Tire performance detection method and system based on tire defect data and binary descriptor

Abstract

The application discloses a tire performance detection method and system based on tire defect data and binary descriptors, and relates to the technical field of tire defect detection. A defect block is marked and a difference feature relative to a normal surface is extracted; a virtual tire model is constructed, difference feature components are assembled and configured in the model, parameter groups are generated through data dimension reduction and correlation is established; a consistent parameter group combination is randomly selected, similar parameters are determined through comparison, parameter volume requirements are optimized and new parameter groups are adjusted; a feature component after dimension reduction is constructed based on a position and a parameter group on the virtual model; a binary descriptor extraction model is trained by adjusting a screenshot in multiple postures for the model; the binary descriptor is input into a double-channel Transformer model for training, so that a tire performance detection model is obtained and is used for real-time detection of a tire surface image; through virtual model simulation and feature dimension reduction, the application reduces training data volume and difficulty, avoids dirt interference, improves detection precision and robustness, and is suitable for tire quality control.

Description

Technical Field

[0001] This invention relates to the field of tire defect detection technology, and in particular to a tire performance detection method and system based on tire defect data and binary descriptors. Background Technology

[0002] With the rapid development of the automotive industry, tires, as a key vehicle component, are directly affected by surface defects (such as bubbles, impurities, cracks, or foreign objects), which impact driving safety and service life. However, traditional tire defect detection mainly relies on manual visual inspection or simple image processing methods. These methods have many limitations: manual inspection is highly subjective, inefficient, and easily affected by fatigue and experience differences, leading to a high rate of missed detections. Simple image processing methods cannot effectively distinguish dirt and grime on tires, resulting in poor detection accuracy. Summary of the Invention

[0003] The purpose of this invention is to provide a tire performance testing method and system with better testing results.

[0004] This invention discloses a tire performance detection method based on tire defect data and binary descriptors, including:

[0005] Step S100: Determine the type of defect on the tire surface, mark the tire defect blocks on the tire surface image based on the defect type, and determine the difference features of the corresponding tire defect blocks relative to the normal tire surface.

[0006] Step S200: Construct a virtual tire model, componentize the differential features to obtain differential feature components, configure the differential feature components in the virtual tire model, perform data volume dimensionality reduction analysis on the differential feature components to obtain differential feature component parameter groups, and establish the correlation between differential feature component parameter groups and differential feature components.

[0007] Step S300: Randomly select several matching combinations of differential feature component parameter groups, and compare the differential feature components corresponding to each combination to determine similar parameters. If the average value of the similar parameters is less than or equal to a preset value, the parameter volume of the differential feature component parameter group is reduced until the average value of the similar parameters is greater than the preset value.

[0008] Step S400: Based on the position of the differential feature components on the virtual tire model and their corresponding differential feature component parameter groups, construct the dimension-reduced feature components on the virtual tire model.

[0009] Step S500: Perform multi-pose adjustment on the virtual tire model and take screenshots to obtain several virtual tire model images. Then, perform binary descriptor extraction model training based on the virtual tire model images.

[0010] Step S600: Input the binary descriptor output by the binary descriptor extraction model into the dual-channel Transformer model for training to obtain the tire performance detection model. Use the tire performance detection model to detect and analyze the tire surface image to determine the defects on the tire surface.

[0011] In some embodiments disclosed in this invention, the method for detecting and analyzing tire surface images using a tire performance detection model includes:

[0012] Step S601: Take a picture of the tire surface to obtain a real-time tire surface image, remove the original tire tread structure in the real-time tire surface image, mark other edges on the tire surface, and map them onto the virtual tire model.

[0013] Step S602: Fill the marked edge contours on the virtual tire model with content to obtain the dimensionality-reduced feature components on the virtual tire model;

[0014] Step S603: Adjust the virtual tire model to the angle shown by the real-time tire surface image, take a screenshot to obtain a real-time virtual tire model image, and input the real-time virtual tire model image into the tire performance detection model for detection and analysis.

[0015] In some embodiments disclosed in this invention, a method for removing the original tire tread structure from a real-time tire performance image includes:

[0016] Step S6012: Determine the tire edge in the real-time tire performance image, determine the portion of the tire mapped in the image, and determine the continuous edges in the mapped tire portion. If the length and curvature change of the continuous edges meet the preset standards, the continuous edges are identified as tread structure edges, and the tread structure edges are removed.

[0017] In some embodiments disclosed in this invention, the method for componentizing differential features includes:

[0018] Step S101: Delineate the tire defect edge of the tire defect block and identify the tire defect edge as the edge frame of the differential feature component.

[0019] In some embodiments disclosed in this invention, the method for performing data volume dimensionality reduction analysis on differential feature groups includes:

[0020] Step S201: Determine the rolling direction line of the tire, and construct several vertical lines within the edge frame of the differential feature component. The two ends of the vertical lines contact the edge frame and are perpendicular to the rolling direction line. Combine the length and position of each vertical line in sequence to obtain the parameter group of the differential feature component.

[0021] In some embodiments disclosed in this invention, the method for filtering combinations of matching differential feature component parameter groups includes:

[0022] Step S301: Analyze the parameter group of the difference feature component, determine the position of the center point of each vertical line, and connect each center point in turn to form a center point polyline.

[0023] Step S302: Determine the turning points of the center point broken line, determine the opening direction of the broken line at each turning point, and combine the opening directions of the broken line in sequence to form a broken line opening sequence.

[0024] Step S303: The broken line opening sequence is identified as the first search tag. The search is performed in the differential feature component parameter group, and the matching differential feature component parameter group is selected. The length of each relative vertical line is used as the second search tag. The corresponding differential feature component parameter group is filtered again, and the finally selected differential feature component parameter group is combined.

[0025] In some embodiments disclosed in this invention, the method for comparing the difference feature components corresponding to each combination includes:

[0026] Step S304: According to the method of coinciding the center point polyline in the parameter group of the difference feature components, the difference feature components are overlapped, and the intersection area between the edge frames of the difference feature components is determined.

[0027] Step S305: Determine the component center of the differential feature component, and radiate several cross-region detection lines outward from the component center as the base point. Mark the intersection point of each cross-region detection line with the edge frame, determine the cross length between the intersection points, and arrange the several cross lengths in order to obtain the cross length sequence.

[0028] Step S306: Based on the area of ​​the cross region and the area ratio of the differential feature components, as well as the length variation fluctuation characteristics of the cross length in the cross length sequence, determine the similarity parameters between the differential feature components.

[0029] In some embodiments disclosed in this invention, the expression for calculating the similarity parameter is as follows:

[0030] ;

[0031] in, For similar parameters, This is the preset maximum area percentage. This represents the area of ​​the overlapping region and the area ratio of the components with different characteristics. For the length variation fluctuation analysis function, the difference in length between adjacent intersections is calculated. If the difference in intersection length is greater than or equal to a preset value, the corresponding intersection length is identified as the first intersection length; otherwise, it is identified as the second intersection length. The ratio of the number of first intersection lengths to the number of second intersection lengths is calculated, and based on the preset number ratio interval to which the ratio belongs, the following is determined: The output area ratio adjustment coefficient. For similarity parameter adjustment coefficients, Adjust constants for similarity parameters.

[0032] In some embodiments disclosed in this invention, a tire performance detection system based on tire defect data and binary descriptors includes:

[0033] The first module is used to determine the type of defect on the tire surface, and based on the defect type, to mark the tire defect blocks on the tire surface image, and to determine the difference features of the corresponding tire defect blocks relative to the normal tire surface.

[0034] The second module is used to construct a virtual tire model, componentize the differential features to obtain differential feature components, configure the differential feature components in the virtual tire model, perform data volume dimensionality reduction analysis on the differential feature components to obtain differential feature component parameter groups, and establish the relationship between differential feature component parameter groups and differential feature components.

[0035] The third module is used to randomly select combinations of several matching differential feature component parameter groups, compare the differential feature components corresponding to each combination, determine similar parameters, determine the parameter volume requirements for the differential feature component parameter groups based on the similar parameters, and adjust the newly generated differential feature component parameter groups based on the parameter volume requirements.

[0036] The fourth module is used to construct dimensionality-reduced feature components on the virtual tire model based on the positions of the differential feature components on the virtual tire model and their corresponding parameter groups.

[0037] The fifth module is used to perform multi-pose adjustments on the virtual tire model, take screenshots of each pose to obtain several virtual tire model images, and train the binary descriptor extraction model based on the virtual tire model images.

[0038] The sixth module is used to input the binary descriptor output by the binary descriptor extraction model into the dual-channel Transformer model for training, thereby obtaining the tire performance detection model. The tire performance detection model is then used to detect and analyze tire surface images to determine tire surface defects.

[0039] This invention discloses a tire performance detection method and system based on tire defect data and binary descriptors, belonging to the field of tire defect detection technology. The method involves: marking defective areas and extracting differential features relative to normal surfaces; constructing a virtual tire model, componentizing and configuring the differential features within the model, generating parameter groups through data dimensionality reduction and establishing associations; randomly selecting matching parameter group combinations, comparing and determining similar parameters, optimizing parameter size requirements, and adjusting new parameter groups; constructing dimensionality-reduced feature components on the virtual model based on position and parameter groups; performing multi-pose adjustments and screenshots on the model to train a binary descriptor extraction model; inputting the binary descriptors into a dual-channel Transformer model for training, obtaining a tire performance detection model for real-time detection of tire surface images. This invention reduces the amount and difficulty of training data through virtual model simulation and feature dimensionality reduction, avoids dirt interference, improves detection accuracy and robustness, and is suitable for tire quality control.

[0040] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the steps of a tire performance detection method based on tire defect data and binary descriptors disclosed in an embodiment of the present invention. Detailed Implementation

[0042] The technical solution of the present invention will be further described below with reference to the accompanying drawings and embodiments.

[0043] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and specific embodiments. It should be understood that the preferred embodiments described herein are only for illustration and explanation of the present invention and should not be construed as limiting the scope of protection of the present invention. Those skilled in the art can make some non-essential improvements and adjustments based on the following content of the present invention. In the present invention, unless otherwise expressly specified and limited, the technical terms used in the present invention should have the ordinary meaning understood by those skilled in the art.

[0044] Example:

[0045] This invention discloses a tire performance detection method based on tire defect data and binary descriptors, see reference. Figure 1 ,include:

[0046] Step S100: Determine the type of defect on the tire surface, mark the tire defect blocks on the tire surface image based on the defect type, and determine the difference features of the corresponding tire defect blocks relative to the normal tire surface.

[0047] First, the type of defect on the tire surface, such as cracks, bulges, or foreign object embedding, is determined through manual observation or expert experience. This step ensures that subsequent processing is optimized for specific defect characteristics. Then, the defect type is marked on the tire surface image. This marking mainly involves delineating the defect edges. Specifically, the edge contour of the defect block is precisely outlined using manual tools (such as annotation software) to form a closed boundary to isolate the defect area and avoid interference from internal filling. Finally, the difference characteristics of the defect block relative to the normal tire surface are determined. These difference characteristics mainly refer to the shape of the edge contour.

[0048] Step S200: Construct a virtual tire model, componentize the differential features to obtain differential feature components, configure the differential feature components in the virtual tire model, perform data volume dimensionality reduction analysis on the differential feature components to obtain differential feature component parameter groups, and establish the correlation between differential feature component parameter groups and differential feature components.

[0049] A virtual tire model is constructed (e.g., using 3D modeling software or parametric surface models to simulate tire geometry) to provide a controlled environment for integrating defect features. The differential features are componentized, that is, continuous edge contour shapes are decomposed into independent reusable components (e.g., frames or sub-contour segments based on defined edges) for modular processing. These components are then configured to their corresponding positions in the virtual model to simulate and reproduce the defects. Data volume reduction analysis is performed on the components, reducing their data volume, generating parameter groups for the differential feature components (e.g., vector or sequence representations), and establishing the association between the parameter groups and the original components. This facilitates efficient storage and retrieval, reduces computational overhead, and retains key information.

[0050] Step S300: Randomly select several matching combinations of differential feature component parameter groups, and compare the differential feature components corresponding to each combination to determine similar parameters. If the average value of the similar parameters is less than or equal to a preset value, the parameter volume of the differential feature component parameter group is reduced until the average value of the similar parameters is greater than the preset value.

[0051] Several matching differential feature component parameter groups are randomly selected, and similarity matching is used to simulate diverse scenarios to avoid bias. For each combination, the corresponding differential feature components are compared, and similarity parameters (such as contour overlap area or shape matching degree) are calculated to quantify the similarity between components. Based on these similarity parameters, parameter size requirements are determined, i.e., the dimensionality or number threshold of the parameter group is optimized to ensure minimal information redundancy. If the similarity parameter is too low, it indicates that the data size of the differential feature component parameter group is too small to guarantee sufficient similarity between the compared differential feature components, and therefore the difference between the differential feature component parameter groups cannot guarantee sufficient difference between the differential feature components. Therefore, the size needs to be increased to improve the distinguishing ability. Finally, based on this requirement, the newly generated parameter groups are constrained and adjusted, such as by truncating or supplementing parameters. This step improves the robustness and efficiency of the parameter groups through iterative optimization and is suitable for dynamic defect data processing.

[0052] Step S400: Based on the position of the differential feature components on the virtual tire model and their corresponding differential feature component parameter groups, construct the dimension-reduced feature components on the virtual tire model.

[0053] By utilizing the position coordinates of differential feature components on the virtual tire model (such as positioning relative to the tread or sidewall) and the corresponding parameter sets (such as length sequences or polyline descriptions), the dimensionality-reduced feature components are reconstructed on the model. Typically, simplified versions of edge contour shape representations are generated through parametric rendering or interpolation methods. This step transforms high-dimensional components into low-dimensional equivalents, preserving positional relevance and core parameters while reducing noise interference, thereby creating standardized features suitable for machine learning input and ensuring data consistency and computational feasibility during subsequent model training.

[0054] In step S500, the virtual tire model is adjusted in multiple poses and screenshots are taken to obtain several virtual tire model images. Then, binary descriptor extraction model training is performed based on the virtual tire model images.

[0055] The virtual tire model undergoes multi-pose adjustments (such as rotation, scaling, or viewpoint transformation) to simulate various angles and deformations under real-world shooting, thereby enhancing the diversity of the dataset. Several virtual tire model images are generated by taking screenshots, forming an expanded training sample set. A binary descriptor extraction model is trained based on these images, typically using variants of algorithms such as BRIEF or ORB, to convert local image features into compact binary strings. These descriptors have rotation invariance and efficient matching properties. The model parameters are optimized through supervised learning, enabling it to extract robust defect edge contour representations from complex tire textures.

[0056] Step S600: Input the binary descriptor output by the binary descriptor extraction model into the dual-channel Transformer model for training to obtain the tire performance detection model. Use the tire performance detection model to detect and analyze the tire surface image to determine the defects on the tire surface.

[0057] The binary descriptor extracted from the model is used as input and fed into a dual-channel Transformer model for training. The dual channels may handle the spatial channel (capturing local contour patterns) and the sequence channel (handling global shape dependencies), respectively. The Transformer models long-distance relationships through a self-attention mechanism, improving the capture of multi-scale edge contours of defects. After training, a tire performance detection model is obtained, which utilizes cross-entropy loss and optimizers such as Adam fine-tuning parameters. Finally, the actual tire surface image is detected and analyzed. The probability or location of defects is output through forward propagation to determine the surface defects. This step achieves end-to-end automated detection. Combining binary efficiency and the powerful representation capabilities of the Transformer, the accuracy and generalization are improved.

[0058] In some embodiments disclosed in this invention, the method for detecting and analyzing tire surface images using a tire performance detection model includes:

[0059] Step S601: Take a picture of the tire surface to obtain a real-time tire surface image, remove the original tire tread structure in the real-time tire surface image, mark other edges on the tire surface, and map them onto the virtual tire model.

[0060] Step S602: Fill the marked edge contours on the virtual tire model with content to obtain the dimensionality-reduced feature components on the virtual tire model.

[0061] Step S603: Adjust the virtual tire model to the angle shown by the real-time tire surface image, take a screenshot to obtain a real-time virtual tire model image, and input the real-time virtual tire model image into the tire performance detection model for detection and analysis.

[0062] In some embodiments disclosed in this invention, a method for removing the original tire tread structure from a real-time tire performance image includes:

[0063] Step S6012: Determine the tire edge in the real-time tire performance image, determine the portion of the tire mapped in the image, and determine the continuous edges in the mapped tire portion. If the length and curvature change of the continuous edges meet the preset standards, the continuous edges are identified as tread structure edges, and the tread structure edges are removed.

[0064] First, the tire edges in the real-time tire surface image are determined. Edge detection algorithms such as Canny or Sobel operators are typically used to extract significant boundary lines in the image to distinguish the tire from the background region and avoid external noise interference. Then, the mapped tire portion in the image is identified, i.e., continuous tire regions are marked based on edge connectivity or region growing methods to ensure the analysis focuses on the tire surface. Subsequently, continuous edges within the mapped tire portion are determined, and sequences of continuous line segments are extracted using chain code tracing or Hough transform. If the length of the continuous edges exceeds a preset threshold (e.g., based on empirical pixels),... If the length standard and the curvature change meets the preset standard (e.g., the rate of curvature change is less than the threshold, indicating relatively straight or regularly curved), then these continuous edges are identified as tread structure edges. This is because the original tire tread usually exhibits periodic, long straight lines or regular arc geometric features, which contrast sharply with random defect edges (such as short, irregular zigzags). Finally, the identified tread structure edges are removed, for example, by masking, edge filling or morphological erosion to remove these lines, thereby eliminating the interference of the tread pattern on defect detection, highlighting the abnormal edges of real defects, and improving the accuracy and robustness of subsequent image analysis.

[0065] In some embodiments disclosed in this invention, the method for componentizing differential features includes:

[0066] Step S101: Delineate the tire defect edge of the tire defect block and identify the tire defect edge as the edge frame of the differential feature component.

[0067] In some embodiments disclosed in this invention, the method for performing data volume dimensionality reduction analysis on differential feature groups includes:

[0068] Step S201: Determine the rolling direction line of the tire, and construct several vertical lines within the edge frame of the differential feature component. The two ends of the vertical lines contact the edge frame and are perpendicular to the rolling direction line. Combine the length and position of each vertical line in sequence to obtain the parameter group of the differential feature component.

[0069] In some embodiments disclosed in this invention, the method for filtering combinations of matching differential feature component parameter groups includes:

[0070] Step S301: Analyze the parameter group of the difference feature component, determine the position of the center point of each vertical line, and connect each center point in turn to form a center point broken line.

[0071] The differential feature component parameter set is analyzed. This parameter set is usually a sequence of data consisting of the lengths and positions of a series of vertical lines. First, the center point position of each vertical line is calculated, that is, the midpoint coordinates of its two endpoints are taken to capture the geometric center trend of the contour. Then, these center points are connected in sequence to form a continuous center point polyline. This polyline is essentially a simplified geometric representation of the parameter set. It is generated by linear interpolation or direct connection methods, which can abstract the overall shape features of the defect edge contour, such as the curvature trend or convex-concave changes. This provides an efficient, low-dimensional structured description for subsequent similarity retrieval, avoids the computational complexity brought about by directly comparing high-dimensional raw data, and highlights the macroscopic morphological differences of the contour.

[0072] Step S302: Determine the turning points of the center point broken line, determine the opening direction of the broken line at each turning point, and combine the opening directions of the broken lines in sequence to form a broken line opening sequence.

[0073] The inflection points of the polyline at the center point are determined by using angle thresholding or quadratic difference calculations to identify points in the polyline where the direction changes significantly. These inflection points represent key turning points in the contour shape. Then, the opening direction of the polyline at each inflection point is determined, i.e., the orientation of the opening (e.g., upward, downward, left, or right) is calculated based on the vectors of adjacent line segments. This can be achieved through vector cross product or angular quadrant determination. Finally, these opening directions are combined in sequence to form a polyline opening sequence. This sequence, as a symbolic string or vector, encodes the topological structure information of the polyline, exhibiting rotation invariance and scale robustness, facilitating rapid matching and indexing. This transforms complex contours into searchable discrete labels, reducing noise interference and improving retrieval efficiency.

[0074] Step S303: The broken line opening sequence is identified as the first search tag. The search is performed in the differential feature component parameter group, and the matching differential feature component parameter group is selected. The length of each relative vertical line is used as the second search tag. The corresponding differential feature component parameter group is filtered again, and the finally selected differential feature component parameter group is combined.

[0075] The broken line opening sequence is identified as the first retrieval label, which is used as a retrieval condition in the database or set of differential feature component parameter groups. Parameter groups with the same or similar sequences are filtered out through string matching or hash lookup. This step initially filters out potentially matching groups, ensuring basic shape consistency based on topological similarity. Then, the length of each opposite vertical line is used as the second retrieval label, which further compares the length differences of the vertical lines at corresponding positions (e.g., using thresholds or normalized distance) to refine the screening, ensuring not only topological similarity but also scale matching. Finally, the selected parameter groups are combined to form pairs or clusters. This step achieves efficient clustering of similar components through multi-level label retrieval, optimizes the computational resources for subsequent comparisons, and ensures the diversity and representativeness of the combinations, providing the basic data for parameter volume optimization.

[0076] In some embodiments disclosed in this invention, the method for comparing the difference feature components corresponding to each combination includes:

[0077] Step S304: According to the method of coinciding the center point polyline in the parameter group of the difference feature components, the difference feature components are overlapped, and the intersection area between the edge frames of the difference feature components is determined.

[0078] Step S305: Determine the component center of the differential feature component, and radiate several cross-region detection lines outward from the component center as the base point. Mark the intersection point of each cross-region detection line with the edge frame, determine the cross length between the intersection points, and arrange the several cross lengths in order to obtain the cross length sequence.

[0079] Step S306: Based on the area of ​​the cross region and the area ratio of the differential feature components, as well as the length variation fluctuation characteristics of the cross length in the cross length sequence, determine the similarity parameters between the differential feature components.

[0080] In some embodiments disclosed in this invention, the expression for calculating the similarity parameter is as follows:

[0081] .

[0082] in, For similar parameters, This is the preset maximum area percentage. This represents the area of ​​the overlapping region and the area ratio of the components with different characteristics. For the length variation fluctuation analysis function, the difference in length between adjacent intersections is calculated. If the difference in intersection length is greater than or equal to a preset value, the corresponding intersection length is identified as the first intersection length; otherwise, it is identified as the second intersection length. The ratio of the number of first intersection lengths to the number of second intersection lengths is calculated, and based on the preset number ratio interval to which the ratio belongs, the following is determined: The output area ratio adjustment coefficient. For similarity parameter adjustment coefficients, Adjust constants for similarity parameters.

[0083] In some embodiments disclosed in this invention, a tire performance detection system based on tire defect data and binary descriptors includes:

[0084] The first module is used to determine the type of defect on the tire surface, and based on the defect type, to mark the tire defect blocks on the tire surface image, and to determine the difference features of the corresponding tire defect blocks relative to the normal tire surface.

[0085] The second module is used to construct a virtual tire model, componentize the differential features to obtain differential feature components, configure the differential feature components in the virtual tire model, perform data volume dimensionality reduction analysis on the differential feature components to obtain differential feature component parameter groups, and establish the relationship between differential feature component parameter groups and differential feature components.

[0086] The third module is used to randomly select combinations of several matching differential feature component parameter groups, compare the differential feature components corresponding to each combination, determine similar parameters, determine the parameter volume requirements for the differential feature component parameter groups based on the similar parameters, and adjust the newly generated differential feature component parameter groups based on the parameter volume requirements.

[0087] The fourth module is used to construct dimensionality-reduced feature components on the virtual tire model based on the positions of the differential feature components on the virtual tire model and their corresponding parameter groups.

[0088] The fifth module is used to perform multi-pose adjustments on the virtual tire model, take screenshots of each pose to obtain several virtual tire model images, and train the binary descriptor extraction model based on the virtual tire model images.

[0089] The sixth module is used to input the binary descriptor output by the binary descriptor extraction model into the dual-channel Transformer model for training, thereby obtaining the tire performance detection model. The tire performance detection model is then used to detect and analyze tire surface images to determine tire surface defects.

[0090] This invention discloses a tire performance detection method and system based on tire defect data and binary descriptors, belonging to the field of tire defect detection technology. The method involves: marking defective areas and extracting differential features relative to normal surfaces; constructing a virtual tire model, componentizing and configuring the differential features within the model, generating parameter groups through data dimensionality reduction and establishing associations; randomly selecting matching parameter group combinations, comparing and determining similar parameters, optimizing parameter size requirements, and adjusting new parameter groups; constructing dimensionality-reduced feature components on the virtual model based on position and parameter groups; performing multi-pose adjustments and screenshots on the model to train a binary descriptor extraction model; inputting the binary descriptors into a dual-channel Transformer model for training, obtaining a tire performance detection model for real-time detection of tire surface images. This invention reduces the amount and difficulty of training data through virtual model simulation and feature dimensionality reduction, avoids dirt interference, improves detection accuracy and robustness, and is suitable for tire quality control.

[0091] Through the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented in hardware or by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. 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 still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A tire performance detection method based on tire defect data and binary descriptors, characterized in that, include: Step S100: Based on the preset defect type, mark the tire defect blocks on the tire surface image and determine the difference features of the tire defect blocks relative to the normal tire surface. Step S200: A virtual tire model is constructed, and differential feature components are configured on the virtual tire model. The data volume of the differential feature components is reduced to obtain the parameter group of the differential feature components. Step S300: Randomly combine the matching differential feature component parameter groups, compare the differential feature components within the combination, and determine the similar parameters. If the average value of the similar parameters is less than or equal to a preset value, reduce the parameter volume of the differential feature component parameter group until the average value of the similar parameters is greater than the preset value. Step S400: Based on the location calibration of the differential feature components and the parameter group of the differential feature components, construct the dimension-reduced feature components on the virtual tire model; Step S500: Adjust the posture of the virtual tire model to obtain several virtual tire model images, and perform binary descriptor extraction model training on the virtual tire model images. Step S600: Input the binary descriptor output by the binary descriptor extraction model into the dual-channel Transformer model for training to obtain the tire performance detection model. Use the tire performance detection model to detect and analyze the tire surface image to determine the defects on the tire surface.

2. The tire performance detection method based on tire defect data and binary descriptors according to claim 1, characterized in that, Methods for detecting and analyzing tire surface images using tire performance detection models include: Step S601: Take a picture of the tire surface to obtain a real-time tire surface image, remove the original tire tread structure in the real-time tire surface image, mark other edges on the tire surface, and map them onto the virtual tire model. Step S602: Fill the marked edge contours on the virtual tire model with content to obtain the dimensionality-reduced feature components on the virtual tire model; Step S603: Adjust the virtual tire model to the angle shown by the real-time tire surface image, take a screenshot to obtain a real-time virtual tire model image, and input the real-time virtual tire model image into the tire performance detection model for detection and analysis.

3. The tire performance detection method based on tire defect data and binary descriptors according to claim 2, characterized in that, Methods for removing the original tire tread structure from real-time tire performance images include: Step S6012: Determine the tire edge in the real-time tire performance image, determine the portion of the tire mapped in the image, and determine the continuous edges in the mapped tire portion. If the length and curvature change of the continuous edges meet the preset standards, the continuous edges are identified as tread structure edges, and the tread structure edges are removed.

4. The tire performance detection method based on tire defect data and binary descriptors according to claim 1, characterized in that, Methods for componentizing differential features include: Step S101: Delineate the tire defect edge of the tire defect block and identify the tire defect edge as the edge frame of the differential feature component.

5. The tire performance detection method based on tire defect data and binary descriptors according to claim 4, characterized in that, Methods for reducing the dimensionality of data for differential feature groups include: Step S201: Determine the rolling direction line of the tire, and construct several vertical lines within the edge frame of the differential feature component. The two ends of the vertical lines contact the edge frame and are perpendicular to the rolling direction line. Combine the length and position of each vertical line in sequence to obtain the parameter group of the differential feature component.

6. The tire performance detection method based on tire defect data and binary descriptors according to claim 5, characterized in that, Methods for randomly combining matching differential feature component parameter sets include: Step S301: Analyze the parameter group of the difference feature component, determine the position of the center point of each vertical line, and connect each center point in turn to form a center point polyline. Step S302: Determine the turning points of the center point broken line, determine the opening direction of the broken line at each turning point, and combine the opening directions of the broken line in sequence to form a broken line opening sequence. Step S303: The broken line opening sequence is identified as the first search tag. The search is performed in the differential feature component parameter group, and the matching differential feature component parameter group is selected. The length of each relative vertical line is used as the second search tag. The corresponding differential feature component parameter group is filtered again, and the finally selected differential feature component parameter group is combined.

7. The tire performance detection method based on tire defect data and binary descriptors according to claim 6, characterized in that, Methods for comparing differential feature components within a combination include: Step S304: According to the method of coinciding the center point polyline in the parameter group of the difference feature components, the difference feature components are overlapped, and the intersection area between the edge frames of the difference feature components is determined. Step S305: Determine the component center of the differential feature component, and radiate several cross-region detection lines outward from the component center as the base point. Mark the intersection point of each cross-region detection line with the edge frame, determine the cross length between the intersection points, and arrange the several cross lengths in order to obtain the cross length sequence. Step S306: Based on the area of ​​the cross region and the area ratio of the differential feature components, as well as the length variation fluctuation characteristics of the cross length in the cross length sequence, determine the similarity parameters between the differential feature components.

8. The tire performance detection method based on tire defect data and binary descriptors according to claim 7, characterized in that, The expression for calculating the similarity parameter is: ； in, For similar parameters, This is the preset maximum area percentage. This represents the area of ​​the overlapping region and the area ratio of the components with different characteristics. For the length variation fluctuation analysis function, the difference in length between adjacent intersections is calculated. If the difference in intersection length is greater than or equal to a preset value, the corresponding intersection length is identified as the first intersection length; otherwise, it is identified as the second intersection length. The ratio of the number of first intersection lengths to the number of second intersection lengths is calculated, and based on the preset number ratio interval to which the ratio belongs, the following is determined: The output area ratio adjustment coefficient. For similarity parameter adjustment coefficients, Adjust constants for similarity parameters.

9. A tire performance detection system based on tire defect data and binary descriptors, characterized in that, include: The first module is used to determine the type of defect on the tire surface, and based on the defect type, to mark the tire defect blocks on the tire surface image, and to determine the difference features of the corresponding tire defect blocks relative to the normal tire surface. The second module is used to construct a virtual tire model, componentize the differential features to obtain differential feature components, configure the differential feature components in the virtual tire model, perform data volume dimensionality reduction analysis on the differential feature components to obtain differential feature component parameter groups, and establish the relationship between differential feature component parameter groups and differential feature components. The third module is used to randomly select combinations of several matching differential feature component parameter groups, compare the differential feature components corresponding to each combination, determine similar parameters, determine the parameter volume requirements for the differential feature component parameter groups based on the similar parameters, and adjust the newly generated differential feature component parameter groups based on the parameter volume requirements. The fourth module is used to construct dimensionality-reduced feature components on the virtual tire model based on the positions of the differential feature components on the virtual tire model and their corresponding parameter groups. The fifth module is used to perform multi-pose adjustments on the virtual tire model, take screenshots of each pose to obtain several virtual tire model images, and train the binary descriptor extraction model based on the virtual tire model images. The sixth module is used to input the binary descriptor output by the binary descriptor extraction model into the dual-channel Transformer model for training, thereby obtaining the tire performance detection model. The tire performance detection model is then used to detect and analyze tire surface images to determine tire surface defects.

Images