Bridge concrete surface crack identification method and system based on wall-climbing robot

Abstract

The application provides a bridge concrete surface crack identification method and system based on a wall-climbing robot, and relates to the technical field of concrete crack detection. The method comprises the following steps: constructing a target detection and semantic segmentation integrated model to identify the concrete surface cracks of a target bridge; obtaining a crack data set according to the results of target detection and semantic segmentation integration, and labeling the crack data set; using the labeled crack data set for crack identification model training and verification to obtain an end-to-end real-time crack identification model; performing mobile terminal quantization deployment on the trained crack identification model; designing a wall-climbing robot suitable for crawling on the surface of bridge concrete; establishing a remote connection between the wall-climbing robot and the mobile terminal; and displaying the bridge crack identification results in real time through the mobile terminal. The application can not only accurately identify the surface cracks of a bridge in real time in an end-to-end manner, but also save time and cost and reduce the safety risks of personnel detection.

Description

Technical Field

[0001] This invention relates to the field of concrete crack detection technology, and in particular to a method and system for identifying cracks on the surface of bridge concrete based on a wall-climbing robot. Background Technology

[0002] Currently, my country has a vast number of concrete beam bridges in service, some of which are in or nearing the end of their service life. Under external loads, cracks inevitably appear in these concrete bridges. Currently, bridge crack detection mainly relies on bridge inspection vehicles as platforms, with manual visual inspection as the primary method, supplemented by tools such as measuring tapes and crack width gauges when necessary. However, this method is susceptible to the subjective bias of engineers, is time-consuming, labor-intensive, and carries a high risk, and also suffers from drawbacks such as inaccurate measurements and incomplete data recording.

[0003] With advancements in intelligent algorithms and computing power, deep learning models have become an effective method for identifying cracks on structural surfaces, attracting widespread attention and research. In summary, crack identification algorithms can be broadly categorized into three types: image classification, target detection, and image segmentation. Currently, some researchers have integrated crack identification algorithms into devices such as drones to achieve real-time crack identification on bridges. However, because drones cannot get close enough to the concrete surface of bridges, and image quality is often poor due to camera shake, the crack identification results are unsatisfactory. Therefore, finding a high-precision and high-efficiency intelligent method for identifying cracks on bridge concrete surfaces is essential. Summary of the Invention

[0004] To address the problems in existing technologies where drones cannot get close to the concrete surface of bridges, and where there is shaking during shooting, resulting in poor image quality and unsatisfactory crack recognition, this invention proposes an intelligent method and system for identifying cracks on the concrete surface of bridges based on a wall-climbing robot.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] On the one hand, a method for identifying cracks on the surface of bridge concrete based on a wall-climbing robot is provided. This method is applied to electronic devices and includes the following steps:

[0007] S1: Construct an integrated model of object detection and semantic segmentation to identify cracks on the concrete surface of the target bridge;

[0008] S2: According to the data format requirements of the object detection and semantic segmentation integrated model, obtain the crack dataset, label the crack dataset, use the labeled crack dataset for model training and validation, and obtain an end-to-end real-time crack recognition model.

[0009] S3: Quantize and deploy the trained crack recognition model on mobile devices;

[0010] S4: A wall-climbing robot suitable for bridge concrete surfaces is designed using a dual adsorption structure of negative pressure suction cup and electromagnet. The wall-climbing robot is remotely connected to a mobile terminal. The wall-climbing robot is placed at a designated position on the bottom or side of the bridge concrete beam. By operating the mobile terminal, the crack identification result corresponding to a specific location on the bridge is displayed in real time, thus completing the intelligent identification of cracks on the bridge concrete surface based on the wall-climbing robot.

[0011] Optionally, in step S1, an integrated model for object detection and semantic segmentation is constructed to identify cracks on the concrete surface of the target bridge, including:

[0012] Based on the YOLOv5 object detection algorithm, a lightweight segmentation network is added to construct an integrated model of object detection and semantic segmentation, which outputs both crack object detection results and crack segmentation results. The crack object detection results and crack segmentation results of the concrete surface of the target bridge are verified, and a binary image of the crack is output.

[0013] Optionally, the crack detection results on the concrete surface of the target bridge are verified with the crack segmentation results, and a binarized image of the crack is output, including:

[0014] The crack target detection result is a crack prediction and location map, which includes the coordinates of the center point of the outer rectangle of the crack and the area of ​​the rectangle; the crack segmentation result is the connected domain of the predicted crack, which includes the coordinates of the center point of the outer rectangle of each connected domain and the area of ​​the connected domain.

[0015] When the coordinates of the center point of the bounding box and the area of ​​the bounding box of the crack target detection result are close to those of the crack segmentation result, the connected component is determined to be a crack; otherwise, it is noise. The connected components determined to be noise are deleted, and only the connected components containing cracks are retained. The crack binarized image is then output.

[0016] Optionally, in step S2, the crack dataset is labeled, including:

[0017] The crack dataset was labeled, and the labeled results were divided into two categories: one is YOLO data format for object detection; the other is binarized mask images for semantic segmentation.

[0018] Optionally, the labeled dataset can be used for training and validation of the crack recognition model, wherein the ratio of training data to validation data in the crack dataset is 8:2.

[0019] Optionally, the wall-climbing robot is a wheeled robot that crawls on the bottom and sides of the bridge's concrete beams using a dual adsorption structure of negative pressure suction cups and electromagnets.

[0020] Optionally, a CCD camera is embedded in the bottom of the wall-climbing robot, which can be remotely connected to a mobile device via Wi-Fi or a SIM card.

[0021] On the one hand, a bridge concrete surface crack recognition system based on a wall-climbing robot is provided. This system is applied to electronic devices and includes:

[0022] The model integration module is used to build an integrated model of object detection and semantic segmentation to identify cracks on the concrete surface of the target bridge.

[0023] The model training module is used to obtain the crack dataset according to the data format requirements of the object detection and semantic segmentation integrated model, and to annotate the crack dataset. The annotated crack dataset is used for model training and validation to obtain an end-to-end real-time crack recognition model.

[0024] The mobile deployment module is used to quantitatively deploy the trained crack recognition model on mobile devices.

[0025] The device is equipped with an identification module and uses a dual adsorption structure of negative pressure suction cup and electromagnet to design a wall-climbing robot suitable for bridge concrete surfaces. The wall-climbing robot is remotely connected to a mobile terminal. The wall-climbing robot is placed at a designated position on the bottom or side of the bridge concrete beam. By operating the mobile terminal, the crack identification result corresponding to a specific location on the bridge is displayed in real time, thus completing the intelligent identification of cracks on the bridge concrete surface based on the wall-climbing robot.

[0026] Optionally, the model training module is further used to add a lightweight segmentation network to the YOLOv5 object detection algorithm, construct an integrated model of object detection and semantic segmentation, and output crack object detection results and crack segmentation results; verify the crack object detection results and crack segmentation results of the concrete surface of the target bridge, and output a binarized image of the crack.

[0027] On the one hand, an electronic device is provided, comprising a processor and a memory, wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-mentioned method for identifying cracks on the concrete surface of a bridge based on a wall-climbing robot.

[0028] On the one hand, a computer-readable storage medium is provided, wherein at least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the above-mentioned method for identifying cracks on the surface of bridge concrete based on a wall-climbing robot.

[0029] The above-described technical solutions of the embodiments of the present invention have at least the following beneficial effects:

[0030] In the above-mentioned solution, the present invention provides an intelligent identification method for cracks on the concrete surface of bridges based on a wall-climbing robot. This method can not only identify cracks on the bridge surface in real time with high accuracy, but also save time and costs and reduce the safety risks of personnel inspection. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of a method for intelligent identification of cracks on the surface of bridge concrete based on a wall-climbing robot, provided by an embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram of the dataset annotation for a method for intelligent identification of cracks on the surface of bridge concrete based on a wall-climbing robot, provided in an embodiment of the present invention.

[0034] Figure 3 This is a structural diagram of a wall-climbing robot, which is provided in an embodiment of the present invention for a method of intelligent identification of cracks on the surface of bridge concrete based on a wall-climbing robot.

[0035] Figure 4 This is a block diagram of an intelligent identification system for cracks on the surface of bridge concrete based on a wall-climbing robot, provided in an embodiment of the present invention.

[0036] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0037] The attached figures are labeled as follows:

[0038] 1. Base plate; 2. Top plate; 3. Negative pressure suction cup with pneumatic pipe; 4. Large negative pressure suction cup; 5. Electromagnet; 6. Electric slide rail; 7. Components with embedded control system; 8. Battery pack; 9. Switch button; 10. Four-wheeled vehicle supporting the base; 11. Roller with displacement sensor; 12. CCD camera; 13. Panoramic camera. Detailed Implementation

[0039] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0040] This invention provides a method for intelligent identification of cracks on the surface of bridge concrete based on a wall-climbing robot. This method can be implemented by an electronic device, which can be a terminal or a server. Figure 1 The flowchart shown is for an intelligent identification method of cracks on the surface of bridge concrete based on a wall-climbing robot. The processing flow of this method may include the following steps:

[0041] S101: Construct an integrated model of object detection and semantic segmentation to identify cracks on the concrete surface of the target bridge;

[0042] In one feasible implementation, the present invention adds a lightweight segmentation algorithm to the YOLOv5 target detection algorithm, and outputs crack target detection results and crack segmentation results simultaneously. The crack target detection results and crack segmentation results of the concrete surface of the target bridge are verified, and a binary image of the crack is output.

[0043] In this embodiment of the invention, after the YOLOv5 (P6.2 version) target detection model backbone network extracts crack features, a lightweight bilateral segmentation network branch is introduced. To improve the crack edge segmentation effect, an attention mechanism module and dilated convolution operation are added, which can simultaneously output the crack target detection result and pixel-level segmentation result.

[0044] In one feasible implementation, the target detection results and crack segmentation results of the concrete surface cracks of the target bridge are verified to eliminate errors, and a binarized image of the cracks is output, including:

[0045] The crack target detection result is a crack prediction and location map, which includes the coordinates of the center point of the outer rectangle of the crack and the area of ​​the rectangle; the crack segmentation result is the connected domain of the predicted crack, which includes the coordinates of the center point of the outer rectangle of each connected domain and the area of ​​the connected domain.

[0046] When the coordinates of the center point of the bounding box and the area of ​​the bounding box of the crack target detection result are close to those of the crack segmentation result, the connected component is determined to be a crack; otherwise, it is noise. The connected components determined to be noise are deleted, and only the connected components containing cracks are retained. The crack binarized image is then output.

[0047] In one feasible implementation, the threshold for determining whether the coordinates of the center point of the rectangle and the area of ​​the rectangle of the crack target detection result and the crack segmentation result are close can be set according to the accuracy requirements.

[0048] In this embodiment of the invention, the binarized image obtained by eliminating errors after verification has higher accuracy.

[0049] S102: According to the data format requirements of the object detection and semantic segmentation integrated model, obtain the crack dataset, label the crack dataset, use the labeled crack dataset for model training and verification, and obtain an end-to-end real-time crack recognition model.

[0050] In one feasible implementation, the crack dataset is labeled, including:

[0051] The labeling results of the crack dataset are divided into two categories: one is YOLO data format used for object detection; the other is binarized mask images used for semantic segmentation.

[0052] One feasible implementation method is, for example Figure 2 As shown, the dataset annotation used for model training in this invention includes:

[0053] (1) Add the dataset to the annotation software.

[0054] (2) The cracks in the image are labeled with rectangular boxes to generate VOC format data, and the VOC format data is converted into YOLO normalized data format for training and validation of the model object detection.

[0055] (3) Polygon annotation is performed on the cracks in the image to generate crack mask images, which are used for training and verification of the semantic segmentation model.

[0056] In one feasible implementation, the labeled dataset is used for training and validation of the crack recognition model, wherein the ratio of training data to validation data in the crack dataset is 8:2, the number of training epochs is 400, and the optimizer is the stochastic gradient descent algorithm with momentum.

[0057] S103: Quantize and deploy the trained crack recognition model on mobile devices;

[0058] S104: A wall-climbing robot suitable for bridge concrete surfaces is designed using a dual adsorption structure of negative pressure suction cup and electromagnet. The wall-climbing robot is remotely connected to a mobile terminal. The wall-climbing robot is placed at a designated position on the bottom or side of the bridge concrete beam. By operating the mobile terminal, the crack identification result corresponding to a specific location on the bridge is displayed in real time, thus completing the intelligent identification of cracks on the bridge concrete surface based on the wall-climbing robot.

[0059] In one feasible implementation, the wall-climbing robot is a wheeled robot that crawls on the bottom and sides of the concrete beam of the bridge through a dual adsorption structure of negative pressure suction cups and electromagnets.

[0060] One feasible implementation method is, for example Figure 3 As shown, the wall-climbing robot device used in this invention includes: a base plate 1, a top plate 2, a negative pressure suction cup with pneumatic pipes 3, a large negative pressure suction cup 4, an electromagnet 5, an electric slide rail 6, components embedded with a control system 7, a battery pack 8, a switch button 9, a four-wheeled vehicle supporting the base 10, rollers with displacement sensors 11, a CCD camera 12, and a panoramic camera 13.

[0061] The base plate 1 is equipped with a negative pressure suction cup 3 with pneumatic pipes, a large negative pressure suction cup 4, an electromagnet 5, a CCD camera 12, etc. The bottom of the base plate 1 is supported on the bearings of the four-wheel vehicle 10, and the top is connected to the top plate 2 via an electric slide rail 6. Negative pressure suction cups 3 with pneumatic pipes are installed at the four corners, and a large negative pressure suction cup 4 with a rotating function is installed in the middle. When the base plate 1 is fixed in place by the negative pressure suction cups 3, the top plate 2 can slide forward via the electric slide rail 6; when the top plate 2 is fixed in place by the negative pressure suction cups 3, the base plate 1 can slide forward via the electric slide rail 6.

[0062] The top plate base 2 includes a negative pressure suction cup 3 with pneumatic pipes, an electromagnet 5, an electric slide rail 6, components with embedded control system 7, a battery pack 8, a switch button 9, a panoramic camera 13, etc. Negative pressure suction cups 3 with pneumatic pipes are installed at the four corners of the top plate base 2.

[0063] The negative pressure suction cup 3, equipped with a pneumatic tube, has two states: on and off. When the power is on, the air compressor connected in series with the pneumatic tube operates. When the suction cup automatically extends and presses tightly against the concrete surface, a vacuum is drawn out inside the suction cup, and the suction cup is in an adsorption state. When the power is off, the suction cup automatically releases and retracts to its original position.

[0064] The working principle of the rotating negative pressure suction cup 4 is the same as that of the negative pressure suction cup 3 with pneumatic tubing. This suction cup is only used when changing the forward direction and can rotate 360 ​​degrees. When it is necessary to change the forward direction, the power supply to the suction cup 4 is turned on, and the suction cup 4 automatically extends and is in the suction state. At this time, it can be rotated to any direction using the rotation function. When the power supply to the suction cup 3 is turned off, the suction cup 3 is in the release state and automatically retracts to its original position.

[0065] Electromagnet 5, because the protective layer of bridge concrete beams is generally no more than 50mm, and the reinforcement bars at the bottom and sides of the beam are densely arranged, can generate electromagnetic attraction with the reinforcement bars in the concrete when the electromagnet 5 is turned on, and can bear part of the weight of the device to a certain extent.

[0066] Electric slide rail 6 is a type of electric sliding device.

[0067] The component 7 has an embedded control system, and the control system includes all control subsystems, such as suction cup 3, suction cup 4, electromagnet 5, electric slide rail 6, roller with displacement sensor 11, CCD camera 12, panoramic camera 13, etc.

[0068] Battery pack 8 provides power to the device.

[0069] Switch button 9, the start and stop button for the device.

[0070] The four-wheeled vehicle 10 supporting the base is able to support the entire device.

[0071] The roller 11, equipped with a displacement sensor, can record the distance the wall-climbing robot travels, thus revealing the exact location of the identified cracks on the bridge.

[0072] Two CCD cameras (12 in total) are installed on either side of the bottom of the base 1 to expand the recognition range. The main optical axis of the camera lens should be perpendicular to the surface of the object being measured. Since the height of the four-wheeled vehicle 10 supporting the base remains constant, the object distance of the camera remains constant. Based on the known parameters inside the CCD camera, the pixel units of the captured image can be converted into standard units such as millimeters.

[0073] Panoramic camera 13 is a 360-degree panoramic camera. This camera is mounted on top of the base 2 and can monitor the surroundings of the wall-climbing robot in real time, especially obstacles.

[0074] In one feasible implementation, the present invention designs a suitable robot for crawling on the concrete surface of bridges, which can be remotely connected to a mobile device via Wi-Fi or a SIM card. The wall-climbing robot employs a dual adsorption structure of negative pressure suction cups and electromagnets. When the negative pressure suction cup control subsystems at the four corners of the base plate are activated, the suction cups automatically extend to just contact the concrete surface and adhere, while the top plate slides forward via an electric rail. When it reaches a fixed position, the negative pressure suction cup control subsystems at the four corners of the top plate activate, and the suction cups automatically extend to just contact the concrete surface and adhere. Then, the negative pressure suction cups at the four corners of the base plate close and automatically retract to their original positions, and the base plate slides forward via the electric rail, alternating between these positions. When a change of direction is required, the large negative pressure suction cup control subsystem in the middle of the base plate activates, and the suction cup automatically extends to just contact the concrete surface and adheres. All other negative pressure suction cup control subsystems are deactivated, and the robot rotates to the specified angle using the rotation function, then repeats the alternating crawling steps.

[0075] In this embodiment of the invention, the proposed intelligent identification method for cracks on the concrete surface of bridges based on a wall-climbing robot can not only identify cracks on the bridge surface in real time with high accuracy, but also save time and costs and reduce the safety risks of personnel inspection.

[0076] Figure 4 This is a block diagram illustrating an intelligent crack recognition system for bridge concrete surfaces based on a wall-climbing robot, according to an exemplary embodiment. (Refer to...) Figure 4 The system 300 includes:

[0077] The model integration module 310 is used to construct an integrated model of target detection and semantic segmentation to identify cracks on the concrete surface of the target bridge.

[0078] The model training module 320 is used to obtain the crack dataset according to the data format requirements of the object detection and semantic segmentation integrated model, and to annotate the crack dataset. The annotated crack dataset is used for model training and validation to obtain an end-to-end real-time crack recognition model.

[0079] Mobile deployment module 330 is used to quantitatively deploy the trained crack recognition model on a mobile device.

[0080] The device is equipped with an identification module 340, which is designed as a wall-climbing robot suitable for bridge concrete surfaces through a dual adsorption structure of negative pressure suction cup and electromagnet. The wall-climbing robot is remotely connected to a mobile terminal. The wall-climbing robot is placed at a designated position on the bottom or side of the bridge concrete beam. By operating the mobile terminal, the crack identification result corresponding to a specific location on the bridge is displayed in real time, thus completing the intelligent identification of cracks on the bridge concrete surface based on the wall-climbing robot.

[0081] Optionally, the model training module 320 is further used to add a lightweight segmentation network on the basis of the YOLOv5 object detection algorithm, construct an integrated model of object detection and semantic segmentation, and output crack object detection results and crack segmentation results; verify the crack object detection results and crack segmentation results of the concrete surface of the target bridge, and output a binary image of the crack.

[0082] Optionally, the target detection module 310 is further configured to generate a crack prediction and location map as the crack target detection result, including the coordinates of the center point of the outer rectangle of the crack and the area of ​​the rectangle; and to generate a connected domain of the predicted crack as the crack segmentation result, including the coordinates of the center point of the outer rectangle of each connected domain and the area of ​​the connected domain.

[0083] When the coordinates of the center point of the bounding box and the area of ​​the bounding box of the crack target detection result are close to those of the crack segmentation result, the connected component is determined to be a crack; otherwise, it is noise. The connected components determined to be noise are deleted, and only the connected components containing cracks are retained. The crack binarized image is then output.

[0084] Optionally, the model training module 320 is further used to annotate the crack dataset, wherein the labeled results are divided into two categories: one is YOLO data format for object detection; the other is a binarized mask image for semantic segmentation.

[0085] Optionally, the labeled dataset can be used for training and validation of the crack recognition model, wherein the ratio of training data to validation data in the crack dataset is 8:2.

[0086] Optionally, the wall-climbing robot is a wheeled robot that crawls on the bottom and sides of the bridge's concrete beams using a dual adsorption structure of negative pressure suction cups and electromagnets.

[0087] Optionally, a CCD camera is embedded in the bottom of the wall-climbing robot, which can be remotely connected to a mobile device via Wi-Fi or a SIM card.

[0088] Figure 5 This is a schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present invention. The electronic device 400 can vary considerably due to differences in configuration or performance. It may include one or more central processing units (CPUs) 401 and one or more memories 402. The memory 402 stores at least one instruction, which is loaded and executed by the processor 401 to implement the steps of the following intelligent identification method for cracks on the surface of bridge concrete based on a wall-climbing robot:

[0089] S1: Construct an integrated model of object detection and semantic segmentation to identify cracks on the concrete surface of the target bridge;

[0090] S2: According to the data format requirements of the object detection and semantic segmentation integrated model, obtain the crack dataset, label the crack dataset, use the labeled crack dataset for model training and validation, and obtain an end-to-end real-time crack recognition model.

[0091] S3: Quantize and deploy the trained crack recognition model on mobile devices;

[0092] S4: A wall-climbing robot suitable for bridge concrete surfaces is designed using a dual adsorption structure of negative pressure suction cup and electromagnet. The wall-climbing robot is remotely connected to a mobile terminal. The wall-climbing robot is placed at a designated position on the bottom or side of the bridge concrete beam. By operating the mobile terminal, the crack identification result corresponding to a specific location on the bridge is displayed in real time, thus completing the intelligent identification of cracks on the bridge concrete surface based on the wall-climbing robot.

[0093] In an exemplary embodiment, a computer-readable storage medium is also provided, such as a memory including instructions that can be executed by a processor in a terminal to complete the aforementioned intelligent identification method for cracks on the concrete surface of bridges based on a wall-climbing robot. For example, the computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, or optical data storage device.

[0094] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0095] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying surface cracks in bridge concrete based on a wall-climbing robot, characterized in that, Includes the following steps: S1: Construct an integrated model of object detection and semantic segmentation to identify cracks on the concrete surface of the target bridge; S2: According to the data format requirements of the object detection and semantic segmentation integrated model, obtain the crack dataset, label the crack dataset, use the labeled crack dataset for model training and validation, and obtain an end-to-end real-time crack recognition model. S3: Quantize and deploy the trained crack recognition model on mobile devices; S4: A wall-climbing robot suitable for bridge concrete surfaces is designed using a dual adsorption structure of negative pressure suction cup and electromagnet. The wall-climbing robot is remotely connected to a mobile terminal. The wall-climbing robot is placed at a designated position on the bottom or side of the bridge concrete beam. By operating the mobile terminal, the crack identification result corresponding to a specific location on the bridge is displayed in real time, thus completing the intelligent identification of cracks on the bridge concrete surface based on the wall-climbing robot. In step S1, an integrated model for target detection and semantic segmentation is constructed to identify cracks on the concrete surface of the target bridge, including: Based on the YOLOv5 object detection algorithm, a lightweight segmentation network is added to construct an integrated model of object detection and semantic segmentation, which outputs both crack object detection results and crack segmentation results. The crack object detection results and crack segmentation results of the concrete surface of the target bridge are verified, and a binarized image of the crack is output. The step of verifying the target crack detection results and crack segmentation results on the concrete surface of the target bridge, and outputting a binarized image of the crack, includes: The crack target detection result is a crack prediction and location map, which includes the coordinates of the center point of the outer rectangle of the crack and the area of ​​the rectangle; the crack segmentation result is the connected domain of the predicted crack, which includes the coordinates of the center point of the outer rectangle of each connected domain and the area of ​​the connected domain. When the coordinates of the center point of the bounding box and the area of ​​the bounding box of the crack target detection result are close to those of the crack segmentation result, the connected component is determined to be a crack; otherwise, it is noise. The connected components determined to be noise are deleted, and only the connected components containing cracks are retained. The crack binarized image is then output.

2. The method according to claim 1, characterized in that, In step S2, the crack dataset is labeled, including: The crack dataset is labeled, and the labeled results are divided into two categories: one is YOLO data format for object detection; the other is a binarized mask image for semantic segmentation.

3. The method according to claim 2, characterized in that, The labeled dataset was used for training and validation of the crack recognition model, with a ratio of 8:2 for training data and validation data in the crack dataset.

4. The method according to claim 1, characterized in that, The wall-climbing robot is a wheeled robot that crawls on the bottom and sides of the bridge's concrete beams using a dual adsorption structure of negative pressure suction cups and electromagnets.

5. The method according to claim 1, characterized in that, The wall-climbing robot has a CCD camera embedded in its bottom and can be remotely connected to a mobile device via Wi-Fi or a SIM card.

6. A bridge concrete surface crack identification system based on a wall-climbing robot, characterized in that, The system is applicable to the method of any one of claims 1-5 above, and the system comprises: The model integration module is used to build an integrated model of object detection and semantic segmentation to identify cracks on the concrete surface of the target bridge. The model training module is used to obtain the crack dataset according to the data format requirements of the object detection and semantic segmentation integrated model, and to annotate the crack dataset. The annotated crack dataset is used for model training and validation to obtain an end-to-end real-time crack recognition model. The mobile deployment module is used to quantitatively deploy the trained crack recognition model on mobile devices. The device is equipped with an identification module and uses a dual adsorption structure of negative pressure suction cup and electromagnet to design a wall-climbing robot suitable for bridge concrete surfaces. The wall-climbing robot is remotely connected to a mobile terminal. The wall-climbing robot is placed at a designated position on the bottom or side of the bridge concrete beam. By operating the mobile terminal, the crack identification result corresponding to a specific location on the bridge is displayed in real time, thus completing the intelligent identification of cracks on the bridge concrete surface based on the wall-climbing robot. The model training module is further used to add a lightweight segmentation network to the YOLOv5 object detection algorithm, construct an integrated model of object detection and semantic segmentation, and output crack object detection results and crack segmentation results; verify the crack object detection results and crack segmentation results of the concrete surface of the target bridge, and output a binarized image of the crack. The model training module is further used to generate a crack prediction and localization map for crack target detection, which includes the coordinates of the center point of the outer rectangle of the crack and the area of ​​the rectangle; and to generate a connected component of the predicted crack for crack segmentation, which includes the coordinates of the center point of the outer rectangle of each connected component and the area of ​​the connected component. When the coordinates of the center point of the bounding box and the area of ​​the bounding box of the crack target detection result are close to those of the crack segmentation result, the connected component is determined to be a crack; otherwise, it is noise. The connected components determined to be noise are deleted, and only the connected components containing cracks are retained. The crack binarized image is then output.

Images