Real-time detection method for tire cord bubbles
By extracting the cord region and replacing gray values in tire X-ray images, combined with Gaussian convolution kernel adaptive threshold segmentation and density gradient analysis, the problem of false detection of cord texture was solved, achieving efficient and accurate bubble detection, adaptable to different working conditions and tire models.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HARBIN INST OF TECH
- Filing Date
- 2026-04-02
- Publication Date
- 2026-06-19
AI Technical Summary
Existing tire cord bubble detection methods are unable to effectively distinguish between cord texture and real bubbles, and are prone to misdetecting cord structure as a defect, resulting in a high false alarm rate and insufficient robustness.
By preprocessing the tire X-ray image, the target area of the cord is extracted, the cord pixels are marked and grayscale values are replaced to generate an intermediate image without cord interference. Adaptive threshold segmentation with Gaussian convolution kernel and multi-scale density gradient analysis are used to determine whether the suspected bubble area is a real bubble.
It significantly improves the accuracy and robustness of tire bubble detection, reduces the false detection rate, adapts to different working conditions and tire models, reduces the dependence on large-scale labeled data and complex deep learning models, and lowers hardware and maintenance costs.
Smart Images
Figure CN122243989A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to a method for real-time detection of air bubbles in tire cords, belonging to the field of computer vision and industrial non-destructive testing technology. Background Technology
[0002] In tire manufacturing, air bubbles within the cord are a critical hidden danger affecting tire safety and durability. As the tire's skeleton material, the presence of air bubbles within the cord leads to localized stress concentration and reduced adhesion between the cord and rubber, potentially causing premature tire blowout in severe cases. Therefore, efficient and accurate detection of air bubbles in the cord is a crucial aspect of tire quality control.
[0003] Currently, tire cord bubble detection mainly relies on manual visual inspection of X-ray images. Inspectors must rely on experience to identify tiny bubble defects from complex cord textures under continuous strong light. This method has significant drawbacks: first, it is inefficient, with a limited daily inspection capacity per person, making it unsuitable for high-speed production lines; second, it is prone to fatigue, as prolonged visual concentration leads to an increased missed detection rate over time; and third, inconsistent standards result in subjective differences in judgment among different inspectors regarding the same defect, making it difficult to guarantee quality stability.
[0004] With the development of machine vision technology, some companies are attempting to replace manual inspection with automated inspection methods. Existing automated inspection methods are mainly divided into two categories:
[0005] The first category is traditional image processing methods, which extract suspected bubble regions through edge detection, texture analysis, contrast enhancement, or morphological filtering. However, tire X-ray images show that the cord structure has regular, high-contrast, and periodic texture features, while bubbles often exhibit localized uniform gray levels and blurred edges. Traditional methods struggle to effectively distinguish between cords and bubbles, easily misjudging the regular textures or intersections of cords as bubble defects, resulting in a high false alarm rate. Furthermore, the uneven gray-level distribution in the cord region and the significant differences in cord specifications between different tire models make algorithms based on fixed thresholds or single features insufficiently robust and difficult to adapt to the diverse operating conditions of actual production lines.
[0006] The second category is deep learning methods, based on models such as convolutional neural networks, which train end-to-end detectors using a large number of labeled samples. While these methods can achieve high accuracy in laboratory environments, they face significant challenges in industrial deployments: First, they require tens of thousands of high-quality labeled defect images, but bubble defects are low-probability events in actual production, making it difficult to obtain positive samples and resulting in extremely high labeling costs. Second, deep learning models are computationally intensive, demanding high levels of hardware resources such as GPUs, making real-time detection difficult on low-cost embedded platforms. Furthermore, the model's generalization ability is limited by the distribution of training data; when tire models or process parameters change, the model is prone to performance degradation, requiring frequent retraining, which is time-consuming and costly.
[0007] In summary, existing tire cord bubble detection methods share a common core problem: they are unable to effectively distinguish between cord texture and actual bubbles, and are prone to misdetecting cord structures as defects. Summary of the Invention
[0008] To address the problem that existing methods struggle to distinguish between cords and air bubbles, and easily misdetect cord texture as a defect, this invention provides a real-time detection method for tire cord air bubbles.
[0009] The present invention provides a method for real-time detection of air bubbles in tire cords, comprising:
[0010] S1. Process the input tire X-ray grayscale image to extract the target area containing the tire cord;
[0011] S2. Mark the curtain line pixels in the target area;
[0012] S3. For the marked curtain line pixels, grayscale values are replaced using a grayscale linear interpolation method based on non-curtain line background pixels in the same column to generate an intermediate image without curtain line interference.
[0013] S4. Segment the intermediate image to obtain a binary mask, which includes multiple suspected bubble regions.
[0014] S5. Calculate the surface density of each suspected bubble region in the binary mask in multiple neighborhoods of different sizes, and determine whether the suspected bubble region is a real bubble based on the gradient decay feature of the surface density as the neighborhood expands.
[0015] Preferably, S5, the method for calculating the surface density of each suspected bubble region in the binary mask within multiple neighborhoods of different sizes, and determining whether a suspected bubble region is a real bubble based on the gradient decay characteristic of the surface density as the neighborhood expands, includes:
[0016] S51. Based on the current binary mask, define a set of scale-increasing neighborhoods for each suspected bubble region;
[0017] S52. Calculate the areal density of the suspected bubble region in each neighborhood. The areal density This is the ratio of the area of the suspected bubble region to the area of its corresponding neighboring region.
[0018] S53, Surface density based on neighboring scales and Calculate the density gradient decay rate ;
[0019] S54, If the density gradient decay rate If the preset threshold is not reached, the corresponding suspected bubble region is deleted from the binary mask, and the process proceeds to S51 until the set number of iterations is completed. The remaining suspected bubble regions in the binary mask are the real bubbles.
[0020] As a preferred option, the preset threshold is .
[0021] Preferably, in S2, the method for marking the curtain line pixels in the target area is as follows:
[0022] Within the target region, a local threshold is dynamically calculated using a vertical Gaussian convolution kernel along the image column direction. A local threshold is generated for each pixel.
[0023] Extract the 95th percentile of grayscale for each column within the target area. , which serves as the column threshold for pixels in the corresponding column;
[0024] Extract global grayscale values from the target region quantiles , which serves as the global threshold for pixels in the target region;
[0025] for , and Set weights and apply them accordingly. , and Perform linear weighted fusion to obtain the final judgment threshold for each pixel. ;
[0026] Within the target area, grayscale values are compared pixel-by-pixel with the corresponding final judgment thresholds, and grayscale values greater than or equal to the corresponding final judgment thresholds are selected. The pixels are labeled as curtain line pixels.
[0027] As a preferred option , and The weights are 0.74, 0.15, and 0.11, respectively.
[0028] Preferably, in S3, the grayscale value of the marked curtain line pixels is replaced using a grayscale linear interpolation method based on non-curtain line background pixels in the same column, generating an intermediate image without curtain line interference.
[0029] For the marked curtain line pixels, perform one round of lateral convolution kernel morphological erosion and two rounds of vertical convolution kernel morphological dilation.
[0030] Select the non-curtain background pixels that are adjacent to the top and bottom of the column containing the processed curtain pixel, and generate replacement grayscale values through one-dimensional linear interpolation. :
[0031]
[0032] in, The column number of the pixel. Indicates the row number of the pixel. The adjacent non-curtain background pixels on the upper side, for The row number, The adjacent non-curtain background pixels on the lower side, The row number it is in. express grayscale value, express The grayscale value.
[0033] Preferably, in S4, the method for segmenting the intermediate image and extracting the suspected bubble region includes:
[0034] A binary mask is obtained by using an adaptive thresholding method with Gaussian convolution kernels on the intermediate image. The binary mask includes multiple suspected bubble regions.
[0035] The beneficial effects of this invention are that it can effectively overcome the interference of local feature differences in images and structural regions (such as joints), and significantly improve the accuracy, robustness and economy of real-time tire bubble detection without relying on large-scale labeled data and complex deep learning models. Attached Figure Description
[0036] Figure 1 This is a flowchart of the method of the present invention;
[0037] Figure 2 X-ray image of a tire;
[0038] Figure 3 Here are the ROI region maps, where (a) is the original image, (b) is the curtain pixel extraction map, and (c) is the curtain interpolation overlay map;
[0039] Figure 4 The results are the labeling results for suspected bubble regions, where (a) represents zero density gradient iterations, (b) represents one density gradient iteration, and (c) represents two density gradient iterations. Detailed Implementation
[0040] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0041] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.
[0042] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.
[0043] The real-time tire cord bubble detection method of this embodiment includes:
[0044] Step 1: Process the input tire X-ray grayscale image to extract the target area containing the tire cords;
[0045] Raw tire X-ray images contain numerous invalid background areas (such as air, clamps, and edge blanks). Directly processing the entire image increases computational burden, and background noise may interfere with subsequent detection. Through image preprocessing and region localization, target regions containing only the tire cord are extracted, and invalid background at image edges is removed, narrowing the scope of subsequent processing. Specifically, this includes:
[0046] Global threshold coarse segmentation is performed on the input grayscale image to identify the main tire region;
[0047] Use tire geometry (such as circular or approximately rectangular) to locate the area where the cords are located;
[0048] The smallest bounding rectangle containing all the cords is cut out as the Region of Interest (ROI), and the resulting cords... area
[0049] Step 2: Mark the curtain line pixels in the target area:
[0050] In X-ray images, the cord filament typically appears as a bright line, but its grayscale is affected by the cord material, X-ray intensity, and tire structure, resulting in drastic local variations. A single global threshold or a simple local threshold is insufficient to accurately separate the cord filament from the background. This step employs an adaptive thresholding method that weights and fuses multi-source statistical information, combining local grayscale, column grayscale quantiles, and global grayscale quantiles to generate a dynamic threshold for each pixel. Specifically, this includes:
[0051] Along the image column direction, use a size of (Approximately in actual engineering) By sliding the vertical Gaussian convolution kernel, the average gray value within the local window corresponding to each pixel is calculated to obtain the local threshold. ;
[0052] For each column of the target area, calculate the 95th percentile of the gray values in that column. , which serves as the column threshold for pixels in the corresponding column;
[0053] Calculate the 99.5th percentile of grayscale values in the entire target area. , which serves as the global threshold for pixels in the target region;
[0054] Through weighted formula Calculate the final decision threshold Within the target area, the grayscale value is compared pixel by pixel with the corresponding final judgment threshold, and the grayscale value that is greater than or equal to the corresponding final judgment threshold is selected. The pixels are labeled as curtain line pixels, and the output is a set of curtain line pixels. .
[0055] This step achieves accurate and robust marking of curtain line pixels in a complex environment with uneven grayscale distribution, varying curtain line thickness, and fluctuating contrast, providing reliable input for subsequent curtain line coverage.
[0056] Step 3: For the marked curtain line pixels, use the grayscale linear interpolation method based on the non-curtain line background pixels in the same column to replace the grayscale values and generate an intermediate image without curtain line interference.
[0057] The curtain-like structure appears as a regular texture in an image. If bubble segmentation is directly applied to an image containing the curtain-like structure, the curtain-like structure may be misidentified as a defect. This step utilizes adjacent non-curtain-like background pixels in the same column to generate new grayscale values through one-dimensional linear interpolation, replacing the original curtain-like pixels, thereby weakening or eliminating the curtain-like structure in the image.
[0058] For the set of curtain pixels Each pixel marked as a curtain Search upwards and downwards along its column for the nearest non-curtain background pixel. and ;
[0059] The alternative grayscale value is calculated using a one-dimensional linear interpolation formula. :
[0060]
[0061] in, The column number of the pixel. Indicates the row number of the pixel. for The row number, The row number it is in. express grayscale value, express The grayscale value.
[0062] Replace the original grayscale with ;
[0063] After traversing all curtain line pixels, an intermediate image free of curtain line interference is obtained. This step completely eliminates the interference of curtain line texture on subsequent bubble detection, making bubbles stand out more in the image and significantly reducing the false detection rate.
[0064] Step 4: Segment the intermediate image to obtain a binary mask, which includes multiple suspected bubble regions;
[0065] After step 3, the interference from the curtain texture in the intermediate image has been largely eliminated. The bubble region appears in the image as a structure with relatively high local gray levels and an approximate clumping shape. To extract the suspected bubble region, this step uses a Gaussian convolution kernel adaptive thresholding method to obtain a binary mask, which includes multiple suspected bubble regions. The core principle is as follows:
[0066] A two-dimensional Gaussian function is used as the weight template to perform weighted smoothing on the image. Compared with the mean convolution kernel, the Gaussian convolution kernel can assign different weights to pixels based on their distance from the center (the closer the distance, the greater the weight), thus focusing more on the neighborhood features of the central pixel when calculating the local threshold and reducing the interference from distant pixels;
[0067] For each pixel in the intermediate image, a Gaussian weighted average is calculated within a preset window centered on that pixel, and this average is used as the local threshold for that pixel. Pixels at different positions within the window contribute different weights, making the threshold more reflective of the true grayscale distribution of the local area.
[0068] The original pixel grayscale is compared with the calculated local adaptive threshold. Pixels with grayscale values higher than the threshold are marked as suspected bubble regions, and pixels with grayscale values lower than the threshold are marked as background.
[0069] This step, using an adaptive threshold for the Gaussian convolution kernel, effectively adapts to uneven lighting and background grayscale fluctuations in the image. In images without curtain interference, it separates the bubble region from the background, forming a complete binary mask. Due to the weight distribution characteristics of the Gaussian convolution kernel, the boundaries of the segmented bubble region are relatively smooth, less prone to excessive fragmentation or adhesion, which is beneficial for subsequent connected component analysis.
[0070] Step 5: Calculate the surface density of the suspected bubble region in multiple neighborhoods of different sizes, and determine whether the suspected bubble region is a real bubble based on the gradient decay characteristic of the surface density as the neighborhood expands.
[0071] Real bubbles exhibit a typical spatial distribution characteristic of being "dense at the center and sparse outwards": within a small neighborhood near the bubble center, the bubble region accounts for a high proportion (high areal density); as the neighborhood expands, the background proportion increases, and the areal density decreases rapidly. False defects (such as residual noise, joints, and texture residues) typically do not exhibit this gradient decay pattern; their areal density decreases slowly with increasing neighborhood size, and may even fluctuate due to the presence of other structures within the neighborhood. This step quantifies the aforementioned spatial distribution characteristics by defining a set of scale-increasing neighborhoods for each suspected bubble region, calculating the areal density within each neighborhood, and the density gradient decay rate between adjacent scales. Simultaneously, an iterative screening strategy is employed to eliminate regions that do not meet the decay criteria round by round, ultimately retaining the real bubbles. Specifically, this includes:
[0072] Step 51: The size and shape of each bubble's suspected area may differ. This applies to the individual suspected areas extracted in Step 4. Define a set of scale-increasing neighborhoods This is used to capture the density distribution characteristics of a region at different spatial scales. For example:
[0073] Step 52: Calculate the areal density of the suspected bubble region in each neighborhood. :
[0074]
[0075] in This is a function for calculating the area of a region.
[0076] areal density This reflects the density of the bubble region within its local environment, its areal density. The spatial distribution of bubbles is transformed into a quantifiable density index, providing fundamental data for subsequent gradient calculations. Higher areal density indicates that the bubbles are more prominent in their neighborhood.
[0077] Step 53: Based on the surface density corresponding to the neighboring scale. and Calculate the density gradient decay rate:
[0078] ;
[0079] The density gradient decay rate quantifies the rate at which the areal density decays with increasing neighborhood. Real bubbles exhibit a higher decay rate (rapidly transitioning from a dense center to a sparse background), while pseudo-defects show a lower decay rate (relatively uniform or scattered distribution). A higher decay rate indicates a more likely real bubble. This step uses the maximum decay rate between adjacent scales as the criterion.
[0080] Step 54, if the density gradient decay rate If the preset threshold is not reached, the corresponding suspected bubble region is deleted from the binary mask, and the process proceeds to step 51 until the set number of iterations is completed. The remaining suspected bubble regions in the binary mask are then considered real bubbles.
[0081] Set a preset threshold, such as 32%. For each suspected bubble region, if its density gradient decay rate does not reach the threshold (i.e., the decay is insufficient), it is considered that the region does not have the typical distribution characteristics of a real bubble and is removed from the binary mask.
[0082] After a deletion, the number and distribution of the remaining regions in the binary mask may change, and the neighborhood statistics (such as whether the neighborhood contains other retained regions) will also change accordingly. Therefore, we return to step 51 to redefine the neighborhood, calculate the surface density and gradient decay rate based on the updated binary mask, and perform a new round of judgment.
[0083] Repeat the above process until the set number of iterations is reached (e.g., 3 times). At this point, the remaining suspected bubble areas in the binary mask are the areas that are finally determined to be real bubbles.
[0084] This application employs multiple iterations to progressively eliminate false defects, stabilizing the judgment results and preventing accidental or missed deletions. When a neighborhood contains multiple suspected regions, deleting one will affect the areal density calculation of other regions; the iteration mechanism can dynamically adjust this, making the judgment more accurate. An upper limit on the number of iterations is set to ensure the algorithm converges within a limited computational load, meeting real-time requirements.
[0085] All regions identified as real bubbles are used as the final detection result. This step replaces subjective experience with quantifiable statistical features to achieve high-precision and interpretable bubble identification, while an iterative screening mechanism adapts to different bubble shapes, improving robustness.
[0086] This application adopts traditional image processing and density gradient iterative algorithms throughout the process, without the need for large-scale labeled datasets or complex deep learning models. It has a short development cycle, low hardware computing power requirements, and significantly reduced deployment and maintenance costs.
[0087] This application uses column-direction weighted adaptive thresholding to accurately mark the cord, and combines one-dimensional linear interpolation to completely eliminate cord structure interference. Multi-scale density gradient iterative judgment can effectively filter out noise and texture pseudo-defects, significantly reducing false detection and false detection rates.
[0088] The algorithm in this application is lightweight, without complex inference operations, and reduces the amount of computation by combining ROI region clipping. It has a fast single-frame processing speed and can be adapted to the high-speed real-time detection requirements of tire production lines.
[0089] This application relies on iterative screening based on quantized density gradient criteria, with objective and unified detection standards, unaffected by subjective human factors, high bubble recognition accuracy, and adaptability to different working conditions and tire models.
[0090] While the invention has been described herein with reference to specific embodiments, it should be understood that these embodiments are merely examples of the principles and applications of the invention. Therefore, it should be understood that many modifications can be made to the exemplary embodiments, and other arrangements can be designed without departing from the spirit and scope of the invention as defined by the appended claims. It should be understood that different dependent claims and features herein can be combined in ways different from those described in the original claims. It is also understood that features described in conjunction with individual embodiments can be used in other embodiments.
Claims
1. A method for real-time detection of air bubbles in tire cords, characterized in that, include: S1. Process the input tire X-ray grayscale image to extract the target area containing the tire cord; S2. Mark the curtain wire pixels in the target area; S3. For the marked curtain line pixels, grayscale values are replaced using a grayscale linear interpolation method based on non-curtain line background pixels in the same column to generate an intermediate image without curtain line interference. S4. Segment the intermediate image to obtain a binary mask, which includes multiple suspected bubble regions; S5. Calculate the surface density of each suspected bubble region in the binary mask in multiple neighborhoods of different sizes, and determine whether the suspected bubble region is a real bubble based on the gradient decay feature of the surface density as the neighborhood expands.
2. The method for real-time detection of air bubbles in tire cords according to claim 1, characterized in that, S5. The method for calculating the surface density of each suspected bubble region in a binary mask within multiple neighborhoods of different sizes, and determining whether the suspected bubble region is a real bubble based on the gradient decay characteristic of the surface density as the neighborhood expands, includes: S51. Based on the current binary mask, define a set of scale-increasing neighborhoods for each suspected bubble region; S52. Calculate the areal density of the suspected bubble region in each neighborhood. The areal density This is the ratio of the area of the suspected bubble region to the area of its corresponding neighboring region. S53, Surface density based on neighboring scales and Calculate the density gradient decay rate ; S54, if the density gradient decay rate If the preset threshold is not reached, the corresponding suspected bubble region is deleted from the binary mask, and the process proceeds to S51 until the set number of iterations is completed. The remaining suspected bubble regions in the binary mask are the real bubbles.
3. The method for real-time detection of air bubbles in tire cords according to claim 2, characterized in that, The preset threshold is .
4. The method for real-time detection of air bubbles in tire cords according to claim 1, characterized in that, In S2, the method for marking curtain wire pixels in the target area is as follows: In the target region, a local threshold is dynamically calculated using a vertical Gaussian convolution kernel along the image column direction. A local threshold is generated for each pixel. Extract the 95th percentile of grayscale for each column in the target area. , which serves as the column threshold for pixels in the corresponding column; Extract global grayscale values from the target region. quantiles , which serves as the global threshold for pixels in the target region; for , and Set weights and apply them accordingly. , and Perform linear weighted fusion to obtain the final judgment threshold for each pixel. ; In the target area, the grayscale value is compared pixel by pixel with the corresponding final judgment threshold, and the grayscale value that is greater than or equal to the corresponding final judgment threshold is selected. The pixels are labeled as curtain line pixels.
5. The method for real-time detection of air bubbles in tire cords according to claim 4, characterized in that, , and The weights are 0.74, 0.15, and 0.11, respectively.
6. The method for real-time detection of air bubbles in tire cords according to claim 1, characterized in that, In S3, for the marked curtain line pixels, a grayscale value replacement method based on grayscale linear interpolation of non-curtain line background pixels in the same column is used to generate an intermediate image without curtain line interference. For the marked curtain line pixels, perform one round of lateral convolution kernel morphological erosion and two rounds of vertical convolution kernel morphological dilation. Select the non-curtain background pixels that are adjacent to the top and bottom of the column containing the processed curtain pixel, and generate replacement grayscale values through one-dimensional linear interpolation. : in, The column number of the pixel. Indicates the row number of the pixel. The adjacent non-curtain background pixels on the upper side, for The row number, The adjacent non-curtain background pixels on the lower side, The row number it is in. express grayscale value, express The grayscale value.
7. The method for real-time detection of air bubbles in tire cords according to claim 1, characterized in that, In S4, the method for segmenting the intermediate image and extracting the suspected bubble region includes: A binary mask is obtained by using a Gaussian convolution kernel adaptive thresholding method on the intermediate image. The binary mask includes multiple suspected bubble regions.
8. A computer-readable storage device storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the real-time tire cord bubble detection method as described in any one of claims 1 to 7.
9. A real-time tire cord bubble detection device, comprising a storage device, a processor, and a computer program stored in the storage device and executable on the processor, characterized in that, The processor executes the computer program to implement the steps of the real-time tire cord bubble detection method as described in any one of claims 1 to 7.
10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps of the real-time tire cord bubble detection method as described in any one of claims 1 to 7.