Method and system for evaluating value of waste tire based on image processing

Abstract

The application discloses a kind of based on image processing waste tyre value evaluation method and system, the method includes: tire sidewall image is carried out gray scale enhancement processing, and is identified and excluded by edge detection and connected domain analysis character and mark area, obtain the region of interest corresponding to flat rubber surface;In the region of interest, adopt multi-direction filter group to enhance crack feature, get single-pixel-wide skeleton image after binarization, morphological processing and skeletonization;According to the number of neighbors in the eight-neighborhood of skeleton pixel, mark intersection node and end node, and construct undirected graph based on node set and continuous skeleton path between nodes;Extract node density, edge density, average node degree, maximum connected component proportion, loop density and average edge length Topological features form feature vector, input classifier after normalization to obtain aging grade, and determine tire value grade in combination with tread remaining depth.

Description

Technical Field

[0001] This application relates to the field of image processing technology, specifically to a method and system for assessing the value of waste tires based on image processing. Background Technology

[0002] In the process of recycling waste tires, it is usually necessary to determine whether a tire is suitable for retreading, direct use, or disposal based on its wear level, rubber aging condition, and structural damage. Current sorting and assessment methods mostly rely on manual visual inspection, tread depth measurement, or simple image recognition. Manual assessment is greatly affected by operator experience, lighting conditions, and tire surface contamination, resulting in inconsistent judgments and making it difficult to meet the rapid grading requirements of large-scale recycling scenarios.

[0003] In existing image detection methods, some schemes can estimate the remaining tread depth by acquiring tire tread images, or identify surface cracks through geometric features such as edges, area, and length. However, the value of used tires depends not only on tread wear but also closely on the aging degree of the sidewall rubber. The sidewall area often contains brand characters, specification markings, and raised patterns. These non-defect textures have a strong edge response in the image and are easily confused with small cracks, leading to false positives or false negatives. At the same time, cracking caused by rubber aging is not simply an increase in the number of cracks, but rather a gradual development from scattered short cracks into interconnected network cracks. Traditional statistical methods based on crack length, width, or area ratio are difficult to accurately reflect the connection relationship and network degree between cracks.

[0004] Therefore, existing waste tire valuation technologies still suffer from problems such as a single valuation dimension, difficulty in eliminating interference from sidewall characters, insufficient quantification of aging degree, and poor stability of grading results, making it difficult to make a relatively objective and reliable judgment on the retreading value and recycling value of tires. Summary of the Invention

[0005] In view of this, this disclosure proposes a method and system for assessing the value of waste tires based on image processing.

[0006] According to one aspect of this disclosure, a method for assessing the value of waste tires based on image processing is provided, including: The tire sidewall image undergoes grayscale enhancement; edge detection and connected component analysis are performed on the enhanced grayscale image to identify and exclude character and label regions, resulting in a region of interest (ROI) corresponding to the flat rubber surface. The identification and exclusion of character and label regions includes: performing edge detection on the enhanced grayscale image to obtain a binary edge image; performing morphological dilation on the binary edge image to connect adjacent edge pixels; labeling connected components in the dilated image, identifying connected components with an area greater than a threshold as character and label regions; performing secondary dilation on the character and label regions to exclude grayscale transition bands around the characters, generating a mask image; the unlabeled areas in the mask image are the ROI. Within the region of interest, a multi-directional filter bank is used to enhance the cracked grayscale image. After binarization and morphological processing of the enhanced cracked image, crack patterns are extracted, and the crack patterns are skeletonized to obtain a single-pixel-wide skeleton image. Traverse the skeleton pixels in the skeleton image and classify them according to the number of skeleton neighbors in the eight-neighbor area: those with more than three neighbors are marked as intersection nodes, and those with one neighbor are marked as end nodes; construct a node set with all intersection nodes and end nodes; after the node set is determined, start from any node and trace outward pixel by pixel along the skeleton pixels until the next node is reached or the tracing path ends; the skeleton pixels traversed on the tracing path form an edge, and finally obtain the edge set; construct the skeleton image as an undirected graph. Topological features including node density, edge density, average node degree, maximum connected component ratio, loop density, and average edge length are extracted from the undirected graph to form a feature vector; the feature vector is then normalized and input into a classifier to output the aging level. The tire value rating is determined based on the aging level and the remaining tread depth.

[0007] In one possible implementation, the node density is the ratio of the total number of nodes in the node set to the area of ​​the region of interest; the edge density is the ratio of the total number of edges in the edge set to the area of ​​the region of interest; the average node degree is the average degree of all nodes in the node set, where the degree of a node is the number of edges directly connected to that node; the maximum connected component percentage is the ratio of the number of nodes in the maximum connected component of the undirected graph to the total number of nodes; the cycle density is the ratio of the number of independent cycles in the undirected graph to the area of ​​the region of interest, where the number of independent cycles is calculated based on the total number of edges, the total number of nodes, and the number of connected components; and the average edge length is the average weight of all edges in the edge set, where the weight of each edge is the total number of pixels traversed by the skeleton path corresponding to that edge.

[0008] In one possible implementation, adjacent intersection nodes whose spatial distance is less than the aggregation distance threshold are aggregated, and the centroid coordinates of each intersection node within the aggregation range are used as the unique nodes of that intersection region and included in the node set.

[0009] In one possible implementation, the multi-directional filter bank is a multi-directional Gabor filter bank, which includes multiple Gabor filter kernels with different directional angles; the Gabor filter kernels with each directional angle are convolved with the enhanced grayscale image in the region of interest to obtain multiple directional response images; the maximum value of each pixel in the multiple directional response images is taken to obtain the crack enhancement image.

[0010] In one possible implementation, the binarization and morphological processing include: binarizing the cracked enhanced image using a local adaptive thresholding method; performing area filtering on the binarization result to remove isolated connected components with an area smaller than the area filtering threshold; and performing morphological closing operations on the filtered image to connect the cracked segments that are broken due to grayscale fluctuations.

[0011] In one possible implementation, the classifier is a support vector machine classifier using a radial basis function kernel function. A one-to-one strategy is used for the four aging levels. A binary classifier is trained for each combination of two aging levels. The aging level is determined by the voting results of all binary classifiers during prediction.

[0012] In one possible implementation, the determination of the aging level label of the training samples of the support vector machine classifier includes: measuring the Shore A hardness value of the rubber surface near the area where the tire sidewall image is captured, and determining the corresponding aging level label according to the preset hardness range in which the Shore A hardness value is located; the regularization parameter and kernel function bandwidth parameter of the support vector machine classifier are jointly determined by grid search and cross-validation.

[0013] In one possible implementation, determining the tire value level based on the aging level and the remaining tread depth includes: when the aging level reaches a preset aging threshold level, classifying the tire value level into the category of unsuitable for retreading; when the aging level is lower than the aging threshold level, classifying the tire value level based on the remaining tread depth.

[0014] According to another aspect of this disclosure, an image processing-based waste tire valuation system is provided, comprising: The region of interest (ROI) acquisition module is used to perform grayscale enhancement processing on a tire sidewall image; perform edge detection and connected component analysis on the enhanced grayscale image to identify and exclude character and label regions, thereby obtaining the ROI corresponding to the flat rubber surface; wherein, the identification and exclusion of character and label regions includes: performing edge detection on the enhanced grayscale image to obtain a binary edge image; performing morphological dilation on the binary edge image to connect adjacent edge pixels into one; performing connected component labeling on the dilated image, identifying connected components with an area greater than an area threshold as character and label regions; performing secondary dilation on the character and label regions to include the grayscale transition bands around the characters in the exclusion range, generating a mask image; the unlabeled areas in the mask image are the ROI; The crack skeleton generation module is used to perform crack enhancement on the enhanced grayscale image in the region of interest using a multi-directional filter bank, extract crack patterns after binarization and morphological processing of the crack enhanced image, and perform skeletonization processing on the crack patterns to obtain a skeleton image with a single pixel width. An undirected graph construction module is used to traverse the skeleton pixels in the skeleton image and classify and label each skeleton pixel according to the number of skeleton neighbors in its eight-neighbor area: those with more than three neighbors are labeled as intersection nodes, and those with one neighbor are labeled as end nodes. A node set is formed by all intersection nodes and end nodes. After the node set is determined, starting from any node, the system traces outward pixel by pixel along the skeleton pixels until the next node is reached or the tracing path terminates. The skeleton pixels traversed on the tracing path form an edge, and finally, an edge set is obtained, thus constructing the skeleton image into an undirected graph. The aging level output module is used to extract topological features from the undirected graph, including node density, edge density, average node degree, maximum connected component ratio, loop density, and average edge length, to form a feature vector; after normalizing the feature vector, it is input into the classifier to output the aging level. The value rating determination module is used to determine the tire value rating based on the aging level and the remaining tread depth.

[0015] This application improves the accuracy of waste tire sidewall aging detection by enhancing grayscale, excluding character and logo areas, and extracting flat rubber surface areas from tire sidewall images, thereby reducing interference from raised text and patterns on crack recognition. Simultaneously, the crack network is constructed into an undirected graph after multi-directional filtering enhancement, binarization, morphological processing, and skeletonization. Topological features such as node density, edge density, average node degree, maximum connected component ratio, loop density, and average edge length are extracted, enabling quantitative characterization of the aging process where rubber cracks develop from scattered cracks to a connected network structure. Compared to methods that rely solely on crack length, area, or remaining tread depth, this invention more accurately reflects the degree of rubber degradation in the tire sidewall and, combined with remaining tread depth, outputs the tire's value grade, which is beneficial for improving the objectivity and stability of waste tire retreading, direct utilization, or end-of-life sorting.

[0016] Other features and aspects of this disclosure will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0017] The accompanying drawings, which are included in and form part of this specification, illustrate exemplary embodiments, features, and aspects of this disclosure together with the specification and serve to explain the principles of this disclosure.

[0018] Figure 1 This is a schematic diagram of a waste tire valuation method based on image processing, provided in an embodiment of this disclosure.

[0019] Figure 2 This is a flowchart of the construction of an undirected graph provided in an embodiment of this disclosure.

[0020] Figure 3 A flowchart illustrating the aging level classification provided in this embodiment of the disclosure.

[0021] Figure 4 This is a schematic diagram of the structure of a waste tire valuation system based on image processing, provided in an embodiment of this disclosure.

[0022] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0023] Various exemplary embodiments, features, and aspects of this disclosure will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0024] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0025] This disclosure provides an image processing-based method for assessing the value of waste tires. This method is applicable to tire recycling and sorting scenarios. It quantitatively assesses the degree of rubber degradation by performing graph topological analysis on the aging crack patterns on the tire sidewall surface, and comprehensively determines the tire's value level by combining this with the remaining tread depth. Cracking patterns refer to the network or dendritic microcracks that form on the sidewall surface of tire rubber after long-term exposure to heat, oxygen, and ozone. The spatial connectivity of these cracks corresponds to the rubber degradation process. The acquisition device can be an industrial area array camera or a mobile terminal camera with a resolution of at least 5 megapixels. The shooting distance is, for example, 20 to 40 centimeters. Illumination uses a low-angle sidelight source, with the angle between the light source direction and the sidewall surface being, for example, 10 to 20 degrees, to create shadows in the crack grooves and improve the grayscale contrast between them and the flat rubber surface. The shooting area is selected from the flat rubber surface in the middle of the sidewall, avoiding areas with large curvature such as the bead and shoulder.

[0026] The implementation process of the method described in this application will be described in detail below with reference to specific embodiments. It should be noted that this embodiment is only used to explain this application and is not intended to limit the scope of protection of this application. Conventional adjustments or substitutions of each step by those skilled in the art without departing from the concept of this application should be included in the scope of protection of this application.

[0027] Figure 1 A flowchart illustrating an image processing-based method for assessing the value of waste tires according to an embodiment of this disclosure is shown. Figure 1 As shown, the method may include: In step S11, the tire sidewall image is subjected to grayscale enhancement processing; edge detection and connected component analysis are performed on the enhanced grayscale image to identify and exclude character and label regions, obtaining the region of interest corresponding to the flat rubber surface; wherein, the identification and exclusion of character and label regions includes: performing edge detection on the enhanced grayscale image to obtain a binary edge image; performing morphological dilation on the binary edge image to connect adjacent edge pixels into one; performing connected component labeling on the dilated image, determining connected components with an area greater than an area threshold as character and label regions; performing secondary dilation on the character and label regions to include the grayscale transition band around the characters in the exclusion range, generating a mask image; the unlabeled area in the mask image is the region of interest.

[0028] Specifically, after converting the acquired color image of the tire sidewall to a grayscale image, a contrast-limited adaptive histogram equalization (CLAHE) method is used for local contrast enhancement. CLAHE divides the image into several sub-blocks, performs histogram equalization on each sub-block separately, and limits the contrast amplification factor to suppress noise amplification. The sub-block size is set to, for example, 8×8 pixels, and the contrast limitation threshold is set to, for example, 2.0. The enhanced grayscale image serves as the unified input for subsequent steps.

[0029] Tire sidewalls are usually embossed with raised characters and patterns such as brand name and model specifications. The edges and textures produced by these high-contrast non-defect features in the image can be easily confused with minor cracks. If not excluded, subsequent crack extraction will generate a large number of false detections around the characters.

[0030] Canny edge detection is performed on the enhanced grayscale image to obtain a binary edge image. The Canny operator filters edge pixels by calculating pixel gradient magnitude and combining it with a dual threshold, with the high threshold set to 120 and the low threshold to 60, for example. Morphological dilation is performed on the binary edge image, using a rectangle as the structuring element, with a size of, for example, 5×5 pixels, connecting spatially adjacent edge pixels into a single unit. The gaps between character strokes are filled into the overall outline after dilation, while crack edges, due to their sparse distribution, still appear as small strip-like areas after dilation. Connected components are labeled in the dilated image, and the area of ​​each connected component is calculated. Connected components with an area greater than the area threshold are identified as character and label regions. Under the conditions of a shooting distance of 20 to 40 cm and an image resolution of 5 megapixels, the area of ​​the connected components of characters after dilation is typically in the range of 500 to 5000 pixels, while the area of ​​crack segments after dilation is typically in the range of 10 to 200 pixels, a difference of more than an order of magnitude. The area threshold is, for example, set to 500 pixels. Optionally, if the image resolution or shooting distance changes, the area distribution of the two types of connected components can be statistically analyzed on a small number of samples, and the median value between the two distributions can be used as the area threshold.

[0031] A secondary dilation process is performed on the selected connected components of characters and identifiers, with a dilation radius of, for example, 12 pixels. This process excludes the grayscale transition band between the raised edges of the characters and the surrounding adhesive surface, generating a mask image. The unmarked areas in the mask image represent the regions of interest (ROIs) corresponding to the flat adhesive surface, and subsequent crack analysis is performed within these regions.

[0032] In step S12, within the region of interest, a multi-directional filter bank is used to enhance the cracked grayscale image. After binarization and morphological processing of the enhanced cracked image, the crack texture is extracted, and the crack texture is skeletonized to obtain a skeleton image with a single pixel width.

[0033] In some embodiments, cracks within the region of interest are represented as thin, dark lines with low grayscale. A multi-directional Gabor filter bank is used to filter the enhanced grayscale image. A Gabor filter is a direction-selective bandpass filter that exhibits a high response to line structures in specific directions and spatial frequencies. The filter bank contains multiple Gabor filter kernels with different directional angles, for example, 0 degrees, 45 degrees, 90 degrees, and 135 degrees; the wavelength parameter is, for example, 5 pixels, corresponding to the typical width range of cracks in the image; the Gaussian envelope standard deviation is, for example, 2.5 pixels; and the aspect ratio parameter is, for example, 0.5. The Gabor filter kernels at each directional angle are convolved with the enhanced grayscale image within the region of interest to obtain multiple directional response images. The maximum value is taken pixel-by-pixel from these multiple directional response images to obtain the crack enhancement image. Taking the maximum value pixel-by-pixel ensures that if a Gabor kernel in a certain direction produces a high response to a crack, that crack is preserved.

[0034] The cracked image is binarized using a local adaptive thresholding method. A neighborhood window, centered on each pixel and with a window size of, for example, 15×15 pixels, is taken. The average grayscale value of the pixels within the window is calculated as the local threshold for that pixel. A pixel whose grayscale value is lower than the local threshold minus an offset is marked as a candidate cracked pixel. The offset is, for example, 5 grayscale levels. Area filtering is performed on the binarized result to remove isolated connected components with areas smaller than the area filtering threshold, for example, 20 pixels. Then, a morphological closing operation is performed, using a 3×3 square as the structuring element to connect the cracked segments caused by grayscale fluctuations. The closing operation consists of dilation and erosion in sequence: first, dilation fills the short-distance crack gaps, and then erosion restores the original width. Morphological skeletonization is then performed on the binary cracked image after the closing operation. For example, the Zhang-Suen thinning algorithm is used, iteratively peeling away the boundary pixels of the binary regions layer by layer until all line widths are reduced to a single pixel, resulting in a skeleton image.

[0035] In step S13, the skeleton pixels in the skeleton image are traversed, and each skeleton pixel is classified and labeled according to the number of skeleton neighbors in its eight-neighborhood: those with no less than three neighbors are labeled as intersection nodes, and those with one neighbor are labeled as end nodes; a node set is formed by all intersection nodes and end nodes. After the node set is determined, starting from any node, the skeleton pixels are traced outward pixel by pixel until the next node is reached or the tracing path ends; the skeleton pixels traversed on the tracing path form an edge, and finally the edge set is obtained, thus constructing the skeleton image as an undirected graph.

[0036] After transforming the pixel-level image representation of crack patterns into a graph structure, graph theory tools can be used to quantify the connectivity and network formation of the crack network. Traditional crack detection methods statistically analyze the geometric features of cracks, such as total crack length, maximum width, or area percentage. These features reflect the number and size of cracks, but cannot describe the connections between cracks. During rubber aging, cracks gradually evolve from scattered short segments into a dense, interconnected grid. This gradual change is reflected in the topological features of the graph. That is, two sets of cracks with the same total length, if one consists of a large number of isolated short cracks and the other consists of a small number of interconnected long cracks, will have drastically different aging degrees, which geometric features cannot distinguish.

[0037] Specifically, Figure 2 A flowchart illustrating the construction of an undirected graph provided in an embodiment of this disclosure is shown, such as... Figure 2 As shown, in S1301, for each skeleton pixel with a value of 1 in the skeleton image, check the number of skeleton neighbors with a value of 1 in its 8 neighborhoods (i.e., the 8 adjacent pixel positions in the top, bottom, left, right and 4 diagonal directions), and record this number as the neighbor count.

[0038] Specifically, the skeleton pixels are classified and labeled according to the neighbor count. When the neighbor count is not less than 3, the pixel is located at the fork or intersection of the crack and is marked as an intersection node. For example, a neighbor count of 3 corresponds to a T-shaped intersection, that is, 3 crack segments converge here, and a neighbor count of 4 corresponds to a cross intersection, that is, 4 crack segments intersect. When the neighbor count is 1, the pixel is located at the end of the crack segment and is marked as an end node. When the neighbor count is 2, the pixel is located in the middle of the crack segment and is not treated as a node, but only as part of the subsequent edge. Isolated pixels with a neighbor count of 0 have been removed in the aforementioned area filtering.

[0039] In a real skeleton image, an intersection region may contain multiple adjacent intersection node pixels. Adjacent intersection nodes with a spatial distance less than an aggregation distance threshold are aggregated, and the centroid coordinates of each intersection node within the aggregation range are used as the unique node for that intersection region and included in the node set. The aggregation distance threshold is determined based on the typical size of the intersection region after skeletonization. Intersection regions generated by the Zhang-Suen thinning algorithm are typically no more than 3×3 pixels, so the aggregation distance threshold is, for example, 3 pixels. In some embodiments, this threshold can be adjusted within the range of 2 to 5 pixels according to the image resolution and the typical width of the cracks. All intersection nodes and end nodes together constitute a node set V, and each node records its coordinate position in the image.

[0040] For example, in a certain local skeleton region there is a cross intersection. After skeletonization, the intersection region consists of 4 adjacent pixels with a neighbor count of no less than 3. The coordinates of these 4 pixels are (100,200), (101,200), (100,201), and (101,201), and the distance between each pair of pixels does not exceed 3 pixels. After aggregation, the centroid (100.5,200.5) is used as the unique node of the intersection region.

[0041] In S1302, after the node set is determined, the connections between nodes are traced along the skeleton lines to construct an edge set, thus building the skeleton image into an undirected graph. Starting from any node, the skeleton pixels are traced outward pixel by pixel until the next node is reached or the tracing path terminates. The skeleton pixels traversed on the tracing path form an edge, with the starting node and the ending node at the two ends; the total number of pixels traversed by the tracing path is the weight of the edge, reflecting the pixel length of the corresponding crack segment. The above process is repeated for all untraced skeleton pixels in the skeleton image until all skeleton pixels are assigned to an edge or marked as a node, finally obtaining the edge set. The set of nodes consisting of all intersection nodes and end nodes. With the above set of edges Jointly construct an undirected graph .

[0042] For example, if a skeleton image in a certain region contains 3 intersection nodes and 2 end nodes, and is connected by 6 skeleton line segments, then a local subgraph containing 5 nodes and 6 edges is constructed.

[0043] In step S14, topological features including node density, edge density, average node degree, maximum connected component ratio, loop density, and average edge length are extracted from the undirected graph to form a feature vector; the feature vector is normalized and then input into a classifier to output the aging level.

[0044] Specifically, Figure 3 A flowchart for classifying aging levels provided in embodiments of this disclosure is shown below. Figure 3 As shown, in S1401, in the undirected graph Six topological features were extracted to quantify the structural differences in the cracked network at different aging stages. The area of ​​the region of interest is denoted as... The number of connected components in the graph is expressed in megapixels. The connected component is the largest subgraph formed by mutually reachable nodes in the graph, obtained by performing a depth-first search traversal on all nodes in the graph.

[0045] Among them, node density , defined as the ratio of the total number of nodes within the region of interest to the area of ​​the region:

[0046] in The total number of nodes. The area of ​​the region of interest is expressed in megapixels. The unit is per 10,000 pixels. In the early stages of aging, the cracks are sparse and have few intersections. Low value; as aging deepens, cracks increase and frequently intersect and branch. The value increases. For example, if there are 120 nodes within a region of interest of 500,000 pixels, then... Each 10,000 pixels.

[0047] Edge density The ratio of the total number of edges in the edge set to the area of ​​the region of interest is defined as follows:

[0048] in Let the total number of edges be . The meaning and unit are the same as above; this feature reflects the quantity density of cracked segments.

[0049] Average node degree , defined as the average degree of all nodes in the node set:

[0050] Degree of a node This represents the number of edges directly connected to the node. End nodes have a degree of 1, T-shaped intersection nodes have a degree of 3, and cross-shaped intersection nodes have a degree of 4. The higher the value, the greater the proportion of intersection nodes and the more branches the cracked network has.

[0051] Among them, the proportion of the largest connected component The maximum connected component of an undirected graph is defined as the ratio of the number of nodes in the maximum connected component to the total number of nodes.

[0052] in The number of nodes in the maximum connected component. The range of values ​​is In the early stages of aging, the cracks are isolated from each other, and there are many small connected components in the diagram. Low value; as aging deepens, cracks become interconnected, and small connected components gradually merge into larger connected networks. It approaches 1; this feature directly reflects the degree of connectivity of the cracked network.

[0053] For example, if an undirected graph contains 80 nodes distributed across 12 connected components, and the largest connected component contains 35 nodes, then As aging deepens, the cracks interconnect, and the largest connected component may expand to contain 70 nodes (assuming the total number of nodes increases to 100). .

[0054] Among them, for loop density An independent cycle is a closed path in the graph that allows you to return to the starting point by walking along its edges. The number of independent cycles is... We obtain this from Euler's formula in graph theory:

[0055] in , , The meaning is the same as before. This reflects the number of closed meshes formed by cracks. Loop density is defined as the ratio of the number of independent loops to the area of ​​the region of interest.

[0056] In the early stages of aging, cracks extend in a tree-like pattern with very few loops; in the later stages of aging, cracks crisscross to form a mesh, and the number of loops increases significantly. For example, if an undirected graph contains 15 nodes, 20 edges, and 2 connected components, then... There are 7 independent loops; if the area of ​​interest is 500,000 pixels, then Each 10,000 pixels.

[0057] Average side length , defined as the average weight of all edges in the edge set:

[0058] in For the edge The weight of the edge is the total number of pixels traversed by the skeleton path corresponding to that edge. The unit is pixels. In the early stages of aging, there are fewer crossover points, and the crack segments between two crossover points are relatively long. The value is large; the intersection points are dense in the later stages of aging, and the cracked segments are frequently truncated. The value decreases.

[0059] The above 6 topological features are arranged as follows: , , , , , The order of the elements forms a 6-dimensional feature vector. .

[0060] In S1402, the feature vector is normalized and then input into the classifier, which outputs the aging level. Specifically, in one embodiment, the six topological features have different dimensions and numerical ranges, and need to be normalized before being input into the classifier. For each dimension of the feature vector, based on the minimum and maximum values ​​of that dimension in the training sample set, the feature value is mapped to the interval between 0 and 1:

[0061] in For the first The original eigenvalues ​​of the dimension, and The first The minimum and maximum values ​​of the dimensional feature in the training sample set. These are the normalized eigenvalues. The normalized eigenvectors are denoted as... A one-dimensional vector with dimensions 1×6.

[0062] In one embodiment, the classifier uses a Support Vector Machine (SVM), and the kernel function is a Radial Basis Function (RBF). SVM constructs a classification hyperplane based on the maximum margin principle, exhibiting good generalization ability for small-sample classification tasks in low- to medium-dimensional feature spaces. It is suitable for scenarios where the feature dimension is 6 and the training sample size is in the hundreds to thousands. The classification objective is to map the normalized feature vectors to one of four aging levels: no obvious aging is labeled as level 1, mild aging as level 2, moderate aging as level 3, and severe aging as level 4.

[0063] SVM is a binary classification method. When dealing with multi-class problems with four aging levels, it adopts a one-vs-one strategy: a binary classifier is trained for every two levels out of the four levels, for a total of [number missing] training classes. There are six binary classifiers. During prediction, each of the six binary classifiers judges and votes on the input normalized feature vector, and the level with the most votes is used as the final output aging level.

[0064] The RBF kernel function for each binary classifier is defined as follows:

[0065] in and These are two normalized feature vectors. It is the Euclidean norm. The kernel function bandwidth parameter controls the rate at which the kernel function value decays as the distance between samples increases. The optimization objective solved during training is:

[0066] The constraints are and ,in The normal vector of the classification hyperplane in the high-dimensional feature space. For bias terms, The implicit feature mapping corresponding to the RBF kernel. For the first The class labels of each training sample. The total number of training samples, For the first Slack variables for each sample. The regularization parameter controls the balance between the classification margin and the penalty for misclassification. The first term in the objective function maximizes the classification margin, while the second term penalizes misclassified samples. Training is solved using the Sequence Minimum Optimization (SMO) algorithm.

[0067] Optionally, the aging grade labels for the training samples are determined as follows: Specifically, the Shore A hardness value is measured on the rubber surface near the area where the tire sidewall image was captured using a Shore A durometer. Shore A hardness is a commonly used hardness index in the rubber industry; a higher value indicates that the rubber is harder and less elastic. The corresponding aging grade label is determined based on the preset hardness range in which the Shore A hardness value falls. For example, a hardness below 60 is grade 1, 60 to 68 is grade 2, 68 to 75 is grade 3, and above 75 is grade 4. The division of the above hardness ranges refers to common experience in the tire retreading industry; in some embodiments, the boundary values ​​of each range can be adjusted according to statistical data from specific application scenarios. At least 200 sidewall images of different tires are collected as training samples for each grade.

[0068] It is understandable that the correlation between the six topological features and the aging level label is that rubber aging causes cracks to gradually evolve from an isolated state to a connected mesh state. This physical process is reflected in the monotonic change trend of topological features (such as increased node density, increased proportion of the largest connected component, increased loop density, and shortened average edge length) and the increase of Shore A hardness. Both are different dimensions of the same aging process.

[0069] In one embodiment, the regularization parameter and kernel function bandwidth parameters It is determined through a combination of grid search and cross-validation. Specifically, it will... The candidate value is set as , The candidate value is set as ,form Several parameter combinations are used. For each combination, 5-fold cross-validation is performed on the training set: the training set is divided into 5 equal parts, and one part is used as the validation set and the remaining 4 parts as the training subset in turn. The average classification accuracy of the 5 validations is recorded, and the parameter combination with the highest accuracy is selected as the final hyperparameter. The trained SVM classifier receives a 1×6 normalized feature vector as input and outputs an integer value as the aging level, which can be 1, 2, 3, or 4. In some embodiments, to prevent overfitting, L2 regularization constraints can be applied to the feature vector during training.

[0070] In step S15, the tire value level is determined based on the aging level and the remaining tread depth. Specifically, the aging level determination result is combined with the remaining tread depth to determine the tire value level. When the aging level reaches a preset aging threshold level, the tire value level is classified as unsuitable for retreading. For example, the aging threshold level is set to level 3, meaning that tires with an aging level of level 3 or 4, even if the tread condition is still good, have rubber material deterioration that no longer meets the safety requirements for retreading and are directly classified as unsuitable for retreading. When the aging level is lower than the aging threshold level, i.e., tires with a level of 1 or 2, the tire value level is determined based on the remaining tread depth.

[0071] The remaining tread depth can be assessed using existing image estimation methods based on shadow features. Optionally, the detection results of visible structural damage such as cuts and bulges on the sidewall can also be included as additional criteria in the comprehensive assessment.

[0072] For example, the assessment of remaining tread depth specifically includes the following steps: The input is a grayscale image of the tire tread crown, with an image size of, for example, 2592×1944 pixels and 8-bit grayscale depth. Gaussian filtering (kernel size 5×5, σ=1.0) is applied to the input image to eliminate interference from sensor noise and uneven illumination. Subsequently, Otsu's method is used to perform global binarization on the image, separating the tread rubber surface (high grayscale area, corresponding to tread blocks) from the groove area (low grayscale area). Connected component analysis is performed on the binarization result, extracting the largest connected components as the tread block area; the groove area is the gap area between the tread blocks. Within the groove area, a rectangular sub-band with a width equal to the actual width of the groove is extracted along the groove extension direction as an analysis window.

[0073] Within the analysis window, grayscale values ​​are scanned line by line along the direction perpendicular to the trench's extension. A typical trench cross-section exhibits a three-segment grayscale profile: high grayscale corresponds to the near-end trench wall (light-facing side) directly illuminated by the light source; low grayscale corresponds to the shadowed area at the bottom of the trench; and high grayscale at the far end corresponds to the top of the far-end trench wall (top of the backlit side) illuminated by the light source. The consecutive pixel segment with the lowest grayscale value in each line is marked as the shadow area. The shadow areas obtained from all scan lines are vertically connected and merged to form a complete shadow band at the bottom of the trench. The average width of the shadow band along the trench's extension direction is calculated and denoted as the shadow width. (Unit: pixels).

[0074] The remaining tread depth is calculated based on the geometric relationship between shadow width and groove depth. Let the incident angle of the light source (the angle between the light source direction and the normal to the tread surface) be... The width of the trench is (Unit: pixels, calculated from the pixel distance between the edges of the tread blocks on both sides of the groove), the shadow width is... (Unit: pixels). When a light source shines from one side of the trench, the bottom of the trench is blocked, forming a shadow. The width of the shadow is... With trench depth The following geometric relationships exist between (unit: pixels):

[0075] because and All values ​​are in pixels, multiplied by a calibration factor. (Unit: mm / pixel) can then be converted to actual physical depth. When the trench depth is shallow, the shadow may not cover the entire trench bottom. In this case, it needs to be adjusted according to the ratio of the shadow width to the trench width: if Then calculate directly using the formula above; if This indicates that the shadow has covered most of the groove bottom area and is quite deep, so the calculation can be done directly using the formula. The typical depth of the main groove of a new tire is about 15-18 mm, and the wear limit is 1.6 mm; the depth range that this method can measure is 1.6-18 mm.

[0076] As an example, determining a tire's value rating based on its aging grade and remaining tread depth includes, where the input may include the aging grade. Remaining tread depth (Unit: mm); Specific judgment rules (preset aging threshold level is 3) may include: if The rubber material is severely deteriorated and is directly classified as "unsuitable for refurbishment" (value level: scrap / shredding). If ,according to Classification: If so, it is "renovable (superior grade)"; If so, then it is "renovable (qualified)"; If the tread is too thin, then it should be used directly (no economical retreading). For example, tire A: , Output "Retreadable (Superior)"; Tire B: , (Deep but aging has reached the threshold), output "Unsuitable for retreading"; Tire C: , The output is "Use directly".

[0077] This method constructs an undirected graph of the sidewall crack network skeleton and extracts topological features, introducing the degree of rubber degradation into tire value assessment. Compared with traditional methods that only count geometric features such as crack length and area, the proportion of the largest connected component and the loop density in the topological features can capture the gradual evolution of cracks from scattered to networked, and has a stronger ability to distinguish different aging stages. The exclusion mechanism for character and label regions uses the difference in connected region area for filtering, which is not dependent on a specific brand template and can be applied to tires of different brands and models.

[0078] Figure 4 A block diagram of an image processing-based waste tire valuation system according to an embodiment of the present disclosure is shown. Figure 4 As shown, the system includes: The region of interest (ROI) acquisition module 4001 is used to perform grayscale enhancement processing on a tire sidewall image; perform edge detection and connected component analysis on the enhanced grayscale image to identify and exclude character and label regions, thereby obtaining the ROI corresponding to the flat rubber surface; wherein, the identification and exclusion of character and label regions includes: performing edge detection on the enhanced grayscale image to obtain a binary edge image; performing morphological dilation on the binary edge image to connect adjacent edge pixels into one; marking connected components in the dilated image, identifying connected components with an area greater than an area threshold as character and label regions; performing secondary dilation on the character and label regions to include the grayscale transition bands around the characters in the exclusion range, generating a mask image; the unmarked areas in the mask image are the ROI. The crack skeleton generation module 4002 is used to perform crack enhancement on the enhanced grayscale image in the region of interest using a multi-directional filter bank, extract crack patterns after binarization and morphological processing of the crack enhanced image, and perform skeletonization processing on the crack patterns to obtain a skeleton image with a single pixel width. The undirected graph construction module 4003 is used to traverse the skeleton pixels in the skeleton image and classify and label them according to the number of skeleton neighbors in the eight-neighbor area of ​​each skeleton pixel: those with no less than three neighbors are labeled as intersection nodes, and those with one neighbor are labeled as end nodes; a node set is formed by all intersection nodes and end nodes. After the node set is determined, starting from any node, the system traces outward pixel by pixel along the skeleton pixels until the next node is reached or the tracing path terminates; the skeleton pixels traversed on the tracing path form an edge, and finally the edge set is obtained, thus constructing the skeleton image into an undirected graph. The aging level output module 4004 is used to extract topological features from the undirected graph, including node density, edge density, average node degree, maximum connected component ratio, loop density, and average edge length, to form a feature vector; after normalizing the feature vector, it is input into a classifier to output the aging level. The value rating determination module 4005 is used to determine the tire value rating based on the aging rating and the remaining tread depth.

[0079] In some embodiments, the functions or modules of the apparatus provided in this disclosure can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0080] Figure 5 A block diagram of an electronic device according to an embodiment of the present disclosure is shown. (Refer to...) Figure 5 Electronic devices can be provided as servers or terminal devices. (See reference...) Figure 5 The electronic device includes a processing component 501, which further includes one or more processors, and memory resources represented by memory 502 for storing instructions, such as application programs, that can be executed by the processing component 501. The application programs stored in memory 502 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 501 is configured to execute instructions to perform the methods described above.

[0081] The electronic device may also include a power supply component 503 configured to perform power management of the electronic device, a wired or wireless network interface 504 configured to connect the electronic device to a network, and an input / output interface 505 (I / O interface). The electronic device can operate on an operating system stored in memory 502.

[0082] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 502 including computer program instructions that can be executed by a processing component 501 of an electronic device to perform the above-described method.

[0083] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or technical improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A method for assessing the value of waste tires based on image processing, characterized in that, include: Perform grayscale enhancement processing on the tire sidewall image; Edge detection and connected component analysis are performed on the enhanced grayscale image to identify and exclude character and label regions, resulting in a region of interest (ROI) corresponding to a flat surface. The identification and exclusion of character and label regions includes: performing edge detection on the enhanced grayscale image to obtain a binary edge image; performing morphological dilation on the binary edge image to connect adjacent edge pixels; labeling connected components in the dilated image, identifying connected components with an area greater than a threshold as character and label regions; performing secondary dilation on the character and label regions to exclude grayscale transition bands around the characters, generating a mask image; the unlabeled areas in the mask image are the ROI. Within the region of interest, a multi-directional filter bank is used to enhance the cracked grayscale image. After binarization and morphological processing of the enhanced cracked image, crack patterns are extracted, and the crack patterns are skeletonized to obtain a single-pixel-wide skeleton image. Traverse the skeleton pixels in the skeleton image and classify them according to the number of skeleton neighbors in the eight-neighbor area: those with more than three neighbors are marked as intersection nodes, and those with one neighbor are marked as end nodes. A node set is formed by all intersection nodes and end nodes. After the node set is determined, start from any node and trace outward pixel by pixel along the skeleton pixels until the next node is reached or the tracing path ends. The skeleton pixels passed on the tracing path form an edge, and finally the edge set is obtained, and the skeleton image is constructed as an undirected graph. Topological features including node density, edge density, average node degree, maximum connected component ratio, loop density, and average edge length are extracted from the undirected graph to form a feature vector; the feature vector is then normalized and input into a classifier to output the aging level. The tire value rating is determined based on the aging level and the remaining tread depth.

2. The method according to claim 1, characterized in that, The node density is the ratio of the total number of nodes in the node set to the area of ​​the region of interest; the edge density is the ratio of the total number of edges in the edge set to the area of ​​the region of interest; the average node degree is the average degree of all nodes in the node set, where the degree of a node is the number of edges directly connected to that node; the maximum connected component percentage is the ratio of the number of nodes in the maximum connected component of the undirected graph to the total number of nodes; the cycle density is the ratio of the number of independent cycles in the undirected graph to the area of ​​the region of interest, where the number of independent cycles is calculated based on the total number of edges, the total number of nodes, and the number of connected components; the average edge length is the average weight of all edges in the edge set, where the weight of each edge is the total number of pixels traversed by the skeleton path corresponding to that edge.

3. The method according to claim 1, characterized in that, Adjacent intersection nodes whose spatial distance is less than the aggregation distance threshold are aggregated, and the centroid coordinates of each intersection node within the aggregation range are used as the unique nodes of the intersection region and included in the node set.

4. The method according to claim 1, characterized in that, The multi-directional filter bank is a multi-directional Gabor filter bank, which contains multiple Gabor filter kernels with different directional angles. The Gabor filter kernels with each directional angle are convolved with the enhanced grayscale image in the region of interest to obtain multiple directional response images. The maximum value of each pixel in the multiple directional response images is taken to obtain the crack enhancement image.

5. The method according to claim 1, characterized in that, The binarization and morphological processing include: binarizing the cracked enhanced image using a local adaptive thresholding method; performing area filtering on the binarization result to remove isolated connected components with an area smaller than the area filtering threshold; and performing morphological closing operations on the filtered image to connect the cracked segments that are broken due to grayscale fluctuations.

6. The method according to claim 1, characterized in that, The classifier is a support vector machine classifier using radial basis function kernel function. It adopts a one-to-one strategy for the four aging levels, and trains a binary classifier for each combination of two aging levels. The aging level is determined by the voting results of all binary classifiers during prediction.

7. The method according to claim 6, characterized in that, The determination of the aging level label for the training samples of the support vector machine classifier includes: measuring the Shore A hardness value of the rubber surface near the area where the tire sidewall image is captured, and determining the corresponding aging level label based on the preset hardness range in which the Shore A hardness value is located; the regularization parameter and kernel function bandwidth parameter of the support vector machine classifier are jointly determined by grid search and cross-validation.

8. The method according to claim 1, characterized in that, The method of determining the tire value level based on the aging level and the remaining tread depth includes: when the aging level reaches a preset aging threshold level, the tire value level is classified into the category of unsuitable for retreading; when the aging level is lower than the aging threshold level, the tire value level is classified according to the remaining tread depth.

9. A waste tire value assessment system based on image processing, characterized in that, include: The region of interest acquisition module is used to perform grayscale enhancement processing on the tire sidewall image; Edge detection and connected component analysis are performed on the enhanced grayscale image to identify and exclude character and label regions, resulting in a region of interest (ROI) corresponding to a flat surface. The identification and exclusion of character and label regions includes: performing edge detection on the enhanced grayscale image to obtain a binary edge image; performing morphological dilation on the binary edge image to connect adjacent edge pixels; labeling connected components in the dilated image, identifying connected components with an area greater than a threshold as character and label regions; performing secondary dilation on the character and label regions to exclude grayscale transition bands around the characters, generating a mask image; the unlabeled areas in the mask image are the ROI. The crack skeleton generation module is used to perform crack enhancement on the enhanced grayscale image in the region of interest using a multi-directional filter bank, extract crack patterns after binarization and morphological processing of the crack enhanced image, and perform skeletonization processing on the crack patterns to obtain a skeleton image with a single pixel width. An undirected graph construction module is used to traverse the skeleton pixels in the skeleton image and classify and label each skeleton pixel according to the number of skeleton neighbors in its eight-neighbor area: those with more than three neighbors are labeled as intersection nodes, and those with one neighbor are labeled as end nodes. A node set is formed by all intersection nodes and end nodes. After the node set is determined, starting from any node, the system traces outward pixel by pixel along the skeleton pixels until the next node is reached or the tracing path terminates. The skeleton pixels traversed on the tracing path form an edge, and finally, an edge set is obtained, thus constructing the skeleton image into an undirected graph. The aging level output module is used to extract topological features from the undirected graph, including node density, edge density, average node degree, maximum connected component ratio, loop density, and average edge length, to form a feature vector; after normalizing the feature vector, it is input into the classifier to output the aging level. The value rating determination module is used to determine the tire value rating based on the aging level and the remaining tread depth.

Images