Special-shaped column identification method based on arc line segment combination

Abstract

The invention relates to the technical field of special-shaped column identification, in particular to a special-shaped column identification method based on arc line segment combination. The method mainly comprises the following steps: acquiring image data of a special-shaped column, and extracting an initial contour through denoising and edge detection; adopting a geometric feature segmentation method, and identifying an arc-shaped line segment and a straight line segment through Hough transform; constructing a geometric model by using a graph neural network, and forming a topological structure of the feature points; matching the geometric model with a standard database by using a graph matching algorithm, and analyzing and identifying the cylinder type through curvature, length and spatial distribution; and optimizing an identification result through a genetic algorithm and outputting a final special-shaped column identification result. According to the method, technologies such as image processing, geometric feature extraction, the graph neural network and the genetic algorithm are utilized, efficient and accurate recognition of the special-shaped column is achieved, and the accuracy and efficiency of special-shaped column recognition are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of special-shaped column recognition, and particularly to a special-shaped column recognition method based on a combination of arc segments. Background Art

[0002] With the development of modern manufacturing processes, special-shaped column structures have been widely used in many fields such as architecture, aerospace, automotive manufacturing, and industrial design. Special-shaped columns usually have complex geometric shapes, and their cross-sections may exhibit various irregular shapes, such as arcs, combinations of multiple straight lines and arcs, free curves, etc. Therefore, the detection and recognition of special-shaped columns have become key steps in the process of precision manufacturing and quality control. The following problems still exist: Traditional recognition methods are mostly based on feature extraction of straight line segments and regular geometric figures. However, in the face of complex arc segment combination structures, it is difficult for existing technologies to achieve high-precision detection and recognition, and there is a lack of accurate modeling of complex geometric forms; The deficiencies in shape segmentation and feature extraction of existing technologies result in low recognition accuracy of special-shaped columns; In existing shape matching technologies, there is a lack of accurate matching methods for the characteristics of complex columns, resulting in large errors in the matching results of special-shaped columns and prone to recognition errors and misjudgments. Summary of the Invention

[0003] To solve the above problems, the present invention provides a special-shaped column recognition method based on a combination of arc segments, which solves the problem of how to achieve high-precision shape segmentation, feature extraction, and accurate matching recognition in the special-shaped column structure composed of complex arc segments and multiple straight lines, thereby improving the accuracy and efficiency of special-shaped column recognition and reducing errors.

[0004] To achieve the above object, the technical solution adopted by the present invention is:

[0005] A special-shaped column recognition method based on a combination of arc segments, comprising the following steps:

[0006] S1: Obtain the image data of the special-shaped column through an image acquisition device, perform preprocessing operations, eliminate the noise in the image through a denoising algorithm, and use an edge detection algorithm to obtain the initial contour number of the special-shaped column;

[0007] S2: Based on the initial contour data, adopt a geometric feature segmentation method of arc segments and straight line segments to disassemble the external form of the special-shaped column into several combined line segments, and identify the arc segment and straight line segment features therein through the Hough transform algorithm;

[0008] S3: Based on the extracted arc and straight line segment features, construct a geometric model of the special-shaped column through a graph neural network algorithm, and establish a topological structure diagram of feature points through the spatial structure relationship of the combination of arcs and straight lines;

[0009] S4: The shape matching algorithm based on graph matching performs shape matching between the geometric model and a known standard special-shaped column database, and identifies the column type corresponding to the special-shaped line segment combination through curvature matching, length comparison, and spatial distribution analysis;

[0010] S5: Based on the identified column type of the special-shaped column, the genetic algorithm is used to optimize the recognition result and output the final special-shaped column recognition result.

[0011] Furthermore, the image data includes the outer contour, surface texture, geometric shape information, color information, and depth information of the special-shaped column.

[0012] Furthermore, the initial contour numbers include multiple characteristic contour points of the outer contour of the special-shaped column extracted by the edge detection algorithm. The distribution of the characteristic contour points is divided into several local contour regions according to the geometric characteristics of the special-shaped column, and each region contains one or more contour lines corresponding to arc segments or straight line segments.

[0013] Furthermore, step S2 includes the following steps:

[0014] Based on the initial contour data, key characteristic points on the contour are selected by analyzing the local curvature change rate, and these characteristic points are classified as high-curvature points, low-curvature points, and inflection points;

[0015] The dynamic window segmentation algorithm is applied to adaptively adjust the size of the segmentation window according to the distribution of characteristic points, local curvature change, and geometric morphology, and perform local curvature and direction analysis on the contour line to initially identify the candidate regions of arc segments and straight line segments;

[0016] In the candidate regions, the weighted Hough transform algorithm is used to extract local line segment features, and the geometric features of arc segments and straight line segments are respectively identified through the voting calculation of each contour point;

[0017] For the identified arc segments and straight line segments, the least squares method is applied to fit and correct their geometric parameters, including the center, radius, and curvature parameters of the arc segments, as well as the direction and length of the straight line segments, and finally the optimized arc segment and straight line segment feature parameters are output.

[0018] Furthermore, step S3 includes the following steps:

[0019] According to the extracted arc segment and straight line segment features, a geometric feature map of the special-shaped column is constructed. The geometric feature map includes the position information of the line segments, line segment types, curvatures, angles, and lengths of the line segments;

[0020] Based on the geometric feature map, an adaptive weighting strategy is adopted to encode the connection relationship between the arc segments and the straight segments in a graph structure. The weight between each edge is dynamically adjusted according to the spatial distance, direction change, and curvature change to form an initial spatial topological structure;

[0021] Based on the constructed spatial topological structure, a graph neural network algorithm is used to construct the geometric model of the special-shaped column. Through the embedding and propagation of node features and edge features, the graph neural network learns the spatial relationship and topological characteristics between different line segments, gradually optimizing the geometric model of the special-shaped column. The finally output geometric model includes the global shape description and local geometric details of the special-shaped column.

[0022] Furthermore, the formula of the geometric model of the special-shaped column is as follows:

[0023]

[0024] where M(υ) represents the geometric model of the special-shaped column, which contains the geometric information of local feature points and their adjacency relationships; h υ represents the geometric feature representation of node v, which is used to describe the local geometric information of a certain point on the column; represents other feature points directly connected to the feature point v on the special-shaped column; c υu and c υw represent the spatial weights between adjacent points; α, β, and γ represent the adjustment parameters in the model, which are used to control the influence weights of different geometric features on the overall model.

[0025] Further, the step S4 includes the following steps:

[0026] Based on the geometric model, using the graph matching algorithm, the topological structure of the geometric model is shape-matched with the models in the standard special-shaped column database, and the comparison is made through the position of feature points, the curvature of line segments, and the spatial distribution characteristics;

[0027] During the shape matching process, a multi-scale feature extraction algorithm is adopted to analyze the different scale information of the arc segments and straight segments in the geometric model;

[0028] Calculate the curvature difference between the matching model and the database model through the curvature matching algorithm, and compare the curvature distribution and curve change trend of the arc segments;

[0029] Based on the line segment length comparison algorithm, analyze the length difference between the identified geometric model and the corresponding arc and straight segments in the standard model, and compensate and adjust the length error;

[0030] Through spatial distribution analysis, the relative positional relationship between the arc segments and straight segments in the geometric model in space is accurately compared. Combining the changes in the spatial relative positions, the global shape matching degree of the model is judged, and finally the recognition result of the special-shaped column is determined.

[0031] Furthermore, the formula of the graph matching algorithm is as follows:

[0032]

[0033] Among them, E match represents the matching degree between the geometric model obtained by the graph matching algorithm and the standard special-shaped column database model; υ i and υ' i respectively represent the feature points in the recognized special-shaped column geometric model and the corresponding feature points in the standard special-shaped column database; d(υ i , υ' i ) represents the spatial position difference between the feature point υ i in the geometric model and the feature point υ' i in the database model; Δk(υ i , υ' i ) represents the curvature difference of the arc segments where the feature points υ i and υ' i are located; Δl(υ i , υ' i ) represents the length difference of the corresponding arc or straight segments in the geometric model and the database model; α 1 , α 2 and α 3 respectively represent the weight parameters of the position difference, curvature difference, and length difference; (υ i , υ j ) represents a pair of feature points.

[0034] Further, when the genetic algorithm is used to optimize the recognition result, by calculating the fitness function of the column type matching result, the optimal column matching result is selected.

[0035] The beneficial effects of the present invention are as follows: The present invention obtains the image data of the special-shaped column through an image acquisition device, and uses a denoising algorithm to eliminate noise, improving the image quality and reducing the errors caused by image noise, making the subsequent contour extraction more accurate. Based on edge detection and Hough transform algorithms, geometric features such as arc segments and straight segments of the special-shaped column can be effectively identified, enabling the complex shape of the special-shaped column to be decomposed into recognizable basic shape combinations for subsequent processing. By constructing a geometric model through a graph neural network algorithm, the spatial structure information of the combination of arcs and straight lines can be fully utilized to establish a more accurate topological structure diagram, enabling the shape and structure information of the special-shaped column to be accurately represented and enhancing the accuracy of shape matching. Based on the graph matching algorithm, the extracted geometric model is shape-matched with a standard database, and through methods such as curvature matching, length comparison, and spatial distribution analysis, the column type corresponding to the special-shaped column can be efficiently and accurately identified, improving the accuracy of identification. Using a genetic algorithm to optimize the preliminary identification results can search for the optimal solution within a large range, further improving the accuracy and reliability of the identification results and ensuring that the final output of the special-shaped column identification results has high precision. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 FIG. is a schematic flowchart of a method for identifying a special-shaped column based on a combination of arc segments according to the present invention.

[0037] Figure 2 FIG. is a schematic flowchart of step S3 provided by an embodiment of the present invention.

[0038] Figure 3 FIG. is a schematic flowchart of step S4 provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] Please refer to Figures 1-3 as shown, the present invention relates to a method for identifying a special-shaped column based on a combination of arc segments.

[0040] Embodiment

[0041] A method for identifying a special-shaped column based on a combination of arc segments includes the following steps:

[0042] S1: Obtain the image data of the special-shaped column through an image acquisition device, perform preprocessing operations, eliminate the noise in the image through a denoising algorithm, and obtain the initial contour number of the special-shaped column using an edge detection algorithm;

[0043] Among them, the image data includes the outer contour, surface texture, geometric shape information, color information, and depth information of the special-shaped column; the initial contour numbers include multiple characteristic contour points of the outer contour of the special-shaped column extracted by an edge detection algorithm, and the distribution of the characteristic contour points is divided into several local contour regions according to the geometric characteristics of the special-shaped column. Each region contains one or more contour lines corresponding to arc segments or straight line segments.

[0044] It should be noted that the image acquisition device needs to have high resolution, a wide viewing angle, and the ability to obtain depth information. Commonly selected devices include high-resolution RGB cameras and RGB-D depth cameras. The latter can capture the color information and depth information of an object simultaneously. To completely obtain the contour and surface features of the special-shaped column, the device can collect images from multiple angles around the special-shaped column. This can be achieved by setting up multiple fixed cameras or using a robotic arm for omnidirectional scanning to ensure that the surface features of the column at different angles can be collected.

[0045] The collected image data includes color information, depth information, and the surface texture of the special-shaped column. The depth information is particularly important for accurately identifying the three-dimensional geometric structure of the special-shaped column. The data of the RGB-D camera will output two matrices simultaneously. One is used to represent the color (RGB) of the column, and the other is used to represent the depth information (D) of the column, that is, the distance from the camera to each point on the column surface. If the device uses a laser scanner, the point cloud data of the column surface can be directly obtained. The point cloud data contains higher-precision three-dimensional spatial information, which is convenient for subsequent contour extraction and shape matching.

[0046] Specifically, the core of edge detection is to extract the contour line of an object through the gray-scale change in the image. Canny edge detection extracts the parts with large intensity changes in the image as edges by calculating the gradient information of the image. The specific steps include: converting the image into a grayscale image to simplify the calculation; using the Sobel operator or other derivative operators to calculate the gradients of each pixel point in the x direction and y direction of the image. This gradient represents the gray-scale change rate of the pixel; to make the edge more refined, the Canny algorithm eliminates non-maximum points and retains the pixels with the maximum local gradient to ensure that the final edge is continuous and accurate; to reduce the influence of noise, a double-threshold screening algorithm is adopted to retain the parts with higher gradient values as strong edges and suppress the parts with lower values.

[0047] In the image after edge detection, a large number of contour line segments are usually extracted. However, these line segments cannot be directly used to identify special-shaped columns, so it is necessary to further extract representative feature contour points. According to the geometric characteristics of special-shaped columns, the contour line is divided into several local regions. The contour of each local region may be a straight line segment or an arc line segment. When segmenting, these local contour regions can be calibrated according to the curvature change points in the image (i.e., the turning points or intersection points on the edge). Each feature point is divided according to the local region, initially forming the overall contour of the special-shaped column. Based on the geometric shape of the image, the distribution area of edge feature points is divided into arc segments and straight line segments. At this time, the least squares method can be used to fit the local edge, and the curvature of the curve is judged to divide the straight line and arc regions. The arc region usually shows continuous curvature changes, while the straight line segment corresponds to the part where the curvature is close to zero.

[0048] S2: Based on the geometric feature segmentation method of arc line segments and straight line segments using the initial contour data, disassemble the external shape of the special-shaped column into several combined line segments, and identify the arc line segment and straight line segment features among them through the Hough transform algorithm;

[0049] Among them, the step S2 includes the following steps:

[0050] Based on the initial contour data, select key feature points on the contour by analyzing the local curvature change rate, and classify these feature points as high-curvature points, low-curvature points, and inflection points;

[0051] It should be noted that the feature points are classified as high-curvature points, low-curvature points, and inflection points, specifically as follows:

[0052] High-curvature points: Points with relatively large curvature values, usually located in places with obvious bending, such as the starting and ending points of arc line segments or some corner positions. High-curvature points indicate that the shape of the local area has changed significantly.

[0053] Low-curvature points: Points with curvature values close to zero, which are usually located on straight line segments, indicating that there is almost no bending in the local area.

[0054] Inflection points: Transition points between two different shapes (such as arc line segments and straight line segments). The curvature change of inflection points is relatively drastic, marking an obvious turning of the geometric shape.

[0055] Apply the dynamic window segmentation algorithm, adaptively adjust the size of the segmentation window according to the distribution of feature points, local curvature changes, and geometric shapes, conduct local curvature and direction analysis on the contour line, and initially identify the candidate regions of arc line segments and straight line segments;

[0056] Specifically, after analyzing the local curvature features, the dynamic window segmentation algorithm is used to preliminarily segment the contour line. The core idea of this algorithm is to dynamically adjust the size of the segmentation window according to the curvature change rate and geometric shape features to adapt to different geometric shapes. The adjustment of the window size is based on the degree of curvature change. For a larger curvature (i.e., the arc region), the window size will be reduced to refine the segmentation; while for a smaller curvature (i.e., the straight line region), the window can be appropriately enlarged. For each window region, based on the curvature and direction analysis, candidate regions of arc segments and straight line segments are identified. The arc region is characterized by a stable and non-zero curvature, while the straight line segment is characterized by a curvature close to zero and a consistent direction. By calculating the angle between contour points within each window, the overall direction of the region is determined. If the direction change is small, it indicates that the region is a straight line segment; while if the direction change is large and the curvature is stable, the region is an arc segment.

[0057] In the candidate regions, the weighted Hough transform algorithm is used to extract local line segment features. By calculating the votes of each contour point, the geometric features of arc segments and straight line segments are respectively identified;

[0058] Specifically, for the arc candidate regions, the Hough circle transform is applied. Each contour point votes for possible centers and radii according to its position in the arc region, and the center and radius with the highest number of votes are the identified arc segment features. In the weighted Hough transform, different weights are assigned to each contour point. The setting of the weights is based on the curvature and the position of the point. Points with higher curvature or closer to the center of the circle will be assigned higher weights, thereby improving the recognition accuracy.

[0059] For the straight line segment candidate regions, the standard Hough line transform is used. Each contour point votes for possible line parameters (slope and intercept) according to its gradient direction, and the parameter combination with the highest number of votes is the identified straight line segment. Similarly, a weighting mechanism is applied, and points closer to the center of the line and with more consistent gradient directions are assigned higher weights to ensure more accurate recognition of the straight line segment.

[0060] For the identified arc segments and straight line segments, the least squares method is applied to fit and correct their geometric parameters, including the center, radius, and curvature parameters of the arc segments, as well as the direction and length of the straight line segments. Finally, the optimized geometric parameter features of the arc segments and straight line segments are output.

[0061] S3: Based on the extracted arc and straight line segment features, a geometric model of the special-shaped column is constructed through the graph neural network algorithm. Through the spatial structure relationship of the combination of arcs and straight lines, a topological structure diagram of the feature points is established;

[0062] Among them, the step S3 includes the following steps:

[0063] According to the extracted arc segment and straight line segment features, a geometric feature map of the special-shaped column is constructed. The geometric feature map includes the position information of the line segments, the line segment type, the curvature of the line segments, the angle, and the length.

[0064] Specifically, the geometric feature map is a graph structure composed of nodes and edges. Each node represents a key feature of the special-shaped column contour (such as an arc segment or a straight line segment), and each edge represents the spatial relationship between these features. Each node contains the following geometric feature information:

[0065] Line segment type: Identifies whether the node is an arc segment or a straight line segment.

[0066] Position information: Includes the starting and ending coordinates of the line segment.

[0067] Line segment curvature: For an arc segment, the reciprocal of the radius is used to represent the curvature; for a straight line segment, the curvature is zero.

[0068] Line segment angle: For a straight line segment, the direction is represented by the slope or angle of the line segment; for an arc segment, the angle is represented by the change in the tangent direction of the segment.

[0069] Line segment length: The physical length of the line segment, directly calculated from the starting and ending coordinates.

[0070] Based on the geometric feature map, an adaptive weighting strategy is adopted to encode the connection relationship between the arc segment and the straight line segment in a graph structure. The weight between each edge is dynamically adjusted according to the spatial distance, direction change, and curvature change, forming an initial spatial topological structure.

[0071] Specifically, on the basis of constructing the geometric feature map, it is necessary to connect the spatial relationships between different line segments (nodes) with edges. To describe this connection relationship, an adaptive weighting strategy can be adopted to dynamically adjust the weight of each edge according to the spatial distance, direction change, and curvature change of the line segments.

[0072] Spatial distance: Calculate the spatial distance between each pair of adjacent line segments. The distance between two line segments can be calculated by the Euclidean distance of their endpoint coordinates. The closer the line segments are, the greater the weight of their edge should be, indicating a closer spatial connection relationship between them. A threshold can be set to ignore line segments with a distance exceeding a certain range.

[0073] Direction change: For straight line segments, the direction difference is determined by analyzing their angular changes. Similar directions (i.e., the included angle between two line segments is small) represent that their geometric shapes are more consistent in space and have a higher weight; while line segments with a larger included angle have a lower weight. For arc segments, the direction change is analyzed through the tangent direction of the arc segment. If the difference in the tangent directions of two arc segments is small, it means they are closer in the overall surface structure, and the weight will increase accordingly.

[0074] Curvature change: For arc segments, the change in curvature is also an important feature. Line segments with a small curvature difference should be given a higher weight in the topological structure because they may belong to the same surface structure. For the connection between a straight line segment and an arc segment, the weight is calculated based on the transition curvature between the straight line segment and the arc segment, and a smoother curvature transition connection should be given a higher weight.

[0075] Based on the above adaptive weighting strategy, the weight of each edge is calculated, and finally a complete spatial topological structure of the geometric feature map is formed. The nodes in this topological structure represent line segments, the edges represent the spatial relationships between these line segments, and the weights of the edges reflect their geometric similarity or spatial adjacency.

[0076] Based on the constructed spatial topological structure, a geometric model of the special-shaped column is constructed using the graph neural network algorithm. Through the embedding and propagation of node features and edge features, the graph neural network learns the spatial relationships and topological characteristics between different line segments, gradually optimizing the geometric model of the special-shaped column. The finally output geometric model includes the global shape description and local geometric details of the special-shaped column.

[0077] Specifically, the spatial topological structure constructed in the previous step is used as the input of the graph neural network. The geometric features (such as position, curvature, direction, etc.) of each node are used as input node features, and the weights of the edges (based on distance, direction, and curvature change) are used as edge features in the graph structure.

[0078] The graph neural network first embeds the geometric features of each node, that is, converts its original geometric information into a high-dimensional feature vector. In each layer of the graph neural network, node features are propagated through edges to adjacent nodes. Each node transmits its own features to its neighbor nodes through edges (i.e., connection relationships), and at the same time receives features from neighbor nodes. The weights of the edges are used to adjust the intensity of information propagation. Higher-weight edges will strengthen the feature propagation between two nodes, thus better learning the close relationships between these nodes. As the number of layers of the graph neural network increases, each node will gradually receive more information from neighbor nodes, gradually establishing the association between local geometric features and the overall structure. For the geometric model of a special-shaped column, the local geometric features of the nodes (such as the length and curvature of arc segments or straight segments) are propagated through the graph neural network, gradually establishing connections with the global shape of the entire special-shaped column. Through multi-layer propagation and feature update, the graph neural network can continuously optimize the spatial topological structure of the special-shaped column. The network adjusts the features of each node by learning the relationships between different nodes (line segments), and gradually constructs a complete geometric model. For example, for a complex special-shaped column, the network will learn which arc segments belong to the same continuous surface and which straight segments form the main framework of the column, thereby optimizing the accuracy of the overall model.

[0079] Furthermore, the formula for the geometric model of the special-shaped column is as follows:

[0080]

[0081] where M(υ) represents the geometric model of the special-shaped column, which contains the geometric information of local feature points and their adjacency relationships; h υ represents the geometric feature representation of node v, which is used to describe the local geometric information of a certain point on the column; represents other feature points directly connected to the feature point v on the special-shaped column; c υu and c uw represent the spatial weights between adjacent points, c υu represents the spatial geometric relationship between the feature point v and its neighbor feature point u, such as the distance, curvature difference, direction change, etc. between them, c uw represents the relationship between the feature point u and its neighbor w, further capturing the relationship of distal feature points in the spatial topological structure to help optimize the global description of the geometric model; α, β, γ represent the adjustment parameters in the model, which are used to control the influence weights of different geometric features on the overall model.

[0082] S4: The shape matching algorithm based on graph matching matches the geometric model with a known standard special-shaped column database, and identifies the column type corresponding to the special-shaped line segment combination through curvature matching, length comparison, and spatial distribution analysis;

[0083] Among them, step S4 includes the following steps:

[0084] Based on the geometric model, using the graph matching algorithm, match the topological structure of the geometric model with the models in the standard special-shaped column database by comparing the position of feature points, the curvature of line segments, and the characteristics of spatial distribution;

[0085] Specifically, perform shape matching between the special-shaped column geometric model constructed in step S3 and the models in the standard special-shaped column database. This step uses the graph matching algorithm to compare the topological structure diagram in the geometric model with the topological structure diagram of the standard column model in the database. The specific steps include: The geometric model of each special-shaped column contains nodes (arc and straight line segments) and edges (representing the connection relationship between arc and straight line segments), which constitute the topological graph of the geometric model. The attributes on the nodes include position information, curvature, angle, length, etc. Use the graph isomorphism or subgraph isomorphism algorithm to match the feature points and line segments in the geometric model with the features in the standard database one by one. Through the graph matching algorithm, possible candidate models of the target special-shaped column in the standard database can be found. The standard special-shaped column database contains various types of special-shaped column models to ensure that the most suitable matching result for the target geometric model can be found.

[0086] During the shape matching process, use the multi-scale feature extraction algorithm to analyze the different scale information of the arc line segments and straight line segments in the geometric model;

[0087] Calculate the curvature difference between the matching model and the database model through the curvature matching algorithm, and compare the curvature distribution and curve change trend of the arc line segments;

[0088] Specifically, analyze the curvature difference between the models through the curvature matching algorithm. This step focuses on the curvature distribution and curve change trend of the arc line segments, as follows:

[0089] Curvature distribution analysis: For each arc line segment in the geometric model, first extract its curvature distribution, including the curvature values of the starting point and the ending point and the change of the middle points. Compare it with the corresponding arc line segment in the standard database and calculate the curvature difference.

[0090] Curve change trend comparison: Not only compare the curvature values point by point, but also analyze the curve change trend of the entire arc line segment. For example, whether there is a consistent increasing or decreasing trend of curvature, or whether the curvature of a certain section shows a similar fluctuation pattern.

[0091] Through curvature matching, the standard arc line segment closest to the target special-shaped column can be accurately found to ensure that the matching arc segment has similar bending characteristics to the standard model.

[0092] Based on the line segment length comparison algorithm, analyze the length differences between the corresponding arcs and straight line segments in the identified geometric model and the standard model, and compensate for and adjust the length errors.

[0093] Specifically, use the line segment length comparison algorithm to compare the lengths of arc segments and straight line segments. The specific operations are as follows:

[0094] Arc segment length comparison: The length of an arc segment is jointly determined by curvature and radian. First, calculate the arc length according to the curvature of the arc segment in the geometric model and the angles of the two endpoints, and then compare it with the arc segment in the standard model. If the length difference between the two is large, it may be necessary to adjust the arc length to match.

[0095] Straight line segment length comparison: For straight line segments, directly calculate the length differences between the corresponding straight line segments in the geometric model and the standard model. If there is a large length error, the system will automatically correct the length or mark this segment as a possible matching error.

[0096] Length error compensation: If the length difference is within a certain range, the system will make fine adjustments to the line segment lengths in the geometric model according to the error compensation strategy to make it more consistent with the standard model. The compensation process will maintain the integrity of curvature and angles to ensure that the overall structure of the model is not affected.

[0097] Through spatial distribution analysis, accurately compare the relative position relationships of the arc segments and straight line segments in the geometric model in space, and combine the changes in spatial relative positions to judge the global shape matching degree of the model, and finally determine the recognition result of the special-shaped column.

[0098] Specifically, through spatial distribution analysis of the arc segments and straight line segments in the geometric model, evaluate their relative positions in three-dimensional space, and further judge the global matching degree of the model. The specific operations are as follows:

[0099] Spatial position comparison: For each arc segment and straight line segment, the system calculates their relative positions in three-dimensional space according to their starting points, ending points and direction vectors. Compare with the corresponding segments in the standard model to check their relative distances and position relationships.

[0100] Analysis of relative position changes: Analyze the spatial distribution differences between arc segments and straight line segments. For example, if the distance or angle change between two line segments conforms to the standard model, the matching degree is higher. If the spatial change between adjacent line segments exceeds the preset threshold, the global shape matching degree will be reduced.

[0101] Calculation of global shape matching degree: The system calculates the global matching degree by comprehensively evaluating the spatial distribution of all line segments. The higher the matching degree, the closer the target geometric model is to the standard model in terms of overall shape.

[0102] Furthermore, the formula of the graph matching algorithm is as follows:

[0103]

[0104] where E match represents the matching degree between the geometric model obtained by the graph matching algorithm and the standard special-shaped column database model; υ i and υ' i respectively represent the feature points in the recognized special-shaped column geometric model and the corresponding feature points in the standard special-shaped column database; d(υ i , υ' i ) represents the spatial position difference between the feature point υ i in the geometric model and the feature point υ' i in the database model; Δκ(υ i , υ' i ) represents the curvature difference of the arc segments where the feature points υ i and υ' i are located; Δl(υ i , υ' i ) represents the length difference of the corresponding arc or straight segments in the geometric model and the database model; α 1 , α 2 and α 3 respectively represent the weight parameters of the position difference, curvature difference and length difference; (υ i , υ j ) represents the feature point pair.

[0105] S5: Based on the identified column type of the special-shaped column, use the genetic algorithm to optimize the recognition result and output the final special-shaped column recognition result; when the genetic algorithm is used to optimize the recognition result, by calculating the fitness function of the column type matching result, the optimal column matching result is selected.

[0106] Specifically, in the previous steps, based on the shape matching algorithm, multiple candidate column type matching results have been initially obtained. These candidate matching results are used as the initial population, and each candidate result can be regarded as an "individual", representing a matching solution for a special-shaped column. For each individual, a binary or real number coding method is used to represent the feature combination of the column. For example, the features of a column can include the curvature, length, relative position, etc. of the arc segments and straight segments. The diversity of the initial population is very important to ensure that different individuals have different geometric feature combinations, thereby increasing the exploration ability during the optimization process.

[0107] The fitness function is the standard for evaluating the quality of each individual (matching result). In this method, the definition of the fitness function is closely related to the matching degree of the geometric features of the special-shaped column, and mainly considers the following aspects:

[0108] Curvature matching degree: Calculate the curvature difference between the arc segments of the target special-shaped column and the candidate columns in the database. The smaller the curvature difference, the more consistent the arc features of the two are, and the higher the fitness value. By normalizing the curvature, the curvature matching degree of each individual can be ensured to be comparable.

[0109] Length matching degree: Compare the similarity between the lengths of the identified arc segments and straight segments and the lengths of the line segments of the known standard column. The smaller the length difference, the higher the fitness value. This part can be calculated by the relative error of the length difference.

[0110] Spatial distribution similarity: Evaluate whether the relative positions and distributions of the arc segments and straight segments in three-dimensional space are consistent with the known columns. Quantify the similarity of the spatial distribution by calculating the relative distances and angles between the feature points.

[0111] Overall shape fitting degree: Calculate the fitting degree of the overall contour based on the combined relationship of all arc segments and straight segments. Quantify the shape similarity between the two by superimposing and comparing the overall contour of the target special-shaped column with the overall contour of the standard model.

[0112] Through the selection, crossover, and mutation operations of the genetic algorithm, multiple candidate matching results are generated. Each candidate result is evaluated by the fitness function, and the optimal match is gradually selected. As the number of iterations increases, the genetic algorithm finally outputs the optimal column matching result. The optimal recognition result will include the complete geometric features of the special-shaped column, including the curvature, length, spatial distribution, etc. of its arc segments and straight segments, and match a specific column type in the standard special-shaped column database.

[0113] Output the final recognition result, including the following: the type of the identified special-shaped column; the matching degree between the recognition result and the standard model (such as the matching scores of curvature, length, and topological structure); error analysis (such as spatial distribution error, geometric error of the match); optimization steps and final convergence information of the recognition process.

[0114] In summary, through the high-resolution image acquisition device combined with the denoising algorithm and edge detection algorithm, the present invention can accurately obtain the contour information of the special-shaped column. Especially by introducing the depth information, the three-dimensional geometric structure can be comprehensively displayed, improving the recognition accuracy. By analyzing the local curvature and dynamically adjusting the segmentation window, the straight segments and arc segments can be effectively distinguished, making the local segmentation of complex geometric shapes more accurate. At the same time, combined with the weighted Hough transform algorithm, these feature line segments can be further accurately identified, ensuring the accuracy and reliability of the recognition result.

[0115] The present invention uses a graph neural network to construct a geometric model of a special-shaped column through feature propagation and learning between nodes and edges. This model can comprehensively describe the global shape and local geometric details of the special-shaped column, providing a solid foundation for subsequent shape matching. Through curvature matching, length comparison, and spatial distribution analysis, the system can perform high-precision shape matching on the special-shaped column, ensuring the coincidence degree between the finally identified special-shaped column and the standard column model in the database. Based on the matching results, a genetic algorithm is introduced for further optimization. Through multiple rounds of iteration, the optimal column matching result is selected, improving the accuracy of the recognition result and outputting detailed geometric parameters and error analysis information.

[0116] The above embodiments are only descriptions of the preferred embodiments of the present invention and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for identifying special-shaped columns based on arc segment combination, characterized in that: The following steps are involved: S1: Obtain image data of the special-shaped column through an image acquisition device, perform preprocessing operations, eliminate noise in the image through a denoising algorithm, and use an edge detection algorithm to obtain the initial contour number of the special-shaped column; S2: Based on the initial contour data, the geometric feature segmentation method of arc segments and straight segments is used to decompose the external form of the special-shaped column into several combined segments, and the arc segment and straight segment features are identified through the Hough transform algorithm; S3: Based on the extracted arc and straight line segment features, the geometric model of the special-shaped column is constructed through the graph neural network algorithm, and the topological structure diagram of the feature points is established through the spatial structural relationship of the combination of arcs and straight lines; S4: performing shape matching of the geometric model with a known standard special-shaped column database based on a graph matching shape matching algorithm, and identifying the column type corresponding to the special-shaped line segment combination through curvature matching, length comparison and spatial distribution analysis; S5: Based on the identified column type of the special-shaped column, a genetic algorithm is used to optimize the recognition result, and the final special-shaped column recognition result is output.

2. The method for identifying special-shaped columns based on arc segment combination according to claim 1, characterized in that: The image data includes the outer contour, surface texture, geometric shape information, color information and depth information of the special-shaped column.

3. The method for identifying special-shaped columns based on arc segment combination according to claim 1, characterized in that: The initial contour data includes a plurality of characteristic contour points of the outer contour of the special-shaped column extracted by an edge detection algorithm. The distribution of the characteristic contour points is divided into a plurality of local contour areas according to the geometric features of the special-shaped column. Each area contains one or more contour lines corresponding to arc segments or straight line segments.

4. The method for identifying special-shaped columns based on arc segment combination according to claim 1, characterized in that: The step S2 comprises the following steps: Based on the initial contour data, the key feature points on the contour are selected by analyzing the local curvature change rate, and these feature points are classified into high curvature points, low curvature points and inflection points; The dynamic window segmentation algorithm is applied to adaptively adjust the size of the segmentation window according to the distribution of feature points, local curvature changes and geometric forms, analyze the local curvature and direction of the contour line, and preliminarily identify the candidate areas of arc segments and straight line segments; In the candidate area, the weighted Hough transform algorithm is used to extract the local line segment features, and the geometric features of the arc segment and the straight line segment are identified by voting for each contour point. For the identified arc segments and straight line segments, the least squares method is used to fit and correct their geometric parameters, including the center, radius and curvature parameters of the arc segments and the direction and length of the straight line segments, and finally the optimized arc segment and straight line segment feature parameters are output.

5. The method for identifying special-shaped columns based on arc segment combination according to claim 1, characterized in that: The step S3 comprises the following steps: According to the extracted arc segment and straight line segment features, a geometric feature map of the special-shaped column is constructed, wherein the geometric feature map includes position information, line segment type, curvature, angle and length of the line segment; Based on the geometric feature graph, an adaptive weighting strategy is adopted to encode the connection relationship between arc segments and straight line segments in a graph structure, and the weight between each edge is dynamically adjusted according to the spatial distance, direction change and curvature change to form an initial spatial topological structure; Based on the constructed spatial topological structure, the geometric model of the special-shaped column is constructed using the graph neural network algorithm. By embedding and propagating node features and edge features, the graph neural network learns the spatial relationship and topological characteristics between different line segments, and gradually optimizes the geometric model of the special-shaped column. The final output geometric model contains the global morphological description of the special-shaped column and its local geometric details.

6. The method for identifying special-shaped columns based on arc segment combination according to claim 5, characterized in that: The formula of the geometric model of the special-shaped column is as follows: Where M(v) represents the geometric model of the special-shaped column, which contains the geometric information of the local feature points and their adjacency relationships; h v Represents the geometric feature representation of node v, which is used to describe the local geometric information of a point on the cylinder; Indicates other feature points on the special-shaped column that are directly connected to the feature point v; c vu and c uw represents the spatial weight between adjacent points; α, β, and γ represent the adjustment parameters in the model, which are used to control the influence weight of different geometric features on the overall model.

7. The method for identifying special-shaped columns based on arc segment combination according to claim 1, characterized in that: The step S4 comprises the following steps: Based on the geometric model, the topological structure of the geometric model is matched with the model in the standard special-shaped column database by using the graph matching algorithm, and the position of feature points, line segment curvature and spatial distribution characteristics are compared; In the shape matching process, a multi-scale feature extraction algorithm is used to analyze the different scale information of arc segments and straight line segments in the geometric model; The curvature difference between the matching model and the database model is calculated through the curvature matching algorithm, and the curvature distribution and curve change trend of the arc segment are compared; Based on the line segment length comparison algorithm, the length difference between the corresponding arc and straight line segments in the identified geometric model and the standard model is analyzed to compensate and adjust the length error; Through spatial distribution analysis, the relative position relationship between the arc segments and straight line segments in the geometric model is accurately compared in space. Combined with the changes in the relative positions in space, the global morphological matching degree of the model is judged to ultimately determine the recognition result of the special-shaped column.

8. The method for identifying special-shaped columns based on arc segment combination according to claim 7, characterized in that: The formula of the graph matching algorithm is as follows: Among them, E match Indicates the matching degree between the geometric model obtained by the graph matching algorithm and the standard special-shaped column database model; v i and v′ i They represent the feature points in the identified special-shaped column geometric model and the corresponding feature points in the standard special-shaped column database; d(v i , v′ i ) represents the feature point v in the geometric model i and feature point v′ in the database model i The spatial position difference between i , v′ i ) represents the feature point v i and v′ i The curvature difference of the arc segment; Δl(v i , v′ i ) represents the length difference between the corresponding arc or straight line segments in the geometric model and the database model; α1, α2 and α3 represent the weight parameters of position difference, curvature difference and length difference respectively; (v i , v j ) represents a pair of feature points.

9. The method for identifying special-shaped columns based on arc segment combination according to claim 1, characterized in that: When the genetic algorithm is used to optimize the recognition result, the fitness function of the column type matching result is calculated to screen out the best column matching result.

Images