Bridge crack intelligent identification method and system based on key points

Abstract

The invention discloses an intelligent bridge crack identification method and system based on key points, relates to the technical field of concrete bridge structure detection, and solves the technical problem that an existing bridge crack identification method is difficult to accurately identify in a complex scene. The method comprises the steps that an original crack image is collected to make an original data set, then the original data set is labeled to obtain a crack data set, and the crack data set comprises a crack true value image; training the improved Po lyHRNet network model by using the crack data set to obtain a crack key point detection model, inputting an original crack image into the crack key point detection model, and outputting to obtain a crack thermodynamic diagram containing crack key points; connecting the key points and generating polylines reflecting crack characteristics; according to the invention, corresponding data extraction, data augmentation, loss calculation and other modules are designed for a crack key point identification task, and the identification accuracy of the model facing interference of concrete surface shedding, holes, marking pen marking and the like is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of concrete bridge structure detection, and in particular to a key point-based bridge crack intelligent identification method and system. Background Art

[0002] During the construction and service life of concrete structures, structural cracks may develop due to multiple factors, including construction defects, environmental effects, material degradation, and mechanical overshoots. Cracks can weaken bearing capacity, and penetrating cracks or network-like cracks, in particular, may indicate underlying problems such as structural deformation or foundation settlement. Cracks also serve as pathways for the penetration of corrosive media such as water and chloride ions, accelerating steel corrosion and causing spalling of the concrete cover, further reducing the lifespan of the structure. Therefore, concrete cracks are both a direct manifestation of structural pathology and a core monitoring target for lifecycle management, necessitating regular inspection and assessment of concrete cracks. However, traditional detection methods are limited by technical means and manual operation, are highly subjective, and are prone to missed detections and false detections, with significant shortcomings in efficiency, accuracy, and safety. With the development of intelligent technology, digital detection systems that integrate multi-source data are gradually replacing traditional models, providing more reliable support for the lifecycle management of concrete structures.

[0003] With the rapid development of new technologies like drones and artificial intelligence, bridge inspection is moving towards digitalization and intelligence. Using computer vision technology to intelligently identify bridge photos, locate cracks, and extract information significantly reduces the labor burden and improves accuracy and objectivity compared to traditional manual inspections.

[0004] For example, the Chinese patent "A Bridge Crack Detection Method Based on Image Overlay and Crack Information Fusion" (Patent Application Number: CN 202010886335.3, Publication Number: CN112053331A) uses image overlay and crack information fusion technology, combined with gradient calculation, symbiotic edge extraction, and seed point growth algorithms, to achieve rapid and high-precision bridge crack detection, avoiding the inefficiencies and information loss caused by feature point calculation and sampling in traditional methods.

[0005] However, this patent extracts crack information based on preset parameters such as gradient threshold, co-occurring edge point spacing, and skeleton length. If crack characteristics in actual bridge images exceed these parameters, problems such as seed point misjudgment, skeleton breakage, or incorrect connections will distort detection results. Furthermore, parameter adjustment requires manual intervention, making the method difficult to adapt to complex and changing engineering scenarios.

[0006] For example, the Chinese patent "Method and System for Extracting Single-Pixel Edge Curves from Crack Images and Measuring Crack Width" (Patent Application No.: CN202311238469.4, Publication No.: CN117197109A) proposes a method for extracting single-pixel edge curves from cracks based on grayscale thresholds and row-by-row and column-by-column pixel traversal. This method generates a crack curve by merging row and column edge point sets to improve edge extraction efficiency.

[0007] However, this patented solution extracts crack edge points based on a preset grayscale threshold and a row-by-column pixel traversal rule. This relies on the accuracy of grayscale segmentation and the regularity of crack morphology. If the crack morphology deviates from the preset rule or the image quality is insufficient, the grayscale threshold will fail and the row-by-column traversal will not accurately capture the edge points, resulting in reduced extraction efficiency and insufficient robustness. Summary of the Invention

[0008] In order to solve the problems existing in the above-mentioned prior art, the present invention provides a key point-based intelligent bridge crack recognition method and system to solve the technical problem that the existing bridge crack recognition method is difficult to accurately identify when facing complex scenes.

[0009] A bridge crack intelligent identification method based on key points, comprising:

[0010] Step 1: Collect original crack images to create an original dataset, and then annotate the original dataset to obtain a crack dataset, which includes a crack ground truth map;

[0011] Step 2: Use the crack dataset to train the improved PolyHRNet network model to obtain a crack key point detection model. Input the original crack image into the crack key point detection model, and output a crack heat map containing crack key points.

[0012] Step 3: After obtaining the key points of the crack from the crack thermal map, the crack identification results are obtained by connecting the key points and generating polylines reflecting the crack characteristics through post-processing operations.

[0013] Furthermore, the step 1 includes:

[0014] Step 1.1: Annotate the images in the original dataset. The original dataset is subjected to operations such as horizontal flipping, random scaling, vertical flipping, and rotation to obtain an augmented crack dataset. The annotated polyline segments are used to generate a crack ground truth map. The generated crack ground truth map is used as a new crack dataset. The crack ground truth map includes information about key points and crack connections and is used for subsequent network model training.

[0015] Step 1.2: Divide the crack dataset into a training set and a validation set for subsequent training of the improved PolyHRNet network model.

[0016] Furthermore, the image processing of the crack key point detection model includes:

[0017] Step 2.1: The original image is first passed through the Conv Block convolution module with a downsampling of 4 times, and then passed through the layer1 convolution module for feature extraction;

[0018] Step 2.2: The feature map output in step 2.1 continuously introduces new low-resolution branches in the process from stage 1 to stage 4, compresses the spatial dimension through downsampling operations, and adjusts the number of channels through 1x1 convolution;

[0019] Step 2.3: In each stage, the high-resolution branch passes features to the low-resolution branch through downsampling, and the low-resolution branch passes semantic information back to the high-resolution branch through upsampling. Stage 3 is repeated four times and stage 4 is repeated three times, ultimately obtaining four output layers from head1 to head4.

[0020] Step 2.4: Head1 to head4 are stacked after upsampling, and then pass through the Neck module, and then output the heat map image containing the crack key points through the head5 output layer.

[0021] Furthermore, the loss function for training the improved PolyHRNet network model is the MSELoss function, and its calculation formula is as shown in the formula:

[0022]

[0023] Where, represents the predicted value of the i-th sample at pixel position p, represents the true value of the i-th sample at pixel position p, N represents the total number of key points, K represents the number of key point categories, and p represents the spatial coordinate of the heat map.

[0024] Furthermore, in the MSELoss loss function, all loss calculations are retained for the zero-value area of the true value map, and complete loss calculations are performed for the area where the true value map exceeds 0.8. A loss threshold filtering mechanism of 0.04 is set. The specific calculation formula is as follows:

[0025]

[0026] Furthermore, in step 2.4, the Neck module uses ordinary convolution, and then obtains the predicted heat map through 4 Basic Block residual modules.

[0027] Furthermore, step 2.1 includes: using the maximum pooling method to downsample the ground truth map of the original image size at different ratios to correspond to the output scale of each layer, wherein the pooling kernel size and step size are set to the downsampling ratio, while reducing the size of the feature map, retaining the feature information of the crack key points, and the feature value of each spatial position (i, j) in the feature map Y output after the maximum pooling layer is calculated as shown in the following formula:

[0028] y ij =max(X (i×s:i×s+k,j×s:j×s+k) )

[0029] Where X represents the input feature map, y ij Represents the value of the output feature map at point (i, j), s represents the pooling kernel step size and the feature map downsampling rate, and k represents the pooling kernel size.

[0030] Furthermore, the step 3 includes:

[0031] Step 3.1: Use the DBSCAN-based keypoint clustering algorithm to effectively identify irregular clusters and process noisy data by analyzing the heatmap density distribution. Connect the sparse keypoints generated by the heatmap clustering and match the endpoints of the resulting polylines after the connection is completed.

[0032] Step 3.2: Use the nearest neighbor-first iterative matching algorithm to cyclically calculate the Euclidean distance of each endpoint combination to establish a global distance matrix, and then select the two closest endpoints to connect them to form a continuous polyline segment;

[0033] Step 3.3: Exclude connected endpoints and iteratively update the distance matrix until the topological reconstruction of all key points is completed;

[0034] Step 3.4: Starting from any endpoint of the polyline, perform vector operations on each pair of adjacent edges. The core detection parameters of the algorithm are defined as the angle θ between the direction vectors of the adjacent edges and the angle α between the direction vectors of the intervals. Determine whether the angle θ between the direction vectors of the adjacent edges or the angle α between the direction vectors of the intervals is greater than 90°. If so, it is determined that there is an abnormal connection between the corresponding adjacent edges, and the abnormal connection is adjusted accordingly. Specifically, it includes:

[0035] Step 3.41: Detect the angle between the direction vectors of adjacent edges as θ. When θ>90° is detected, mark the connection as an abnormal crack topology. At the abnormal connection, prioritize retaining the edge segments with shorter geometric lengths and delete redundant long edges.

[0036] Step 3.42: For a composite connection structure consisting of three consecutive edges, wherein the three edges include edge AB, edge BC, and edge CD, the extended angle α between the first and last edges AB and edge CD is calculated. If α>90°, the middle transition edge BC is determined to be a redundant connection, and the edge BC deletion operation is performed at the same time. If α<90°, the middle transition edge BC is determined to be a non-redundant connection.

[0037] Further, step 4 is included:

[0038] The dedicated evaluation metric used is CKS (Crack Keypoint Similarity) to evaluate the crack key point detection model. In CKS, the F1-score value is calculated by the precision and recall rate. The calculation formula of the F1-score value is as follows:

[0039]

[0040] Where, F 1_CKS Expressed as F1-score value, P CKS Expressed as spatial accuracy, R CKS Expressed as spatial recall, where P CKS and R CKS The specific calculation method is as follows:

[0041]

[0042] Where N p Expressed as the number of predicted key points, G(p i ) is expressed as the predicted key point p i The activation value of the real heat map, N g Expressed as the number of predicted key points, P(g i ) is expressed as the predicted key point p i Activation value in the real heat map;

[0043] Different from the normal calculation method of precision and recall, spatial precision uses the predicted key point coordinates to reversely query the response intensity of the real annotation, and uses the true value of the point on the real heat map corresponding to the predicted key point coordinates as the spatial precision. Spatial recall uses the true key point coordinates to reversely query the response intensity in the predicted heat map, and uses the predicted value of the point on the predicted heat map corresponding to the true key point coordinates as the spatial recall.

[0044] A key point-based intelligent bridge crack identification system includes: a data set production module, a model training module, a model evaluation module and an identification module. The data set production module is used to collect original crack images to produce an original data set, and then annotate the original data set to obtain a crack data set, wherein the crack data set includes a crack truth map; the model training module is used to use the crack data set to train a PolyHRNet network model to obtain a crack key point detection model, input the original crack image into the crack key point detection model, and output a crack heat map containing crack key points; the model evaluation module is used to evaluate the crack key point detection model using a special evaluation index CKS; the crack identification module is used to obtain crack key points from the crack heat map, connect the key points through post-processing operations, and generate polylines reflecting crack characteristics.

[0045] The beneficial effects of the present invention include:

[0046] 1. This paper develops a crack key point detection model and designs corresponding data extraction, data augmentation, loss calculation and other modules for the crack key point identification task, further improving the model's recognition accuracy when facing interference such as concrete surface peeling, holes, and marker pen marks.

[0047] 2. The improved PolyHRNet model, used as a crack keypoint detection model, has only 9.64M parameters, compared to 28M for the U-net network commonly used for crack semantic segmentation. In terms of the CKS evaluation metric, PolyHRNet achieved an accuracy of 0.81, a recall of 0.94, and a final F1 value of 0.87. This model can achieve 37fps on an RTX2080s graphics card, while other lightweight segmentation models generally have inference speeds below 10fps on embedded devices. This model achieves both high precision and real-time performance for crack detection while being lightweight.

[0048] 3. The present invention uses a crack key point connection optimization method that combines heat maps with key point clustering. This method adopts a dual-modal clustering strategy and adds an angle constraint mechanism to effectively solve the path confusion problem of adjacent or intersecting cracks. By simulating the physical laws of material fracture to design algorithm constraints, a connection method that is more in line with the actual crack expansion law is achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 This is a flow chart of a key point-based intelligent bridge crack identification method involved in an embodiment of the present application.

[0050] Figure 2 This is a schematic diagram of the PolyHRNet network structure involved in the embodiments of this application.

[0051] Figure 3 This is a schematic diagram of the Neck module involved in an embodiment of the present application.

[0052] Figure 4 This is an effect diagram of the original crack image involved in the embodiment of the present application after the crack key point detection model outputs the crack key point.

[0053] Figure 5 This is a schematic diagram of the endpoint connection involved in the embodiment of the present application.

[0054] Figure 6 This is Example 1 of the angle correction involved in the embodiment of the present application.

[0055] Figure 7 FIG2 is an example of angle correction according to an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of this application.

[0057] like Figure 1 As shown, a bridge crack intelligent identification method based on key points includes:

[0058] Step 1: Collect original crack images to create an original dataset, and then annotate the original dataset to obtain a crack dataset, which includes a crack ground truth map;

[0059] Step 2: Use the crack dataset to train the improved PolyHRNet network model to obtain a crack key point detection model. Input the original crack image into the crack key point detection model and output a crack thermal map containing crack key points. The structure of the PolyHRNet network model is as follows: Figure 2 As shown;

[0060] Step 3: After obtaining the key points of the crack from the crack thermal map, the crack identification results are obtained by connecting the key points and generating polylines reflecting the crack characteristics through post-processing operations.

[0061] In another embodiment, step 1 includes:

[0062] Step 1.1: Annotate the images in the original dataset. The original dataset is subjected to operations such as horizontal flipping, random scaling, vertical flipping, and rotation to obtain an augmented crack dataset. The annotated polyline segments are used to generate a crack ground truth map. The generated crack ground truth map is used as a new crack dataset. The crack ground truth map includes information about key points and crack connections and is used for subsequent network model training.

[0063] Step 1.2: Divide the crack dataset into a training set and a validation set for subsequent training of the improved PolyHRNet network model.

[0064] In another embodiment, the image processing by the crack key point detection model includes:

[0065] Step 2.1: The original image is first passed through the Conv Block convolution module with a downsampling of 4 times, and then passed through the layer1 convolution module for feature extraction;

[0066] Step 2.2: The feature map output in step 2.1 continuously introduces new low-resolution branches in the process from stage 1 to stage 4, compresses the spatial dimension through downsampling operations, and adjusts the number of channels through 1x1 convolution;

[0067] Step 2.3: In each stage, the high-resolution branch passes features to the low-resolution branch through downsampling, and the low-resolution branch passes semantic information back to the high-resolution branch through upsampling. Stage 3 is repeated four times and stage 4 is repeated three times, ultimately obtaining four output layers from head1 to head4.

[0068] Step 2.4: After upsampling, head1 to head4 are stacked, and then passed through the Neck module, and the heat map image containing the crack key points is output through the head5 output layer. The output effect is as follows Figure 4 shown.

[0069] The loss function for training the improved PolyHRNet network model is the MSELoss function, and its calculation formula is as shown in the formula:

[0070]

[0071] Where, represents the predicted value of the i-th sample at pixel position p, represents the true value of the i-th sample at pixel position p, N represents the total number of key points, K represents the number of key point categories, and p represents the spatial coordinate of the heat map.

[0072] In the MSELoss loss function, all loss calculations are retained for the zero-value area of the true value map, and full loss calculations are performed for the area where the true value map exceeds 0.8. A loss threshold filtering mechanism of 0.04 is set. The specific calculation formula is as follows:

[0073]

[0074] In step 2.4, the Neck module uses ordinary convolution, and then obtains the predicted heat map through 4 Basic Block residual modules. The specific structure of the Neck module is as follows: Figure 3 shown.

[0075] Step 2.1 includes: using the maximum pooling method to downsample the ground truth map of the original image size at different ratios to correspond to the output scale of each layer, where the pooling kernel size and step size are set to the downsampling ratio, while reducing the size of the feature map, while retaining the feature information of the crack key points. The feature value of each spatial position (i, j) in the feature map Y output after the maximum pooling layer is calculated as shown in the following formula:

[0076] y ij =max(X (i×s:i×s+k,j×s:j×s+k) )

[0077] Where X represents the input feature map, y ij Represents the value of the output feature map at point (i, j), s represents the pooling kernel step size and the feature map downsampling rate, and k represents the pooling kernel size.

[0078] The improved PolyHRNet model, used as a crack keypoint detection model, has only 9.64MB of parameters, compared to 28MB for the U-Net network commonly used for crack semantic segmentation. Using the CKS evaluation metric, PolyHRNet achieved an accuracy of 0.81, a recall of 0.94, and a final F1 score of 0.87. This model can achieve 37fps on an RTX2080s graphics card, while other lightweight segmentation models generally experience inference speeds below 10fps on embedded devices. This model achieves both high precision and real-time performance for crack detection while being lightweight. Specifically, the 9.64MB parameter memory usage in the model is directly output from the code, demonstrating the model's lightweight nature. The accuracy, recall, and F1 scores are calculated using a formula specifically designed for the CKS evaluation metric. The final F1 score demonstrates high precision. The 37fps frame rate indicates 37 images processed per second, also directly output from the network code, demonstrating real-time performance.

[0079] In another embodiment, step 3 includes:

[0080] Step 3.1: Use the DBSCAN-based keypoint clustering algorithm to effectively identify irregular clusters and process noisy data by analyzing the heatmap density distribution. Connect the sparse keypoints generated by the heatmap clustering and match the endpoints of the resulting polylines after the connection is completed.

[0081] Step 3.2: Use the nearest neighbor-first iterative matching algorithm to cyclically calculate the Euclidean distance of each endpoint combination to establish a global distance matrix, and then select the two closest endpoints to connect them to form a continuous polyline segment;

[0082] Step 3.3: Exclude connected endpoints and iteratively update the distance matrix until the topological reconstruction of all key points is completed;

[0083] Step 3.4: Starting from any endpoint of the polyline, perform vector operations on each pair of adjacent edges. The core detection parameters of the algorithm are defined as the angle θ between the direction vectors of the adjacent edges and the angle α between the direction vectors of the intervals. Determine whether the angle θ between the direction vectors of the adjacent edges or the angle α between the direction vectors of the intervals is greater than 90°. If so, it is determined that there is an abnormal connection between the corresponding adjacent edges, and the abnormal connection is adjusted accordingly. Specifically, it includes:

[0084] Step 3.41: Detect the angle between the direction vectors of adjacent edges as θ. When θ>90° is detected, mark the connection as an abnormal crack topology, such as Figure 5 The angle θ formed by the BC edge and the CD edge is shown. At the abnormal connection, the edge segments with shorter geometric lengths are preferentially retained while the redundant long edges are deleted.

[0085] Step 3.42: For a composite connection structure consisting of three consecutive edges, the three edges include edge AB, edge BC, and edge CD connected in sequence, calculate the extension angle α between the first and last edges AB and edge CD. If α>90°, the middle transition edge BC is determined to be a redundant connection, and the edge BC deletion operation is performed at the same time. Specifically, Figure 6 As shown, if α<90°, the middle transition edge BC is determined to be non-redundantly connected, as shown in Figure 7 shown.

[0086] In another embodiment, step 4 is included:

[0087] The dedicated evaluation metric used is CKS (Crack Keypoint Similarity) to evaluate the crack key point detection model. In CKS, the F1-score value is calculated by the precision and recall rate. The calculation formula of the F1-score value is as follows:

[0088]

[0089] Where, F 1_CKSExpressed as F1-score value, P CKS Expressed as spatial accuracy, R CKS Expressed as spatial recall, where P CKS and R CKS The specific calculation method is as follows:

[0090]

[0091] Where N p Expressed as the number of predicted key points, G(p i ) is expressed as the predicted key point p i The activation value of the real heat map, N g Expressed as the number of predicted key points, P(g i ) is expressed as the predicted key point p i Activation value in the real heat map;

[0092] Different from the normal calculation method of precision and recall, spatial precision uses the predicted key point coordinates to reversely query the response intensity of the real annotation, and uses the true value of the point on the real heat map corresponding to the predicted key point coordinates as the spatial precision. Spatial recall uses the true key point coordinates to reversely query the response intensity in the predicted heat map, and uses the predicted value of the point on the predicted heat map corresponding to the true key point coordinates as the spatial recall.

[0093] In another embodiment, a key point-based intelligent bridge crack identification system is provided, comprising: a data set production module, a model training module, a model evaluation module and an identification module, wherein the data set production module is used to collect original crack images to produce an original data set, and then annotate the original data set to obtain a crack data set, wherein the crack data set includes a crack true value map; the model training module is used to use the crack data set to train an improved PolyHRNet network model to obtain a crack key point detection model, input the original crack image into the crack key point detection model, and output a crack heat map containing crack key points; the model evaluation module is used to evaluate the crack key point detection model using a special evaluation index CKS, and the crack identification module is used to obtain crack key points from the crack heat map, connect the key points through post-processing operations, and generate polylines reflecting crack characteristics.

[0094] The above-described embodiments merely represent specific implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of protection of the present application. It should be noted that a person skilled in the art would be able to make numerous variations and improvements without departing from the technical concept of the present application, and all such variations and improvements fall within the scope of protection of the present application.

Claims

1. A bridge crack intelligent identification method based on key points, characterized in that: The following steps are involved: Step 1: Collect original crack images to create an original dataset, and then annotate the original dataset to obtain a crack dataset, which includes a crack ground truth map; Step 2: Use the crack dataset to train the improved PolyHRNet network model to obtain a crack key point detection model. Input the original crack image into the crack key point detection model, and output a crack heat map containing crack key points. Step 3: After obtaining the key points of the crack from the crack thermal map, the crack identification results are obtained by connecting the key points and generating polylines reflecting the crack characteristics through post-processing operations.

2. The bridge crack intelligent identification method based on key points according to claim 1 is characterized in that: The step 1 comprises: Step 1.1: Annotate the images in the original dataset. The original dataset is subjected to operations such as horizontal flipping, random scaling, vertical flipping, and rotation to obtain an augmented crack dataset. The annotated polyline segments are used to generate a crack ground truth map. The generated crack ground truth map is used as a new crack dataset. The crack ground truth map includes information about key points and crack connections and is used for subsequent network model training. Step 1.2: Divide the crack dataset into a training set and a validation set for subsequent training of the improved PolyHRNet network model.

3. The bridge crack intelligent identification method based on key points according to claim 1 is characterized in that: The image processing of the crack key point detection model includes: Step 2.1: The original image is first passed through the Conv Block convolution module with a downsampling of 4 times, and then passed through the layer1 convolution module for feature extraction; Step 2.2: The feature map output in step 2.1 continuously introduces new low-resolution branches in the process from stage 1 to stage 4, compresses the spatial dimension through downsampling operations, and adjusts the number of channels through 1x1 convolution; Step 2.3: In each stage, the high-resolution branch passes features to the low-resolution branch through downsampling, and the low-resolution branch passes semantic information back to the high-resolution branch through upsampling. Stage 3 is repeated four times and stage 4 is repeated three times, ultimately obtaining four output layers from head1 to head4. Step 2.4: Head1 to head4 are stacked after upsampling, and then pass through the Neck module, and then output the heat map image containing the crack key points through the head5 output layer.

4. The bridge crack intelligent identification method based on key points according to claim 1 is characterized in that: The loss function for training the improved PolyHRNet network model is the MSELoss function, and its calculation formula is as shown in the formula: Where, represents the predicted value of the i-th sample at pixel position p, represents the true value of the i-th sample at pixel position p, N represents the total number of key points, K represents the number of key point categories, and p represents the spatial coordinate of the heat map.

5. The bridge crack intelligent identification method based on key points according to claim 4 is characterized in that: In the MSELoss loss function, all loss calculations are retained for the zero-value area of the true value map, and full loss calculations are performed for the area where the true value map exceeds 0.

8. A loss threshold filtering mechanism of 0.04 is set. The specific calculation formula is as follows:

6. The bridge crack intelligent identification method based on key points according to claim 3 is characterized in that: In step 2.4, the Neck module uses ordinary convolution, and then obtains the predicted heat map through 4 Basic Block residual modules.

7. The bridge crack intelligent identification method based on key points according to claim 3 is characterized in that: Step 2.1 includes: using the maximum pooling method to downsample the ground truth map of the original image size at different ratios to correspond to the output scale of each layer, where the pooling kernel size and step size are set to the downsampling ratio, while reducing the size of the feature map, while retaining the feature information of the crack key points. The feature value of each spatial position (i, j) in the feature map Y output after the maximum pooling layer is calculated as shown in the following formula: and ij =max(X (i×s:i×s+k,j×s:j×s+k) ) Where X represents the input feature map, y ij Represents the value of the output feature map at point (i, j), s represents the pooling kernel step size and the feature map downsampling rate, and k represents the pooling kernel size.

8. The bridge crack intelligent identification method based on key points according to claim 1 is characterized in that: The step 3 includes: Step 3.1: Use the DBSCAN-based keypoint clustering algorithm to effectively identify irregular clusters and process noisy data by analyzing the heatmap density distribution. Connect the sparse keypoints generated by the heatmap clustering and match the endpoints of the resulting polylines after the connection is completed. Step 3.2: Use the nearest neighbor-first iterative matching algorithm to cyclically calculate the Euclidean distance of each endpoint combination to establish a global distance matrix, and then select the two closest endpoints to connect them to form a continuous polyline segment; Step 3.3: Exclude connected endpoints and iteratively update the distance matrix until the topological reconstruction of all key points is completed; Step 3.4: Starting from any endpoint of the polyline, perform vector operations on each pair of adjacent edges. The core detection parameters of the algorithm are defined as the angle θ between the direction vectors of adjacent edges and the angle α between the direction vectors of the intervals. Determine whether the angle θ between the direction vectors of adjacent edges or the angle α between the direction vectors of the intervals is greater than 90°. If so, it is determined that there is an abnormal connection between the corresponding adjacent edges, and the abnormal connection is adjusted accordingly.

9. The bridge crack intelligent identification method based on key points according to claim 1 is characterized in that: Includes step 4: The dedicated evaluation metric used is CKS (Crack Keypoint Similarity) to evaluate the crack key point detection model. In CKS, the F1-score value is calculated by the precision and recall rate. The calculation formula of the F1-score value is as follows: Where, F 1_CKS Expressed as F1-score value, P CKS Expressed as spatial accuracy, R CKS Expressed as spatial recall, where P CKS and R CKS The specific calculation method is as follows: Where N p Expressed as the number of predicted key points, G(p i ) is expressed as the predicted key point p i The activation value of the real heat map, N g Expressed as the number of predicted key points, P(g i ) is expressed as the predicted key point p i Activation values in the true heatmap.

10. A bridge crack intelligent identification system based on key points, characterized in that: include: A data set production module, a model training module, a model evaluation module and an identification module. The data set production module is used to collect original crack images to produce original data sets, and then annotate the original data sets to obtain crack data sets, and the crack data sets include crack true value maps; the model training module is used to use the crack data sets to train the improved PolyHRNet network model to obtain a crack key point detection model, input the original crack images into the crack key point detection model, and output a crack thermal map containing crack key points; the model evaluation module is used to evaluate the crack key point detection model using a special evaluation index CKS, and the crack identification module is used to obtain crack key points from the crack thermal map, and then connect the key points through post-processing operations to generate polylines reflecting crack characteristics.

Images