Bridge crack detection method based on unmanned aerial vehicle device

Abstract

The application provides a bridge crack detection method based on a UAV device, comprising the following steps: S1, controlling a multi-rotor UAV to approach a bridge to be detected, then capturing bridge surface images of different to-be-detected areas by a visual camera, and collecting corresponding position data; S2, for each bridge surface image, using an improved YOLOv5 model to perform image recognition and feature extraction to obtain a crack target, and performing feature quantization based on the position data and a crack connected domain to obtain surface feature information of the crack target; S3, controlling the multi-rotor UAV to be adsorbed on the bridge surface at the crack target through a suction cup on a detection arm, and starting an ultrasonic depth detector to collect acoustic feedback data and analyze the crack depth value through the acoustic feedback data; and S4, integrating the surface feature information and the crack depth value to generate a bridge crack detection report. The beneficial effect is that the application can realize comprehensive detection of bridge cracks in multiple dimensions, high efficiency and high precision.

Description

Technical Field

[0001] This invention relates to the technical field of bridge inspection and drone applications, and more specifically, to a method for detecting bridge cracks based on a drone device. Background Technology

[0002] With the rapid development of my country's transportation infrastructure, bridges, as key nodes, are of paramount importance for their structural health and safe operation. Cracks are one of the most common forms of damage to bridge structures. Timely and accurate detection of parameters such as length, width, area, and depth is fundamental to assessing the degree of structural damage, predicting remaining lifespan, and developing scientific maintenance strategies.

[0003] Currently, bridge crack detection mainly relies on traditional manual inspections, bridge inspection vehicles, and the increasingly popular drone-based visual inspection. Manual inspections and bridge inspection vehicles have inherent limitations, including low efficiency, high cost, high risk (such as high-altitude or cross-sea operations), and difficulty in accessing certain areas. While drone-based non-contact visual inspection solutions can achieve rapid, large-scale image acquisition and crack identification, their detection capabilities are typically limited to a two-dimensional plane: image processing algorithms can obtain surface geometric features such as crack length, width, and area, but cannot acquire depth information within the structure. Depth is a crucial three-dimensional parameter for assessing the severity of cracks and determining whether they are shallow surface cracks or penetrating cracks; its absence results in incomplete detection conclusions, making it difficult to support in-depth structural safety assessments.

[0004] On the other hand, crack depth detection currently relies mainly on ground-based, manually operated contact or semi-contact equipment such as ultrasonic flaw detectors and crack microscopes. These methods require inspectors to reach the vicinity of the crack for point-to-point operation, making it impossible to synchronize with UAV visual inspection. This results in a broken inspection process, time-consuming and labor-intensive operations, and inconsistent data in time and space. Furthermore, existing technologies lack integrated devices capable of stably and reliably performing contact depth detection on UAV platforms. UAVs are typically used only as image acquisition platforms, and their multi-functional fusion detection capabilities have not been fully utilized.

[0005] Therefore, the urgent challenge in the existing technology is: how to design a bridge crack detection method based on UAV devices that can continuously and automatically complete crack identification, accurate two-dimensional size measurement and three-dimensional non-destructive detection in one system, so as to achieve comprehensive detection of bridge cracks in a multi-dimensional, efficient and high-precision manner. Summary of the Invention

[0006] The technical problem to be solved by this invention is how to achieve comprehensive detection of bridge cracks in a multi-dimensional, efficient and high-precision manner. In order to overcome the defects of the above-mentioned existing technology (or related technology), this invention provides a bridge crack detection method based on a drone device. This invention provides a bridge crack detection method based on a drone device. A visual camera and an ultrasonic depth detector are pre-configured on a multi-rotor drone. At least one detection arm is connected to the fuselage of the multi-rotor drone, and a suction cup is mounted at the end of each detection arm. The bridge crack detection method includes the following steps: Step S1: Control the multi-rotor drone to approach the bridge to be inspected, and then use the vision camera to capture bridge surface images of different areas to be inspected on the bridge to be inspected, and collect position data between the multi-rotor drone and the corresponding area to be inspected when capturing each bridge surface image. Step S2: For each bridge surface image, an improved YOLOv5 model is used to perform image recognition and feature extraction on the bridge surface image to obtain the crack target, and feature quantization is performed based on the location data and the crack connected domain to obtain the surface feature information of the crack target; Step S3: Control the multi-rotor UAV to attach to the bridge surface at the crack target using the suction cup on the detection arm, and activate the ultrasonic depth detector to collect acoustic feedback data and analyze the acoustic feedback data to obtain the crack depth value. Step S4: Integrate the surface feature information and the crack depth value to generate a bridge crack detection report.

[0007] The bridge crack detection method based on an unmanned aerial vehicle (UAV) device of the present invention has the following advantages compared with the prior art: In this invention, bridge surface images and location data are acquired in step S1, crack target identification and surface feature information analysis are performed in step S2, crack depth values ​​are calculated in step S3, and a bridge crack detection report is generated in step S4. This achieves full automation of the bridge crack detection process. By integrating a visual camera and an ultrasonic depth detector into a UAV platform, crack identification, location, surface feature extraction, and depth measurement can be completed in a single flight mission, significantly improving detection efficiency and safety, and reducing reliance on manual inspection at height. The generated bridge crack detection report contains diverse information on surface features and crack depth values, which can help engineers conduct multi-dimensional bridge condition analysis and achieve comprehensive, efficient, and high-precision detection of bridge cracks.

[0008] In one possible implementation, the improvements to the YOLOv5 model used in step S2 include introducing a CA attention mechanism, introducing a Ghost lightweight module to replace the target proportion CBS structure, and replacing the GIoU loss function with an Alpha-DIoU loss function.

[0009] Compared with existing technologies, the above-mentioned technical solution can enhance the ability to capture subtle crack features through the CA attention mechanism, reduce the computational load of the YOLOv5 model through the Ghost lightweight module, making it suitable for real-time operation on airborne or edge devices, and improve the localization accuracy of crack target bounding boxes through the Alpha-DIoU loss function, making it particularly suitable for crack target detection in complex backgrounds.

[0010] In one possible implementation, step S2, the process of obtaining the crack target through image recognition and feature extraction, includes: Step A1: The bridge surface image is enhanced using the multi-scale MSR algorithm to obtain the enhanced image; Step A2: The improved Otsu algorithm, which introduces target weight coefficients, is used to perform region image segmentation on the enhanced image that is biased towards the crack target to obtain a binarized image; Step A3: Perform a morphological closure operation on the binarized image to connect the fractured parts of the crack target to form a crack connected region.

[0011] Compared with existing technologies, the above-mentioned technical solution can effectively improve the usability and segmentation accuracy of the image where the crack target is located through the multi-layer image processing flow of steps A1, A2 and A3. It overcomes the problem of uneven illumination by using the multi-scale MSR algorithm, and improves the detection rate of shallow cracks and blurred cracks by introducing weight coefficients in the improved Otsu algorithm to make the regional image segmentation more biased towards the crack target. Furthermore, the morphological closure operation can repair the fracture crack to form a complete crack connected domain, laying the foundation for subsequent geometric measurement.

[0012] In one possible implementation, the multi-rotor drone is pre-equipped with a ranging sensor. In step S1, for each area to be detected, when the multi-rotor drone hovers and takes an image of the bridge surface of the area to be detected, the straight-line distance between the visual camera and the bridge surface is measured by the ranging sensor and included in the position data.

[0013] Compared with existing technologies, the above technical solution can provide key spatial information for image size calibration through real-time ranging. The ranging sensor can directly obtain the actual distance between the visual camera and the bridge surface, avoiding the error caused by relying on image to estimate the distance, and providing a reliable basis for the subsequent conversion of pixel coordinates to actual size.

[0014] In one possible implementation, in step S2, the conversion coefficient between pixel coordinates and actual size is calculated based on each of the straight-line distances, combined with the camera focal length and imaging principle, and included in the position data.

[0015] Compared with existing technologies, the above technical solution establishes a precise conversion relationship between pixel coordinates and actual physical dimensions, making image-based crack size measurement meaningful in practical engineering. This conversion coefficient integrates imaging geometry principles and measured distances, ensuring that the measurement results of two-dimensional parameters such as length, width, and area are accurate and reliable.

[0016] In one possible implementation, in step S2, the crack connected regions of the crack target are marked and noise points with areas smaller than a preset value are filtered out. The total number of pixels in the crack connected regions is counted and combined with the conversion coefficient to obtain the actual area included in the surface feature information.

[0017] Compared with existing technologies, the above-mentioned technical solution can accurately quantify the crack area. By marking the crack connected components and filtering out noise, image interference is eliminated. In addition, the pixel area can be converted into the actual area by combining the conversion coefficient, providing key area parameters for assessing the degree of crack damage.

[0018] In one possible implementation, in step S2, the skeleton of the crack connected region of the crack target is extracted and the number of pixels of the skeleton is counted. Then, the actual length is obtained by combining the conversion coefficient and included in the surface feature information.

[0019] Compared with existing technologies, the above technical solution can provide an accurate method for measuring crack length. Skeleton extraction can accurately reflect the center path of the crack target, overcoming the problem of measuring the length of curved and bifurcated cracks. Combined with the conversion coefficient, it can realize the automated and high-precision calculation of the length of complex crack shapes.

[0020] In one possible implementation, in step S2, the width value at each pixel is obtained by calculating the distance from each pixel on the image where the skeleton is located to the background, and then the average width, maximum width and minimum width are obtained by combining the conversion coefficient and included in the surface feature information.

[0021] Compared with existing technologies, the above technical solution can achieve multi-index quantification of crack width. By calculating the distance from each pixel of the skeleton to the background edge, not only the average width can be obtained, but also the maximum and minimum widths can be extracted, comprehensively reflecting the crack width distribution and providing more detailed width data for structural safety assessment.

[0022] In one possible implementation, in step S3, ultrasonic waves are continuously released to the bridge surface through the ultrasonic depth detector, allowing the ultrasonic waves to propagate inside the concrete structure. The waveform of the first wave signal is received through the ultrasonic depth detector, and the phase reversal point of the first wave is obtained based on the waveform of the first wave signal. The crack depth value is obtained based on the principle of geometric acoustics in combination with the pre-calibrated concrete wave velocity.

[0023] Compared with existing technologies, the above-mentioned technical solution can achieve non-destructive and accurate measurement of crack depth. By using ultrasonic phase reversal point detection technology and combining it with geometric acoustic principles, the crack depth value inside the wall can be directly calculated, which makes up for the deficiency of pure visual inspection in obtaining three-dimensional depth information, and enables crack assessment to move from two-dimensional to three-dimensional.

[0024] In one possible implementation, an acceleration detection module is provided at the center of the suction cup at the end of the detection arm. In step S3, after the multi-rotor UAV is controlled to adhere to the bridge surface at the crack target, the acceleration detection module collects vibration data of the bridge under natural conditions and under external excitation conditions when vehicles pass by, and obtains structural performance information by analyzing time-domain and frequency-domain characteristics, which is then added to the bridge crack detection report.

[0025] Compared with existing technologies, the above-mentioned technical solution can expand the dimensions of structural safety assessment for crack detection. By collecting vibration data of the bridge under natural and external excitation conditions through the acceleration detection module, the impact of crack targets on structural stiffness and dynamic characteristics can be analyzed, providing dynamic data support for the assessment of bridge health status and remaining life. Attached Figure Description

[0026] Figure 1 This is a schematic diagram of the overall structure of the multi-rotor UAV of the present invention; Figure 2 This is a flowchart illustrating the overall steps of the present invention; Figure 3 This is a flowchart of the image recognition and feature extraction steps of the present invention; Figure 4 This is a schematic diagram illustrating the lens imaging principle during the calculation of the conversion coefficient in this invention. Figure 5 This is a schematic diagram showing the position when confirming the reversal point in this invention; Figure 6 This is a schematic diagram of the concrete structure experimental verification of the present invention; Figure 7 This is a depth diagram used in calculating the crack depth value according to the present invention; Explanation of reference numerals in the attached figures: 1. Visual camera; 2. Ultrasonic depth detector; 3. Buffer spring; 4. Detector arm; 5. Suction cup; 6. Miniature linear slide rail. Detailed Implementation

[0027] First, those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention. Those skilled in the art can make adjustments as needed to adapt to specific application scenarios.

[0028] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0029] See Figure 1 This invention discloses a bridge crack detection method based on a drone device. The drone device includes a drone platform carrying a multi-rotor drone, such as a hexacopter drone. The drone platform has stable hovering and position holding capabilities. A high-resolution visual camera 1 with autofocus, such as a 20-megapixel camera, and an ultrasonic depth detector 2 are integrated on the multi-rotor drone body. At least one foldable detection arm 4 is connected to the body via a buffer spring 3. The end of the detection arm 4 is equipped with a silicone suction cup 5. A micro vacuum pump interface is integrated in the center of the suction cup. An acceleration detection module consisting of a three-axis accelerometer and a gyroscope is also embedded on the surface of the suction cup 5.

[0030] See Figure 2 The bridge crack detection method disclosed in this invention mainly includes the following steps: Step S1: Control the multi-rotor drone to approach the bridge to be inspected, and then use the vision camera 1 to capture images of the bridge surface of different areas to be inspected on the bridge, and collect position data between the multi-rotor drone and the corresponding area to be inspected when capturing each bridge surface image. Step S2: For each bridge surface image, the improved YOLOv5 model is used to perform image recognition and feature extraction on the bridge surface image to obtain the crack target, and the surface feature information of the crack target is obtained by feature quantization based on the location data and the crack connected domain. Step S3: Control the multi-rotor UAV to attach to the bridge surface at the crack target location via the suction cup 5 on the detection arm 4, and start the ultrasonic depth detector 2 to collect acoustic feedback data and analyze the acoustic feedback data to obtain the crack depth value. Step S4: Integrate surface feature information and crack depth values ​​to generate a bridge crack detection report.

[0031] In this embodiment of the invention, in step S1, the operator plans the flight path through the ground control station, controls the multi-rotor drone to fly to the vicinity of the bridge to be inspected, such as the bottom plate of the box girder, the pier, or the bridge tower, and maintains a safe distance. After the multi-rotor drone enters the hovering mode, the vision camera 1 begins to continuously capture images of the bridge surface at a preset frame rate, such as 1 frame per second, and transmits them back to the ground control station in real time through 5G or a dedicated image transmission system. At the same time, the laser rangefinder, such as the TOF sensor, carried by the multi-rotor drone measures the straight-line distance from the optical center of the vision camera 1 to the shooting point on the bridge surface in real time, and records it together with the timestamp of each frame of the bridge surface image, GPS / RTK positioning data, and drone attitude angle to form a complete location data packet.

[0032] In this embodiment of the invention, in step S2, the ground control station automatically processes the received bridge surface images, mainly including crack identification, image enhancement and segmentation, and feature quantization. Specifically, during crack identification, an improved YOLOv5 model is used to detect crack targets in the bridge surface images in real time. This YOLOv5 model has undergone transfer learning during the training phase and has been fine-tuned using a large number of bridge crack images. Its structural improvements include: The Coordinate Attention (CA) mechanism is introduced into the Backbone and Neck parts of the YOLOv5 model, making the YOLOv5 network pay more attention to the linear texture features of the crack target. The original CBS (Conv-BN-Silu) structure was partially replaced by the Ghost lightweight module, and the size of the YOLOv5 model was compressed to facilitate deployment while ensuring accuracy. Replacing the bounding box regression loss function with the Alpha-DIoU loss function improves the localization accuracy and convergence speed of small target cracks.

[0033] See Figure 3 In this embodiment of the invention, during image enhancement and segmentation, local processing is performed on the identified crack areas, specifically including: Step A1: The multi-scale MSR algorithm is used to normalize the illumination of the bridge surface image and enhance the visibility of cracks in the shadow or reflective areas to obtain the enhanced image. Step A2: Use the improved Otsu algorithm for threshold segmentation. When calculating the inter-class variance, assign a weight coefficient k (k>1, such as 1.2) to the crack target to make the threshold more biased towards crack pixels. Then, combine K-means clustering to perform morphological correction on the segmentation results to obtain a more accurate binarized image. Step A3: Perform a morphological closure operation on the binarized image to connect the fractures caused by surface stains or uneven lighting, forming a complete fracture connected region.

[0034] In this embodiment of the invention, an improved YOLOv5 model is used to traverse all pixels in the binarized image. All pixels with the same pixel value (i.e., 0 or 1 in the binarized image) are grouped into a connected region. The connected region is assigned a unique identifier to distinguish it from other non-connected regions. The Seed-Filling method is used to traverse the binarized image until B(X,Y)=1. The calculation logic is as follows: B(X,Y) is used as a seed and assigned a label. All adjacent foreground pixels in the seed are pushed onto a stack. Then, the top pixel is popped from the stack and assigned the same label. All foreground pixels adjacent to the top pixel are pushed onto the stack. The above steps are repeated until the stack is empty.

[0035] In this embodiment of the invention, the connected component labeling method can traverse all connected components in the binarized image, label them, and return parameters such as the number of pixels in each group of connected components. Therefore, this embodiment proposes a noise reduction method based on connected components. The principle is to set a threshold for the number of pixels in each connected component group based on the connected component labeling. The method traverses the labeled connected components in the binarized image and returns the number of pixels in each of these connected component groups. For connected components with a number of pixels significantly greater than the threshold, they are labeled as true (i.e., the connected component is classified as the target region) and assigned a true label. For connected components with a number of pixels less than the threshold, they are labeled as false (i.e., the connected component is classified as the background region) and assigned a false label. The code filters out the groups with false labels. These groups are usually residual binary image noise in the background, and the number of pixels in their connected components is significantly less than the threshold, thereby completing the noise reduction of the binarized image.

[0036] In this embodiment of the invention, feature quantization specifically includes conversion coefficient calculation, area calculation, length calculation, and width calculation. Specifically, when calculating the conversion coefficient, see [link to relevant documentation]. Figure 4 Based on the imaging principle of a convex lens, the following formula can be derived:

[0037] in, The object distance, or straight-line distance, from the multi-rotor UAV to the target in the crack can be directly obtained from the ranging sensor carried by the multi-rotor UAV. Indicates the distance. Indicates the focal length of visual camera 1; Then, based on the principle of similar triangles, the following formula can be derived:

[0038]

[0039] in, Indicates the actual size of the crack target. This indicates the size of the crack target after imaging; The size of the crack target after imaging. It can also be obtained from the following formula:

[0040] in, Indicates the pixel size of the ranging sensor. Indicates image resolution; The following formula can then be derived from the above formula:

[0041] in, This represents the conversion coefficient. Therefore, when the straight-line distance from the lens position of the multi-rotor UAV's vision camera 1 to the crack target is measured using a range sensor, the actual parameters of the crack target can be calculated.

[0042] In this embodiment of the invention, when calculating the area, the cv2.connectedComponentsWithStats function in the OpenCV library is used to mark the connected regions, filter out noise regions with pixel areas smaller than a preset threshold such as 50 pixels, count the total number of pixels in the target connected regions, and multiply by the conversion coefficient to obtain the actual area of ​​the crack target.

[0043] In this embodiment of the invention, when calculating the length, the connected region markers of the crack feature are first subjected to dilation and erosion operations to obtain its skeleton structure. Then, a thinning process is performed to obtain a more refined skeleton. Subsequently, the shape skeleton is extracted. In Python, this can be achieved by the morphology.medial_axis() function in the skimage library. Then, the number of pixels in the skeleton is counted, and the actual size represented by a single pixel is calculated based on the conversion coefficient. Finally, the number of pixels in the skeleton is counted and multiplied by the conversion coefficient to obtain the actual length of the crack target.

[0044] In this embodiment of the invention, when calculating the width, the skeleton is first calculated using skimage.morphology.medial_axis(image, mask=None, return_distance=True), and return_distance=True is set therein to obtain the distance values ​​between all pixels on the central axis and the background points. Similarly, the points processed here are pixels, and the actual distance values ​​need to be calculated using conversion coefficients to determine the actual distance represented by the pixels. Then, the distance values ​​between all pixels on the central axis and the background points are calculated. This method can help us calculate the crack width more accurately and improve the accuracy of the evaluation of bridge structural safety. Then, by calling the np.mean(width), np.max(width), and np.min(width) functions, the average width, maximum width, and minimum width of the crack target can be obtained. The maximum width of the crack target is used as a reference for subsequent bridge damage assessment.

[0045] In this embodiment of the invention, in step S3, when structural dynamic testing is required, the operator controls the multi-rotor drone to slowly approach the identified crack target location until the suction cup 5 at the end of the detection arm 4 contacts the bridge surface. Then, a micro vacuum pump is activated to create negative pressure inside the suction cup 5, which stably adsorbs the multi-rotor drone onto the bridge surface. At this time, the acceleration detection module is in close contact with the bridge surface and synchronously collects vibration data for at least 30 seconds, including environmental excitations (such as wind, background vibration) and optional external excitations (such as coordinating vehicles to pass over the bridge surface). The structural dynamic characteristics at this location are evaluated through time domain analysis (peak value, root mean square) and frequency domain analysis (FFT transformation, extraction of dominant frequency and harmonics). The data is archived for later use.

[0046] In this embodiment of the invention, in step S3, when it is necessary to obtain the crack depth value, the ultrasonic depth detector 2 is activated. It includes a pair of ultrasonic transducers with a center frequency of 50kHz, which are symmetrically installed on the miniature linear slide rail 6. The two ultrasonic transducers are controlled to move synchronously in opposite directions from the initial position with equal steps. Each step of movement transmits and receives an ultrasonic signal. The received first wave waveform is analyzed in real time to detect whether its phase has reversed, that is, from a positive peak to a negative peak. When the first wave phase reversal point is detected, the current position of the two ultrasonic transducers is recorded. Combined with the concrete ultrasonic wave velocity obtained in advance on site and the geometric trajectory of the ultrasonic transducer movement, the crack depth value is automatically calculated and output according to the diffraction wave time difference positioning principle.

[0047] In this embodiment of the invention, see Figure 5 and Figure 6The ultrasonic transmitting transducer emits ultrasonic waves at mounting point P1 on the load cell. The ultrasonic waves propagate within the concrete structure. Within the maximum detection time threshold, it is determined whether the ultrasonic receiving transducer can receive the wave signal at mounting point P2 on the load cell. If it can, the first wave signal received by the ultrasonic receiving transducer is recorded, and the waveform of the first wave signal is analyzed. The first wave signal waveform includes two types: a downward convex waveform and an upward convex waveform. Then, along the circular measuring line, with the detection point as the center, the ultrasonic transmitting transducer and ultrasonic receiving transducer move in opposite directions with equal steps to the next position, such as... Figure 5 As shown; because the curvature of wall cracks is generally too large, and due to the limitations of crack physical characteristics and structural design, the probe arm 4 is designed with a large-curvature guide rail moving in equal steps. According to the principle, the object that the vision camera 1 is aimed at is a crack-free wall surface, and the purpose is to detect the position of the first wave phase reversal using equal-step movement. Assuming the curvature of the probe line is infinitely large, after theoretical calculation and verification, the actual minimum depth point of the crack differs from the theoretically tested minimum depth point by only 5% in accuracy. For the actual crack morphology and depth of bridges, this difference is negligible. Therefore, the large-curvature symmetrical distribution guide rail method is feasible. Experimental verification on a concrete structure is shown below. Figure 6 As shown.

[0048] In this embodiment of the invention, based on Figure 6 The distance relationships shown in the figure are used to calculate the actual range and location of the lowest crack depth point using the following formulas:

[0049]

[0050] in, This indicates the range of the actual lowest depth point of the crack. This indicates the location of the actual lowest depth point of the crack. In a high-curvature survey line experiment, the initial position of a single location is 19 cm from the crack. The phase reversal point of the first wave is symmetrically located at the crack detection point at 32.5 cm. Therefore, the depth value measured at this point is:

[0051]

[0052] in, This indicates the range of depth values ​​at the lowest depth point in a high-curvature survey line experiment. The depth value of the lowest depth point in the large curvature survey line experiment; The experiment concluded that:

[0053]

[0054] in, Indicates the error in the principle test. This represents the actual test error, which is negligible because the crack depth in reality is too small.

[0055] See Figure 7 In this embodiment of the invention, when calculating the crack depth, the bottom tip of the crack target is represented as point C, and the straight-line distance from point C to detection point A is represented as... , This is the crack depth value to be determined, and the distance from point B to point C is... It is determined by the following method: An ultrasonic transducer emits ultrasonic waves at position B. The time T at which the ultrasonic receiving transducer receives the wave signal at position D is determined, which is the first wave phase reversal point. The ultrasonic waves emitted by the ultrasonic transducer at position B pass through the bottom tip of the crack target, i.e., point C, and then diffract at point C. They are then received by the ultrasonic receiving transducer at position D. This wave signal is a diffracted wave. Therefore, the path of wave propagation and S are: S = distance from point B to point C. + Distance from point C to point D Since location points D and B are symmetrically arranged from detection point A, therefore, the distance... = Therefore, the distance can be obtained using the following formula. :

[0056] in, This indicates the ultrasonic wave velocity of the cylindrical concrete determined by pretreatment. The arc lengths of the measured detection points A and B are then expressed as arc length AB. The central angle ∠AOB corresponding to arc length AB is calculated using the following formula:

[0057] in, Indicates the radius of the circular survey line; Then, the length of chord AB is obtained based on ∠AOB. And the angle ∠BAC:

[0058]

[0059] For triangle ABC, the lengths of its two sides are: and Therefore, ∠BAC is obtained. Thus, using the Law of Sines, the sine value of ∠ACB, sin∠ACB, is obtained as follows:

[0060] Determine the angle ∠ACB based on the sine of ∠ACB, sin∠ACB, and then determine the angle ∠ABC using the following formula:

[0061] The crack depth can be obtained using the cosine theorem based on the following formula. :

[0062] This enables the calculation of the depth of longitudinal cracks in cylindrical concrete.

[0063] In this embodiment of the invention, in step S4, all collected data, including crack images, two-dimensional dimensions, depth, acceleration spectrum, and location information, are automatically uploaded to the cloud or local database. A report generation template is called to integrate the multi-dimensional data of the same crack target into a structured bridge crack detection report. The report includes a crack location diagram, a detailed dimension data table, depth values, a preliminary assessment of the safety level based on width and depth thresholds, and a summary of vibration characteristics. It supports exporting to PDF or CAD-compatible formats for maintenance units to use for decision-making.

[0064] In the description of this invention, the references to "one embodiment," "some embodiments," "in this embodiment," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0065] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting bridge cracks based on an unmanned aerial vehicle (UAV) device, characterized in that, A visual camera and an ultrasonic depth detector are pre-configured on a multi-rotor drone. At least one detection arm is attached to the fuselage of the multi-rotor drone, and a suction cup is mounted at the end of each detection arm. The bridge crack detection method includes the following steps: Step S1: Control the multi-rotor drone to approach the bridge to be inspected, and then use the vision camera to capture bridge surface images of different areas to be inspected on the bridge to be inspected, and collect position data between the multi-rotor drone and the corresponding area to be inspected when capturing each bridge surface image. Step S2: For each bridge surface image, an improved YOLOv5 model is used to perform image recognition and feature extraction on the bridge surface image to obtain the crack target, and feature quantization is performed based on the location data and the crack connected domain to obtain the surface feature information of the crack target; Step S3: Control the multi-rotor UAV to attach to the bridge surface at the crack target using the suction cup on the detection arm, and activate the ultrasonic depth detector to collect acoustic feedback data and analyze the acoustic feedback data to obtain the crack depth value. Step S4: Integrate the surface feature information and the crack depth value to generate a bridge crack detection report.

2. The bridge crack detection method according to claim 1, characterized in that, The improvements to the YOLOv5 model used in step S2 include introducing a CA attention mechanism, introducing a Ghost lightweight module to replace the target ratio CBS structure, and replacing the GIoU loss function with an Alpha-DIoU loss function.

3. The bridge crack detection method according to claim 1, characterized in that, In step S2, the process of obtaining the crack target through image recognition and feature extraction includes: Step A1: The bridge surface image is enhanced using the multi-scale MSR algorithm to obtain the enhanced image; Step A2: The improved Otsu algorithm, which introduces target weight coefficients, is used to perform region image segmentation on the enhanced image that is biased towards the crack target to obtain a binarized image; Step A3: Perform a morphological closure operation on the binarized image to connect the fractured parts of the crack target to form a crack connected region.

4. The bridge crack detection method according to claim 1, characterized in that, The multi-rotor drone is pre-equipped with a ranging sensor. In step S1, for each area to be detected, when the multi-rotor drone hovers and takes a picture of the bridge surface of the area to be detected, the straight-line distance between the visual camera and the bridge surface is measured by the ranging sensor and included in the position data.

5. The bridge crack detection method according to claim 4, characterized in that, In step S2, the conversion coefficient between pixel coordinates and actual size is calculated based on the straight-line distances and combined with the camera focal length and imaging principle, and included in the position data.

6. The bridge crack detection method according to claim 5, characterized in that, In step S2, the crack connected regions of the crack target are marked and noise points with an area smaller than a preset value are filtered out. The total number of pixels in the crack connected regions is counted and combined with the conversion coefficient to obtain the actual area included in the surface feature information.

7. The bridge crack detection method according to claim 5, characterized in that, In step S2, the skeleton of the crack connected region of the crack target is extracted and the number of pixels of the skeleton is counted. Then, the actual length is obtained by combining the conversion coefficient and included in the surface feature information.

8. The bridge crack detection method according to claim 7, characterized in that, In step S2, the width value at each pixel is obtained by calculating the distance from each pixel on the image where the skeleton is located to the background. Then, the average width, maximum width, and minimum width are obtained by combining the conversion coefficient and included in the surface feature information.

9. The bridge crack detection method according to claim 1, characterized in that, In step S3, ultrasonic waves are continuously released to the bridge surface through the ultrasonic depth detector, allowing the ultrasonic waves to propagate inside the concrete structure. The waveform of the first wave signal is received through the ultrasonic depth detector, and the phase reversal point of the first wave is obtained based on the waveform of the first wave signal. The crack depth value is obtained based on the principle of geometric acoustics in combination with the pre-calibrated concrete wave velocity.

10. The bridge crack detection method according to claim 1, characterized in that, An acceleration detection module is provided at the center of the suction cup at the end of the detection arm. In step S3, after the multi-rotor UAV is controlled to adhere to the bridge surface at the crack target, the acceleration detection module collects vibration data of the bridge under natural conditions and under external excitation conditions when vehicles pass by, and obtains structural performance information by analyzing time-domain and frequency-domain characteristics and adding it to the bridge crack detection report.

Images