pixel area of ​​each connected region is compared with the preset area threshold in turn. Connected regions whose actual pixel area is less than the area threshold are identified as pseudo noise and completely removed from the global binarized image. All remaining connected regions are uniformly used as candidate defect connected regions for the inner tube.

[0048] It should be noted that the preset area threshold is calculated based on the minimum detectable defect physical diameter specified in the inner tube production line quality inspection standards, combined with the pre-calibrated imaging resolution of the current image acquisition system. Specifically, the minimum detectable defect diameter ranges from 0.5 mm to 1.0 mm, and the imaging resolution is ten pixels per millimeter. During the conversion, the physical radius of the minimum detectable defect is first calculated, then the area of ​​the circle within that radius is calculated, and finally, this physical area is multiplied by the square of the imaging resolution to obtain the corresponding pixel area threshold. In a preferred embodiment of the present invention, when the minimum detectable defect diameter is 0.8 mm and the resolution is ten pixels per millimeter, the calculated area threshold falls within the range of twenty to eighty pixels.

[0049] Specifically, for each candidate defective connected region, the coordinates of all internal and edge pixels contained in the connected region are retrieved. The center position, major and minor axis dimensions, and deflection angle of the ellipse are adjusted with the constraint of completely encompassing all pixels. The ellipse with the smallest coverage area is selected as the minimum bounding ellipse of the candidate defective connected region. At the same time, the pixel length value corresponding to the major axis diameter of the ellipse is obtained as the major axis length of the minimum bounding ellipse.

[0050] The axial direction of the inner tube corresponds to a pre-calibrated fixed vertical reference direction in the differential contrast image. This reference direction corresponds one-to-one with the physical axis of the inner tube. Based on this vertical reference direction, the minimum angle between the extension direction of the minor circumscribed ellipse and the reference direction is calculated. This angle is used as the direction deflection parameter of the candidate defect connected domain.

[0051] For all pixels in the candidate defective connected region, the centroid position of the connected region is first determined by the average position of all pixel coordinates. Then, the distribution shape is fitted based on the position distribution of all pixels relative to the centroid. The dispersion parameter of the connected region is obtained by the overall dispersion of all pixels from the centroid. The elongation parameter of the connected region is obtained by the length ratio of the longest extension direction to the shortest extension direction of the pixel distribution.

[0052] It is important to note that, for all pixel coordinates of a candidate defective connected component, the centroid of the component is first determined by averaging all pixel coordinates. Then, a covariance matrix of all pixels relative to this centroid is constructed. This covariance matrix contains the second-order central moments and mixed central moments of the pixels in the row and column directions. Solving this covariance matrix yields two eigenvalues: the larger eigenvalue represents the dispersion along the principal extension direction of the pixel distribution, and the smaller eigenvalue represents the dispersion perpendicular to the principal extension direction. The larger eigenvalue is divided by the smaller eigenvalue, and the resulting ratio is used as the elongation parameter of the connected component. Simultaneously, the sum of the larger and smaller eigenvalues ​​is calculated, and the square root of this sum is used as the dispersion parameter of the connected component.

[0053] The preset number of pixels is a fixed number of pixels pre-calibrated based on the image acquisition resolution and the width of the typical edge transition zone of the rubber surface defect. For each edge pixel in the connected region of the candidate defect, the edge tangent direction is first determined based on the direction of the adjacent edge pixels. Then, the direction perpendicular to the tangent and pointing outward of the connected region is calculated as the edge normal direction of the point. The preset number of pixel positions are extended outward along the normal direction. The gray values ​​of all extended points are collected one by one in the differential contrast image. The gray values ​​obtained by all edge point extensions together constitute the gray values ​​of the pixels in the extended region.

[0054] Specifically, the preset number of pixels is pre-calibrated based on the effective width of the point spread function of the image acquisition system. In a preferred embodiment of the present invention, the preset number is three to five pixels, that is, three to five pixels are expanded outward along the edge normal direction, and the gray values ​​at these points are collected for subsequent edge sharpness analysis.

[0055] For each edge normal direction, the gray value of the extended region is collected, and the gray value change of adjacent pixels is compared sequentially from the inside to the outside of the connected component along the normal. The pixel distance required for the gray value to transition from the gray value inside the connected component to the background gray value is calculated. Then, the average value of the distance values ​​corresponding to all edge points is taken, and the quantized value corresponding to the average distance is used as the edge sharpness parameter of the candidate defective connected component.

[0056] The actual pixel area, orientation deflection parameter, dispersion parameter, elongation parameter, and edge sharpness parameter corresponding to the same candidate defect connected region are bound and collected one by one to form a set of feature parameters corresponding to the connected region. After all the feature parameter sets corresponding to the candidate defect connected regions are summarized, they together constitute the morphological feature parameters of the inner tube.

[0057] In summary, the system adapts to the overall size of the inner tube through block detection, achieves accurate segmentation based on local grayscale data, and optimizes image quality through morphological processing. Area filtering removes irrelevant noise, accurately pinpointing suspected defect areas and effectively reducing interference from invalid data. This clears the way for subsequent feature parameter extraction and accurate defect determination, ensuring the stability of the entire detection process.

[0058] Parameters are extracted from multiple dimensions, including size, orientation, distribution pattern, and edge state, comprehensively encompassing various appearance characteristics of defects. The combination of multiple features forms complete characterization data, clearly distinguishing differences between different defect categories. This provides sufficient basis for subsequent conditional comparison to identify genuine defects, significantly improving overall recognition and differentiation capabilities.

[0059] S4. Compare the morphological feature parameters with the preset defect judgment conditions item by item to determine the actual defect area of ​​the inner tube, and record the position coordinates and defect type mark of the actual defect area in the inner tube. In this embodiment of the invention, the step of comparing the morphological feature parameters with preset defect judgment conditions item by item to determine the actual defect area of ​​the inner tube, and recording the position coordinates and defect type label of the actual defect area in the inner tube, includes: The morphological feature parameters are compared with the preset defect judgment conditions one by one to obtain the independent comparison results of the inner tube, and the independent comparison results are weighted and evaluated to obtain the comprehensive score value of the inner tube. The comprehensive score is compared with a preset comprehensive judgment threshold. When the comprehensive score is not less than the comprehensive judgment threshold, the candidate defect connected region is determined to be the real defect region of the inner tube; otherwise, it is determined to be a false defect region and is removed. The actual defect area is matched with a preset defect type determination table to determine the defect type label of the actual defect area; Extract the centroid coordinates and the vertex coordinates of the smallest bounding rectangle of the actual defect area, and map the centroid coordinates and the vertex coordinates of the smallest bounding rectangle to the spatial coordinate system of the inner tube to obtain the position coordinates of the actual defect area.

[0060] Specifically, the preset defect judgment conditions are sub-judgment rules formulated in advance by statistically analyzing the morphological feature data of a large number of real inner tube surface defect samples. Each morphological feature parameter corresponds to an independent numerical judgment interval and a pre-defined weight ratio. The weight ratio is pre-determined based on the effectiveness of each feature in distinguishing between real defects and pseudo-defects. For each candidate defect connected region corresponding to the morphological feature parameter, each feature parameter is compared with the judgment interval of the corresponding item to obtain the independent comparison result of that item. The independent comparison result is presented in the form of a conformity value. Then, according to the weight ratio of each parameter, the conformity values ​​of all independent comparison results are weighted and summed to obtain the comprehensive score value corresponding to the candidate defect connected region.

[0061] It should be noted that the morphological feature parameters include at least the actual pixel area, elongation parameter, and edge sharpness parameter. The preset defect judgment conditions include independent judgment intervals for each parameter and their corresponding weight percentages. Specifically, the weight of the actual pixel area is set to 0.4, the weight of the elongation parameter is set to 0.35, and the weight of the edge sharpness parameter is set to 0.25, with the sum of the three weights being 1. For each candidate defect connected region, each feature parameter is first compared with its corresponding judgment interval to obtain a conformity score for each parameter. Then, each conformity score is multiplied by its corresponding weight, and the products are summed to obtain the comprehensive score value of the connected region. The preset comprehensive judgment threshold is set to 0.6. When the comprehensive score value is greater than or equal to 0.6, the candidate defect connected region is judged as a real defect region; otherwise, it is judged as a false defect region and discarded.

[0062] The preset comprehensive judgment threshold is a defect judgment boundary value pre-determined based on the pass standard of the inner tube finished product's factory inspection and historical test data. The comprehensive score value corresponding to each candidate defect connected region is compared with the preset comprehensive judgment threshold. When the comprehensive score value is not less than the comprehensive judgment threshold, the corresponding candidate defect connected region is judged as a real defect area of ​​the inner tube. When the comprehensive score value is less than the comprehensive judgment threshold, the corresponding candidate defect connected region is judged as a pseudo defect area and is completely removed from the set of all candidate defect connected regions, and will no longer participate in the subsequent processing.

[0063] The preset defect type judgment table is a classification and comparison table pre-established based on the typical morphological feature combination rules corresponding to different types of inner tube surface defects such as scratches, bubbles, and missing glue. Each type of defect in the table corresponds to a set of exclusive feature value range combinations. The various morphological feature parameters of the actual defect area are matched one by one with the feature range combinations of each type of defect in the table. The number of feature matching items for each category is counted. The defect category with the most matching items is the defect type label corresponding to the actual defect area.

[0064] It is important to note that the preset defect type determination table pre-loads the characteristic parameter combination ranges corresponding to three typical defect types: scratches, bubbles, and missing glue. The specific determination rules are as follows: when the elongation parameter of a real defect area is greater than or equal to 3.0 and the edge sharpness parameter is greater than or equal to 0.8, the defect is marked as a scratch; when the elongation parameter of a real defect area is less than 2.0, the dispersion parameter is greater than 5 pixels, and the edge sharpness parameter is less than 0.5, the defect is marked as a bubble; when the elongation parameter of a real defect area is between 2.0 and 3.0, and the edge sharpness parameter is greater than or equal to 0.6, the defect is marked as missing glue. Each morphological characteristic parameter of the real defect area to be determined is matched against the above rules one by one. The type corresponding to the rule with the highest matching degree is the defect type label for that real defect area.

[0065] The centroid coordinates and minimum bounding rectangle vertex coordinates of each real defect area in the differential contrast image are extracted. The spatial coordinate system of the inner tube is a pre-established entity coordinate system that corresponds one-to-one with the physical size of the inner tube. This coordinate system has the center of the end face at one end of the inner tube as the origin, the inner tube axis as the vertical axis, and the circumferential unfolding direction as the horizontal axis. The coordinate system has a fixed conversion ratio from the pixel coordinates calibrated in the image acquisition stage to the physical spatial coordinates. Combining the correspondence between the coordinate origin of the differential contrast image and the coordinate origin of the inner tube entity, the centroid coordinates and minimum bounding rectangle vertex coordinates at the pixel level are converted one by one into physical coordinates under the inner tube spatial coordinate system to obtain the position coordinates of the real defect area.

[0066] In summary, multi-feature weighted evaluation improves the accuracy of judgment, reliably filters out false defects, and accurately locates the true defects. Based on preset standards, it quickly completes defect category classification and simultaneously achieves precise coordinate positioning. The resulting judgment results and location information are detailed and standardized, directly supporting subsequent topology mapping and quality rating work, ensuring the orderly progress of the entire inspection process.

[0067] S5. Perform topological reconstruction on the location coordinates and the defect type markers to obtain the defect distribution map of the inner tube, and make a joint decision on the defect distribution map based on the number and type of the actual defect areas to obtain the quality grade of the inner tube; In this embodiment of the invention, the step of performing topological reconstruction on the position coordinates and the defect type marker to obtain the defect distribution map of the inner tube includes: The position coordinates are mapped to the two-dimensional unfolded plane coordinate system of the inner tube to obtain the plane coordinates of the actual defect area; Based on the planar coordinates, the spatial distance between the actual defective areas is calculated to obtain the spatial distance value of the actual defective areas; By using the spatial distance value, adjacency determination and connectivity analysis are performed on the actual defect area to obtain the spatial adjacency relationship diagram of the actual defect area; Extract the defect clusters from the spatial adjacency graph, and count the types and number of defects in the defect clusters to obtain the cluster type composition information of the defect clusters; The cluster type composition information is associated with the spatial distribution location of the defect cluster to generate a defect distribution map of the inner tube.

[0068] The process of jointly determining the quality grade of the inner tube by considering the quantity and type of the actual defective areas and performing a joint decision on the defect distribution map includes: The total number of real defect areas in the defect cluster is counted to obtain the defect density value of the real defect area; The cluster type composition information is subjected to key type identification to obtain the type identification result of the real defect area; Based on the defect density value and the type identification result, the cluster hazard quantification of the defect distribution map is performed to obtain the cluster hazard quantification value of the defect distribution map; The quality level of the inner tube is obtained by globally weighting and aggregating the cluster hazard quantification values ​​and mapping them to a level.

[0069] Specifically, the two-dimensional unfolded plane coordinate system of the inner tube is completely unified with the coordinate system of the differential contrast image generated in the early stage. The origin, axis and circumferential extension direction of the two are in one-to-one correspondence. The position coordinates of the inner tube space coordinate system that have been converted for each real defect area are retrieved. Based on the fixed correspondence between the spatial physical size and the planar pixel size pre-calibrated in the imaging process, the position coordinates in the spatial coordinate system are completely converted to the two-dimensional unfolded plane coordinate system. After the conversion, the exclusive planar coordinates of each real defect area are obtained.

[0070] Retrieve the planar coordinates corresponding to each of the real defect areas. Select any two different real defect areas in sequence, and read the values ​​corresponding to the horizontal and vertical axes of the planar coordinates of the two areas respectively. Calculate the interval length between the center points of the two areas according to the calculation logic of the distance between two points of the planar coordinates. The calculated interval length is the spatial distance value between the two real defect areas. After traversing all the real defect areas of pairwise combinations, the spatial distance value corresponding to all paired combinations is obtained.

[0071] A unified adjacency judgment distance standard is pre-set. This standard is fixed in advance according to the process requirements for judging inner tube defect clusters. The spatial distance value corresponding to each pair of real defect areas is compared with the adjacency judgment distance standard. Two real defect areas with a spatial distance value less than the standard are judged as spatially adjacent. Two real defect areas with a spatial distance value greater than or equal to the standard are judged as independent and have no adjacency relationship. The adjacency judgment results of all real defect areas are uniformly compiled and drawn to form a spatial adjacency relationship diagram of real defect areas that can intuitively reflect the distance relationship of each defect area.

[0072] It should be noted that the unified adjacency determination distance standard is preset based on the actual operating conditions of the inner tube and the process requirements for judging appearance quality. In a preferred embodiment, this adjacency determination distance standard is set to 15 millimeters on the two-dimensional unfolded plane. When converted to pixel distance, combined with the current imaging resolution (10 pixels per millimeter), the corresponding pixel distance is 150 pixels. When the spatial distance between two actual defect areas is less than 15 millimeters (i.e., 150 pixels), the two defects are determined to be spatially adjacent and are grouped into the same defect cluster.

[0073] The entire process traverses all real defect regions within the spatial adjacency graph. Multiple real defect regions that are adjacent to each other are grouped into the same defect cluster, while single real defect regions that are not adjacent to each other are grouped into independent defect clusters. For each defect cluster, the total number of real defect regions contained within the cluster is counted, and the defect type marker bound to each real defect region within the cluster is identified. All non-repeating defect types appearing within the cluster are summarized, and the total number of defects and the types of defects contained in the cluster are uniformly collected and recorded to form the cluster type composition information of the corresponding defect cluster.

[0074] The cluster type composition information corresponding to each defect cluster is retrieved. At the same time, the planar coordinates corresponding to all real defect areas within the defect cluster are retrieved to determine the overall spatial distribution of the defect cluster. The cluster type composition information of the same defect cluster is bound and associated with the spatial distribution of the cluster. According to the arrangement rules of the two-dimensional unfolded planar coordinate system, all the associated defect clusters are completely drawn on the same plane. Finally, a defect distribution map of the inner tube that can completely display the location, quantity, and type of all defect clusters is generated.

[0075] Specifically, the process involves traversing all the defect clusters within the defect distribution map, counting the number of actual defect areas contained in each defect cluster, summing up the counts of actual defect areas within all defect clusters to obtain the total number of actual defect areas covered by the entire defect distribution map, and then performing a corresponding conversion based on the total unfolded area of ​​the inner tube corresponding to the two-dimensional unfolded plane coordinate system. This conversion yields the number of actual defect areas distributed per unit plane area, and the number of defects per unit area is the defect density value of the actual defect area.

[0076] A pre-defined list of key defect types is prepared, which includes categories of defects that severely affect the performance of inner tubes, such as missing glue, cracks, and deep air bubbles. The cluster type composition information corresponding to each defect cluster is retrieved sequentially, and the defect type markers recorded in the cluster are compared one by one with the pre-defined list of key defect types. All defect types within the cluster that fall within the list are marked, and the corresponding quantity of each type of key defect within the cluster is recorded. After all comparisons and markings are completed, the results are summarized to form a complete identification result of the type of real defect area.

[0077] Pre-set corresponding basic hazard scores for defect density values ​​in different intervals and for different combinations of key defects. Match the defect density values ​​to the corresponding intervals to read the basic scores, and then combine the types and quantities of key defects in the type identification results to add the corresponding hazard bonus scores. Add the basic scores and all bonus scores together to obtain the hazard score corresponding to a single defect cluster. After traversing all defect clusters in the defect distribution map and calculating their respective hazard scores, all scores are uniformly collected and summarized. The overall score obtained is the cluster hazard quantification value of the defect distribution map.

[0078] The inner tube is pre-divided into multiple fixed quality ranges according to the factory quality inspection specifications. Each quality range corresponds to a specific product quality grade. At the same time, a pre-set global aggregation weight is assigned to the hazard score of each defect cluster. The hazard scores corresponding to all defect clusters are accumulated and aggregated according to their respective weights to obtain the total hazard score of the entire inner tube. The total hazard score is then compared and matched with the boundary standards of each quality range in turn. After matching the corresponding range, the product grade identifier bound to that range is read. This identifier is the quality grade of the inner tube.

[0079] It should be noted that the mapping of the quality grades is carried out in accordance with the relevant national standards for the appearance quality of rubber products or the internal quality inspection specifications formulated by the enterprise. In a specific implementation plan, three quality grade ranges are pre-defined. The specific judgment logic is as follows: when the defect density value on the entire defect distribution map is less than or equal to 0.1 defects per square centimeter, and none of the defect clusters contain key types of defects such as missing rubber and cracks, the inner tube is judged to be a Grade 1 product; when the defect density value is less than or equal to 0.5 defects per square centimeter, or when there is only a single repairable defect, the inner tube is judged to be a Grade 2 product; when the defect density value is greater than 0.5 defects per square centimeter, or when any defect cluster contains a through-crack or severe missing rubber, the inner tube is judged to be a substandard product.

[0080] When calculating the hazard quantification value of a cluster, each defect cluster is quantified separately. First, the corresponding base hazard score is read based on the defect density range of the cluster. Then, a corresponding hazard bonus score is added based on the types and quantities of key defect types contained in the cluster. The base score and the bonus score are added together to obtain the hazard score of the cluster. After all defect clusters have been calculated, a pre-set global aggregation weight is assigned to each cluster. This weight is proportional to the coverage area of ​​the cluster on the unfolded plane. The hazard scores of all clusters are multiplied by their respective weights, and the products are summed to obtain the global weighted aggregation score of the entire defect distribution map. Finally, this global weighted aggregation score is compared with the boundary standards of the three quality level ranges mentioned above. The level identifier corresponding to the matched range is the final quality level of the inner tube.

[0081] In summary, by standardizing and converting defect coordinates, the spatial relationships between defects can be accurately determined, and various defect clusters can be rationally classified. A distribution map is generated by comprehensively summarizing the number and category information of clusters and combining it with their distribution locations, intuitively displaying the overall defect distribution. This provides comprehensive and intuitive data support for subsequent overall quality assessment, thus improving the overall inspection and evaluation system.

[0082] By combining the density of defect distribution with the comprehensive evaluation of defective product categories, the overall severity of harm can be accurately quantified. Utilizing a weighted approach to integrate various evaluation data, product grading can be quickly completed. The rating results accurately reflect the actual product quality, and the judgment standards are unified and objective, effectively ensuring the accurate implementation of subsequent sorting and classification work and perfecting the entire quality inspection process.

[0083] S6. Map the quality grade to a screening control command to drive the sorting execution mechanism to perform real-time screening of the inner tube.

[0084] In this embodiment of the invention, mapping the quality grade to screening control instructions to drive the sorting execution mechanism to perform real-time screening of the inner tube includes: According to the preset quality grade classification rules, the quality grades are classified to determine the grade category corresponding to the inner tube; Based on the grade category, a preset filtering strategy mapping table is matched to determine the target filtering strategy corresponding to the grade category, and a corresponding filtering control instruction is generated based on the target filtering strategy. The screening control command is output to the sorting execution mechanism, which drives the sorting execution mechanism to perform a screening action on the inner tube that matches the quality grade.

[0085] Specifically, the preset quality grade classification rules are based on the grading standards compiled in advance according to the quality inspection specifications for rubber inner tubes. The rules define the exclusive grade category boundary for each quality grade. The quality grade calculated for the current inner tube is compared and matched one by one with the boundaries of each grade recorded in the preset quality grade classification rules. After the matching is completed, the unique grade category to which the inner tube belongs is locked.

[0086] The preset screening strategy mapping table is a pre-compiled lookup table. Each grade category in the table is bound to a unique set of target screening strategies. The grade category corresponding to the current inner tube is retrieved, and the target screening strategy that completely matches the grade category is searched in the screening strategy mapping table. Based on the equipment action sequence, execution component action form, and action triggering time recorded in the target screening strategy, a unique corresponding screening control instruction is generated. The screening control instruction completely records all the action information that the sorting mechanism needs to perform.

[0087] A real-time signal communication channel is established between the sorting execution mechanism and the image processing host through a fixed signal transmission line. The generated screening control command is transmitted to the signal receiving unit built into the sorting execution mechanism through the signal transmission channel. After receiving the screening control command, the sorting execution mechanism reads the action information recorded in the command and drives the corresponding mechanical execution parts inside the mechanism to complete the corresponding pushing, diversion, sorting and storage screening actions according to the command requirements, thus completing the grading and sorting operation for the current inner tube.

[0088] In summary, by quickly classifying products according to established standards and accurately generating corresponding control commands through preset strategy tables, sorting equipment can be smoothly linked to complete the corresponding sorting actions. This achieves seamless integration between quality inspection and physical sorting, automating the entire sorting process, effectively reducing manual intervention, and significantly improving the overall efficiency of inspection and sorting operations.

[0089] like Figure 2 As shown, Embodiment 2 is a functional block diagram of a machine vision-based real-time screening system for inner tube surface defects provided in an embodiment of the present invention.

[0090] The machine vision-based real-time defect screening system 100 for inner tube surfaces described in this invention can be installed in an electronic device. Depending on the functions implemented, the machine vision-based real-time defect screening system 100 may include an image acquisition module 101, a contrast difference module 102, a parameter quantization module 103, a defect marking module 104, a grade determination module 105, and a real-time screening module 106. The module described in this invention can also be called a unit, referring to a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0091] In this embodiment, the functions of each module / unit are as follows: The image acquisition module 101 is used to perform a circumferential unfolding scan of the surface of the inner tube to obtain a multi-angle image sequence of the inner tube. The contrast difference module 102 is used to perform contrast difference synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate a differential contrast image of the inner tube. The parameter quantization module 103 is used to extract the connected components of the surface anomaly candidate region in the differential contrast image to obtain the candidate defect connected components of the inner tube, and to perform morphological quantization on the candidate defect connected components to obtain the morphological feature parameters of the inner tube. The defect marking module 104 is used to compare the morphological feature parameters with the preset defect judgment conditions item by item to determine the actual defect area of ​​the inner tube, and record the position coordinates of the actual defect area in the inner tube and the defect type mark. The grade determination module 105 is used to perform topological reconstruction on the position coordinates and the defect type marker to obtain the defect distribution map of the inner tube, and to make a joint decision on the defect distribution map based on the number and type of the actual defect areas to obtain the quality grade of the inner tube. The real-time screening module 106 is used to map the quality grade into screening control instructions, driving the sorting execution mechanism to perform real-time screening of the inner tube.

[0092] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0093] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0094] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0095] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0096] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A machine vision-based real-time screening method for inner tube surface defects, characterized in that, The method includes: S1. Perform a circumferential unfolding scan on the surface of the inner tube to obtain a multi-angle image sequence of the inner tube; S2. Perform contrast difference synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate the differential contrast image of the inner tube; S3. Connected component extraction is performed on the candidate regions of surface anomalies in the differential contrast image to obtain the candidate defect connected components of the inner tube, and morphological quantization is performed on the candidate defect connected components to obtain the morphological feature parameters of the inner tube. S4. Compare the morphological feature parameters with the preset defect judgment conditions item by item to determine the actual defect area of ​​the inner tube, and record the position coordinates and defect type mark of the actual defect area in the inner tube. S5. Perform topological reconstruction on the location coordinates and the defect type markers to obtain the defect distribution map of the inner tube, and make a joint decision on the defect distribution map based on the number and type of the actual defect areas to obtain the quality grade of the inner tube; S6. Map the quality grade to a screening control command to drive the sorting execution mechanism to perform real-time screening of the inner tube.

2. The real-time screening method for inner tube surface defects based on machine vision as described in claim 1, characterized in that, The step of performing a circumferential unfolding scan of the inner tube surface to obtain a multi-angle image sequence of the inner tube includes: The inner tube is mounted on the rotary drive assembly, and the inner tube is driven to rotate at a constant speed around the axis of the rotary drive assembly. During the rotation process, the surface images of the inner tube are acquired at equal angular intervals along the circumference of the inner tube by the image acquisition component, resulting in a circumferential image sequence of the inner tube; A circular unfolding mapping is performed on each frame of the circumferential image sequence, and the mapping results are spliced ​​in time to obtain the panoramic unfolded image of the inner tube. A sliding window extraction is performed on the panoramic unfolded image to obtain a multi-angle image sequence of the inner tube.

3. The real-time screening method for inner tube surface defects based on machine vision as described in claim 1, characterized in that, The step of performing contrast differential synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate a differential contrast image of the inner tube includes: Extract the pixel values ​​of each frame in the multi-angle image sequence at the same surface position to obtain the multi-angle pixel value sequence of the inner tube; Extreme value detection is performed on the multi-angle pixel value sequence to determine the maximum and minimum pixel values ​​of the inner tube; The initial difference value of the inner tube is obtained by performing a difference operation between the maximum pixel value and the minimum pixel value; The initial difference value is mapped to a preset gray value range to obtain the normalized difference value of the inner tube; By reorganizing the spatial positions of the normalized difference values ​​according to the actual spatial arrangement order of the inner tube, a difference contrast image of the inner tube is generated.

4. The real-time screening method for inner tube surface defects based on machine vision as described in claim 1, characterized in that, The step of extracting connected components from the candidate surface anomaly regions in the differential contrast image to obtain the candidate defect connected components of the inner tube includes: Based on the circumferential diameter and axial length of the inner tube, the differential contrast image is divided into equidistant grids to obtain local detection sub-regions of the inner tube; The grayscale values ​​of pixels within the local detection sub-region are statistically analyzed and the outlier degree is measured to obtain the local mean and local standard deviation of the local detection sub-region. The local detection sub-region is binarized and segmented using the local mean and the local standard deviation to obtain a local binarized sub-map of the local detection sub-region. The local binarized sub-images are stitched together and fused, and morphological opening is performed on the fused image to obtain the global binarized image of the inner tube. Connected component labeling is performed on the global binarized image to obtain the edge pixel set of the inner tube, and the pixel area of ​​the region enclosed by the edge pixel set is calculated to obtain the actual pixel area of ​​the inner tube. Connected regions whose actual pixel area is less than a preset area threshold are discarded as pseudo-noise, and the remaining connected regions are retained as candidate defect connected regions of the inner tube.

5. The real-time screening method for inner tube surface defects based on machine vision as described in claim 4, characterized in that, The step of performing morphological quantization on the connected components of the candidate defects to obtain the morphological feature parameters of the inner tube includes: Extract the minimum bounding ellipse of the candidate defect connected domain, and obtain the length of the major axis of the minimum bounding ellipse; The directional deflection parameters of the candidate defect connected domain are obtained by directional calculation of the angle between the length of the major axis and the axial direction of the inner tube. The edge pixel set of the candidate defective connected region is fitted with a second moment to obtain the dispersion parameter and elongation parameter of the candidate defective connected region. Extend a predetermined number of pixels outward along the edge normal direction of the candidate defect connected region, and collect the grayscale values ​​of the pixels in the extended region; Based on the edge normal direction, the gradient magnitude of the gray value is analyzed by gradient decay to obtain the edge sharpness parameter of the candidate defect connected domain; The morphological feature parameters of the inner tube are obtained by combining the actual pixel area, the orientation deflection parameter, the dispersion parameter, the elongation parameter, and the edge sharpness parameter.

6. The real-time screening method for inner tube surface defects based on machine vision as described in claim 1, characterized in that, The step of comparing the morphological feature parameters with preset defect judgment conditions item by item to determine the actual defect area of ​​the inner tube, and recording the position coordinates and defect type label of the actual defect area in the inner tube, includes: The morphological feature parameters are compared with the preset defect judgment conditions one by one to obtain the independent comparison results of the inner tube, and the independent comparison results are weighted and evaluated to obtain the comprehensive score value of the inner tube. The comprehensive score is compared with a preset comprehensive judgment threshold. When the comprehensive score is not less than the comprehensive judgment threshold, the candidate defect connected region is determined to be the real defect region of the inner tube; otherwise, it is determined to be a false defect region and is removed. The actual defect area is matched with a preset defect type determination table to determine the defect type label of the actual defect area; Extract the centroid coordinates and the vertex coordinates of the smallest bounding rectangle of the actual defect area, and map the centroid coordinates and the vertex coordinates of the smallest bounding rectangle to the spatial coordinate system of the inner tube to obtain the position coordinates of the actual defect area.

7. The real-time screening method for inner tube surface defects based on machine vision as described in claim 1, characterized in that, The step of performing topological reconstruction on the location coordinates and the defect type markers to obtain the defect distribution map of the inner tube includes: The position coordinates are mapped to the two-dimensional unfolded plane coordinate system of the inner tube to obtain the plane coordinates of the actual defect area; Based on the planar coordinates, the spatial distance between the actual defective areas is calculated to obtain the spatial distance value of the actual defective areas; By using the spatial distance value, adjacency determination and connectivity analysis are performed on the actual defect area to obtain the spatial adjacency relationship diagram of the actual defect area; Extract the defect clusters from the spatial adjacency graph, and count the types and number of defects in the defect clusters to obtain the cluster type composition information of the defect clusters; The cluster type composition information is associated with the spatial distribution location of the defect cluster to generate a defect distribution map of the inner tube.

8. The real-time screening method for inner tube surface defects based on machine vision as described in claim 7, characterized in that, The process of jointly determining the quality grade of the inner tube by considering the quantity and type of the actual defective areas and performing a joint decision on the defect distribution map includes: The total number of real defect areas in the defect cluster is counted to obtain the defect density value of the real defect area; The cluster type composition information is subjected to key type identification to obtain the type identification result of the real defect area; Based on the defect density value and the type identification result, the cluster hazard quantification of the defect distribution map is performed to obtain the cluster hazard quantification value of the defect distribution map; The quality level of the inner tube is obtained by globally weighting and aggregating the cluster hazard quantification values ​​and mapping them to a level.

9. The real-time screening method for inner tube surface defects based on machine vision as described in claim 8, characterized in that, The step of mapping the quality grade to screening control instructions to drive the sorting execution mechanism to perform real-time screening of the inner tubes includes: According to the preset quality grade classification rules, the quality grades are classified to determine the grade category corresponding to the inner tube; Based on the grade category, a preset filtering strategy mapping table is matched to determine the target filtering strategy corresponding to the grade category, and a corresponding filtering control instruction is generated based on the target filtering strategy. The screening control command is output to the sorting execution mechanism, which drives the sorting execution mechanism to perform a screening action on the inner tube that matches the quality grade.

10. A machine vision-based real-time screening system for defects on the surface of inner tubes, characterized in that, The system for implementing the machine vision-based real-time screening method for inner tube surface defects as described in claim 1 includes: The image acquisition module is used to perform a circumferential unfolding scan of the surface of the inner tube to obtain a multi-angle image sequence of the inner tube. The contrast difference module is used to perform contrast difference synthesis on the reflected light intensity at each angle at the same surface position in the multi-angle image sequence to generate the differential contrast image of the inner tube. The parameter quantization module is used to extract the connected components of the surface anomaly candidate regions in the differential contrast image to obtain the candidate defect connected components of the inner tube, and to perform morphological quantization on the candidate defect connected components to obtain the morphological feature parameters of the inner tube. The defect marking module is used to compare the morphological feature parameters with preset defect judgment conditions one by one to determine the actual defect area of ​​the inner tube, and record the position coordinates of the actual defect area in the inner tube and the defect type mark. The grade determination module is used to perform topological reconstruction on the location coordinates and the defect type markers to obtain the defect distribution map of the inner tube, and to make a joint decision on the defect distribution map based on the number and type of the actual defect areas to obtain the quality grade of the inner tube. The real-time screening module is used to map the quality grade into screening control instructions, driving the sorting execution mechanism to perform real-time screening of the inner tube.

Images