A feature enhancement and semantic segmentation bridge crack identification method and device

Abstract

The application relates to the technical field of bridge structure monitoring, and discloses a bridge crack identification method and device based on feature enhancement and semantic segmentation, which comprises the following steps: obtaining bridge-related original data; preprocessing the original data to obtain preprocessed image data; performing enhancement processing on the preprocessed image data to obtain an enhanced feature map; performing crack region identification on the enhanced feature map by using an improved semantic segmentation model to obtain a crack segmentation image; performing parameter calculation on the crack segmentation image, combining associated data to verify the result, and obtaining a crack detection result; and arranging the crack detection result to obtain a bridge crack detection report. The application can capture crack features, effectively separate cracks from background interference, optimize crack edge segmentation effect, reduce crack breaking phenomenon, reduce the situation of small crack missed detection and background misjudgment, and improve the accuracy and integrity of crack identification.

Description

Technical Field

[0001] This invention relates to the field of bridge structure monitoring technology, specifically to a method and apparatus for identifying bridge cracks using feature enhancement and semantic segmentation. Background Technology

[0002] As a core component of transportation infrastructure, the structural health of bridges directly affects traffic safety and transportation efficiency. Cracks are one of the most common defects in bridge concrete structures. Under the long-term influence of loads and environmental erosion (such as temperature changes, humidity effects, and chemical corrosion), cracks will continue to expand, eventually leading to a decrease in structural strength, durability failure, and even collapse. Therefore, timely and accurate identification of bridge cracks and assessment of their parameters (length, width, and area) is a crucial aspect of bridge maintenance.

[0003] With the expansion of bridge construction, my country now ranks among the world's top countries in terms of the total number of highway bridges. Traditional manual inspection methods (such as visual inspection and measurement) suffer from low efficiency, high labor intensity, strong subjectivity, and high missed detection rates, and are also difficult to adapt to the inspection needs in dangerous environments such as high altitudes and complex terrains. In recent years, automatic inspection technology based on computer vision and deep learning has gradually become a research hotspot. Bridge images are collected using equipment such as drones and high-definition cameras, and semantic segmentation models (such as UNet and FCN) are used to identify crack areas. However, in actual inspection scenarios, bridge images are often affected by factors such as uneven lighting, background texture interference (such as pores and stains on concrete surfaces), and narrow and discontinuous cracks. This leads to problems such as insufficient feature extraction, low semantic segmentation accuracy, and serious missed detection of small cracks in existing technologies, making it difficult to meet the requirements of engineering practice for inspection accuracy and reliability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a method and apparatus for identifying bridge cracks using feature enhancement and semantic segmentation, in order to solve the aforementioned technical problems.

[0005] Firstly, a method for identifying bridge cracks using feature enhancement and semantic segmentation is provided, including: Obtain raw data related to the bridge; The original data is preprocessed to obtain preprocessed image data; The preprocessed image data is then enhanced to obtain an enhanced feature map; An improved semantic segmentation model is used to identify crack regions in the enhanced feature map, resulting in a crack segmentation image. The parameters of the crack segmentation image are calculated, and the results are verified by combining the associated data to obtain the crack detection results; The crack detection results are compiled to obtain a bridge crack detection report.

[0006] Furthermore, the acquisition of bridge-related raw data includes: The original image of the target area of ​​the bridge is obtained through image acquisition equipment, and the corresponding environmental parameters are obtained through the built-in sensor of the acquisition equipment or the external environmental monitoring module. The original image and environmental parameters are associated and stored in the bridge detection database according to the correspondence. The bridge detection database stores historical detection data and bridge structural parameters.

[0007] Furthermore, the preprocessing of the original data to obtain preprocessed image data includes: Based on the intrinsic and extrinsic parameters of the image acquisition device, the original image is subjected to distortion correction to obtain the corrected image; An adaptive median filtering algorithm is used to suppress noise in the corrected image, resulting in preprocessed image data.

[0008] Furthermore, the enhancement processing of the preprocessed image data to obtain an enhanced feature map includes: Based on the illumination-related information in the environmental parameters, illumination correction is performed on the preprocessed image data to obtain an illumination-equalized image. The illumination equalization image is sampled at multiple scales and edge enhancement is performed to obtain a multi-scale edge-enhanced image; The multi-scale edge enhancement image is fused across scales using a pyramid pooling module to obtain an enhanced feature map.

[0009] Furthermore, the step of using an improved semantic segmentation model to identify crack regions in the enhanced feature map to obtain a crack segmentation image includes: Construct an improved semantic segmentation model; The enhanced feature map is encoded and decoded using the improved semantic segmentation model to obtain a crack region probability map. The probability map is thresholded to convert it into a binary image, and then morphological processing is used to fill holes and connect broken segments to obtain a crack segmentation image.

[0010] Furthermore, the parameter calculation for the crack segmentation image includes: The crack centerline was obtained using a skeleton extraction algorithm; Sampling points are selected along the center line of the crack, and the vertical width of the crack at each sampling point is calculated to determine the average width and maximum width of the crack. The number of pixels in the crack area is counted, and the actual area of ​​the crack is calculated by combining the conversion relationship between pixel area and actual area.

[0011] Furthermore, the result verification based on associated data includes: The calculated crack parameters are compensated and corrected based on the environmental parameters. Retrieve historical inspection data for the same area from the bridge inspection database, calculate the deviation between the current inspection results and the historical results, and mark areas where the deviation exceeds the preset range; The reliability of the segmentation results is evaluated using accuracy, recall, and F1 score. If the reliability does not meet the preset standard, the feature enhancement and semantic segmentation process is re-executed.

[0012] Furthermore, the process of organizing the crack detection results to obtain a bridge crack detection report includes: The crack segmentation images, crack parameters, and reliability assessment results are compiled to generate a standardized inspection report that includes basic bridge information, inspection-related information, crack parameters, and maintenance recommendations.

[0013] Secondly, a bridge crack identification device based on feature enhancement and semantic segmentation is provided, which is based on a bridge crack identification method based on feature enhancement and semantic segmentation as described in any of the preceding claims, including: The acquisition module is configured to acquire raw data related to the bridge. The preprocessing module is configured to preprocess the original data to obtain preprocessed image data; The enhancement processing module is configured to perform enhancement processing on the preprocessed image data to obtain an enhanced feature map; The recognition module is configured to use an improved semantic segmentation model to identify crack regions in the enhanced feature map, thereby obtaining a crack segmentation image; The verification module is configured to calculate parameters on the crack segmentation image and verify the results by combining them with associated data to obtain the crack detection results. The processing module is configured to process the crack detection results to obtain a bridge crack detection report.

[0014] Thirdly, a terminal is provided, including a processor and a memory storing program instructions, the processor being configured to, when executing the program instructions, perform a bridge crack identification method based on feature enhancement and semantic segmentation as described in any of the preceding claims.

[0015] The invention employing the above technical solution has the following advantages: This invention, through multi-scale feature enhancement and improved semantic segmentation model, can capture crack features, effectively separate cracks from background interference, optimize crack edge segmentation, reduce crack breakage, decrease the occurrence of missed small cracks and misjudgments of the background, and improve the accuracy and completeness of crack identification.

[0016] Thanks to its adaptive illumination correction and dual-branch attention mechanism, the model can effectively cope with complex detection environments such as different lighting and humidity. Even in adverse scenarios such as backlighting, shadows, and concrete surfaces with many stains, it can still maintain stable recognition performance without the need for additional manual intervention to adjust parameters.

[0017] This invention effectively reduces the impact of factors such as shooting angle and environmental interference on parameter measurement through image distortion correction, environmental parameter compensation and historical data verification mechanisms. This makes the measurement results of parameters such as crack length, width and area more reliable and consistent, and can meet the actual needs of bridge maintenance engineering teams for detection accuracy.

[0018] This invention automates the entire process from image acquisition, preprocessing, feature enhancement, semantic segmentation to parameter calculation and report generation, reducing manual intervention, improving detection efficiency, adapting to the batch detection needs of large-scale bridge clusters, and reducing detection manpower costs.

[0019] The standardized test report output by this invention is comprehensive and formatted in a standardized manner. The data can be directly connected to the bridge maintenance management platform, providing maintenance personnel with clear information on crack distribution, parameters, and maintenance suggestions. This helps in making maintenance decisions, promptly identifying potential safety hazards in bridge structures, reducing bridge safety risks, and extending the service life of bridges. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0021] Figure 1 This is a flowchart of a bridge crack identification method based on feature enhancement and semantic segmentation according to the present invention; Figure 2 This is a flowchart of a bridge crack identification device based on feature enhancement and semantic segmentation according to the present invention; Figure 3 This is a schematic diagram of the structure of a terminal according to the present invention. Detailed Implementation

[0022] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0023] like Figures 1-3 As shown, a bridge crack identification method based on feature enhancement and semantic segmentation according to the present invention includes: Step S01: Obtain the original data related to the bridge; Step S02: Preprocess the original data to obtain preprocessed image data; Step S03: Perform enhancement processing on the preprocessed image data to obtain enhanced feature maps; Step S04: Use an improved semantic segmentation model to identify crack regions in the enhanced feature map to obtain a crack segmentation image; Step S05: Calculate the parameters of the crack segmentation image and verify the results by combining them with the associated data to obtain the crack detection results; Step S06: Organize the crack detection results to obtain the bridge crack detection report.

[0024] In this embodiment, obtaining raw data related to the bridge includes: The original image of the target area of ​​the bridge is obtained through image acquisition equipment, and the corresponding environmental parameters are obtained through the built-in sensor of the acquisition equipment or the external environmental monitoring module. The original images are associated with environmental parameters according to their correspondence and stored in the bridge detection database, which stores historical detection data and bridge structural parameters.

[0025] Specifically, the server acquires raw image data of the bridge target area through image acquisition devices (such as drones equipped with high-definition cameras or fixed monitoring cameras); at the same time, it acquires corresponding environmental parameter information through built-in sensors of the acquisition devices or external environmental monitoring modules.

[0026] Raw image data: includes RGB images of the concrete surface of the bridge (beams, piers, bridge deck), with image resolution meeting the detection accuracy requirements, and supports input of single images or continuous frames.

[0027] Environmental parameter information: including light intensity, shooting distance, shooting angle, and atmospheric humidity during shooting, which are used for subsequent image correction and result compensation.

[0028] Data storage and association: The server associates the raw image data with environmental parameter information by timestamp and stores it in the bridge inspection database. The database also stores historical inspection data and bridge structural parameters (such as concrete strength grade and component size) to provide a reference for subsequent result verification.

[0029] In this embodiment, the original data is preprocessed to obtain preprocessed image data, including: Based on the intrinsic and extrinsic parameters of the image acquisition device, distortion correction is performed on the original image to obtain the corrected image; An adaptive median filtering algorithm is used to suppress noise in the corrected image, resulting in preprocessed image data.

[0030] Specifically, preprocessing of raw image data: The server preprocesses the original image to eliminate noise interference and image distortion, providing high-quality input data for feature enhancement and semantic segmentation.

[0031] Image distortion correction: The server corrects the distortion of the original image based on the intrinsic parameters (focal length, pixel size) and extrinsic parameters (shooting angle, distance) of the acquisition device using the perspective transformation formula (1), eliminating the geometric distortion caused by the shooting angle. The specific formula is as follows: (1) Where (X,Y,Z) are the three-dimensional coordinates of the pixels in the original image, and (X',Y') are the two-dimensional coordinates of the pixels after correction; , For camera focal length, , The coordinates of the camera's principal point are obtained in advance through camera calibration.

[0032] Noise suppression: An adaptive median filtering algorithm is used to suppress noise in the corrected image. The size of the filtering window is dynamically adjusted according to the gray-level distribution in the pixel neighborhood. Formula (2) is the window selection rule: (2) in, Let be the standard deviation of gray levels in the neighborhood of pixel (x,y). Using a preset threshold, this method effectively removes Gaussian noise and salt-and-pepper noise while preserving crack details.

[0033] In this embodiment, the preprocessed image data is enhanced to obtain an enhanced feature map, including: Based on the illumination-related information in the environmental parameters, illumination correction is performed on the preprocessed image data to obtain an illumination-equalized image. Multi-scale sampling and edge enhancement are performed on the illumination-equalized image to obtain a multi-scale edge-enhanced image; A pyramid pooling module is used to perform cross-scale fusion on multi-scale edge enhancement images to obtain enhanced feature maps.

[0034] Specifically, multi-scale feature enhancement processing: The server employs a three-level feature enhancement strategy enhanced by an adaptive environment correction multi-scale sampling attention mechanism to highlight crack features, suppress background interference, and generate enhanced feature maps.

[0035] Adaptive illumination correction: Based on the illumination intensity parameters obtained in the previous steps, an improved Retinex algorithm is used for illumination correction to eliminate the influence of uneven illumination on crack characteristics.

[0036] First, convert the image from RGB space to HSV space and separate the luminance channel. Then, the specific brightness channel is adaptively adjusted using formula (3): (3) in, For reference light intensity, The average illumination intensity of the current image is used; finally, the adjusted HSV image is converted back to RGB space to obtain the illumination-equalized image.

[0037] Multi-scale image sampling: The illumination-corrected image is sampled at multiple scales to generate image pyramids of multiple scales, each corresponding to the feature extraction requirements of cracks at different scales; edge enhancement (using the Laplacian operator) is performed on the image at each scale to highlight the crack edge features.

[0038] Feature enhancement guided by attention mechanism: Construct a dual-branch attention module with channel attention and spatial attention to enhance features after multi-scale sampling.

[0039] Channel attention branch: The weights of each feature channel are calculated by squeezing the excitation module. Formula (4) is the weight calculation process: (4) in, This is the result of global average pooling of the feature map. , For the weights of the fully connected layer, It is the ReLU activation function. It is the Sigmoid activation function. This represents the attention weight for each channel.

[0040] Spatial attention branch: Spatial features are extracted through convolution operations, and the spatial attention of each pixel is calculated. Formula (5) is used to calculate the spatial attention weights. (5) in, The result of global max pooling of the feature map. This is a 3×3 convolution operation. Spatial attention weights Feature weighted fusion: The channel attention weights and spatial attention weights are multiplied element-wise with the multi-scale feature map to obtain the enhanced multi-scale feature map.

[0041] Cross-scale feature fusion: The pyramid pooling module is used to fuse enhanced feature maps of multiple scales. Global and local features are extracted through pooling operations of different sizes, and finally enhanced feature maps of a unified scale are generated.

[0042] In this embodiment, an improved semantic segmentation model is used to identify crack regions in the enhanced feature map to obtain a crack segmentation image, including: Construct an improved semantic segmentation model; By using an improved semantic segmentation model to encode and decode the enhanced feature map, a probability map of the crack region is obtained. The probability map is thresholded to convert it into a binary image, and then morphological processing is used to fill holes and connect broken fragments to obtain a crack segmentation image.

[0043] In this embodiment, parameter calculation is performed on the crack segmentation image, including: The crack centerline was obtained using a skeleton extraction algorithm; Sampling points are selected along the center line of the crack, and the vertical width of the crack at each sampling point is calculated to determine the average width and maximum width of the crack. The number of pixels in the crack area is counted, and the actual area of ​​the crack is calculated by combining the conversion relationship between pixel area and actual area.

[0044] Specifically, improve the crack region identification of the semantic segmentation model: Semantic segmentation of the enhanced feature map is performed based on the improved UNet++ model, achieving accurate separation of the crack region from the background.

[0045] Improved semantic segmentation model construction: A bilateral attention module is added to the encoding end of the basic UNet++ to enhance the model's ability to capture crack features; an adaptive feature fusion module is introduced at the decoding end to solve the problem of feature scale mismatch between the encoding and decoding ends; the model loss function adopts a combination loss function of Dice loss + cross-entropy loss, and formula (6) is used to calculate the combination loss: (6) in, These are the weighting coefficients. This is the Dice loss (used to address class imbalance). Cross-entropy loss (to optimize classification accuracy).

[0046] Feature encoding and decoding: The encoding end downsamples the enhanced feature map through multiple convolutional blocks (each convolutional block includes two 3×3 convolutional layers + BN layer + ReLU activation function) to extract deep semantic features; the decoding end fuses the deep semantic features with the shallow detail features of the encoding end through upsampling (transposed convolution) and adaptive feature fusion module to gradually restore the image resolution.

[0047] Crack region prediction and refinement: The model output layer uses the Sigmoid activation function to obtain a probability map of the crack region (pixel value 0~1); a probability threshold is set to convert the probability map into a binary image (crack region is 1, background is 0); morphological processing (dilation and erosion operation) is performed on the binary image to fill the internal holes of the crack and connect the broken crack segments.

[0048] In this embodiment, result verification is performed by combining associated data, including: The calculated crack parameters are compensated and corrected based on environmental parameters; Retrieve historical inspection data for the same area from the bridge inspection database, calculate the deviation between the current inspection results and the historical results, and mark areas where the deviation exceeds the preset range; The reliability of the segmentation results is evaluated using accuracy, recall, and F1 score. If the reliability does not meet the preset standard, the feature enhancement and semantic segmentation process is re-executed.

[0049] Specifically, crack parameter calculation and result verification: The server calculates parameters for the segmented crack areas and verifies the results by combining historical data and environmental parameters to ensure the accuracy of the parameters.

[0050] Crack parameter calculation: Length calculation: The centerline of the crack is obtained using a skeleton extraction algorithm. The actual length of the crack is calculated by converting pixel distance to actual distance (determined based on shooting distance and image resolution).

[0051] Width calculation: Select several sampling points evenly along the center line of the crack, calculate the vertical width of the crack at each sampling point, and take the average value as the average width of the crack; record the maximum width value.

[0052] Area calculation: Count the number of pixels in the crack area, and calculate the actual area of ​​the crack by combining the conversion relationship between pixel area and actual area.

[0053] Result verification: Environmental parameter compensation: Based on the environmental parameters (such as light intensity and humidity) obtained in the previous steps, the calculated crack parameters are compensated and corrected.

[0054] Historical data comparison: retrieve historical inspection data for the same area from the bridge inspection database, calculate the deviation between the current inspection result and the historical results, and automatically mark the area as suspicious if the deviation exceeds a preset threshold, prompting manual review.

[0055] Reliability assessment: The reliability of the segmentation results is assessed using precision, recall, and F1 score, with formulas (7) to (9) as the assessment metrics: (7) (8) (9) Among them, TP is a true positive (correctly identified crack pixels), FP is a false positive (background pixels misidentified as cracks), and FN is a false negative (crack pixels missed). When the F1 score is lower than the preset standard, the feature enhancement and segmentation process is automatically re-executed.

[0056] In this embodiment, the crack detection results are processed to obtain a bridge crack detection report, including: The crack segmentation images, crack parameters, and reliability assessment results are compiled to generate a standardized inspection report that includes basic bridge information, inspection-related information, crack parameters, and maintenance recommendations.

[0057] Specifically, the output includes the recognition results and report generation: The server organizes the crack identification results (segmented images, crack parameters, and reliability assessment results), generates a standardized bridge crack detection report, and pushes it to the bridge maintenance management platform.

[0058] Output formats include: visualized segmented images (with labeled crack areas, lengths, and widths), parameter statistics tables in Excel format, and inspection reports in PDF format; the inspection reports include basic bridge information, inspection time, environmental parameters, crack parameters, maintenance recommendations, etc.

[0059] Data push: The test results are pushed to the bridge maintenance management platform through the API interface, allowing managers to view them in real time; at the same time, the data is stored in the bridge test database, and historical test records are updated to provide data support for long-term maintenance decisions.

[0060] In other embodiments, a bridge crack identification device based on feature enhancement and semantic segmentation is provided, comprising a bridge crack identification method based on any of the preceding embodiments, including: The acquisition module is configured to acquire raw data related to the bridge. The preprocessing module is configured to preprocess the raw data to obtain preprocessed image data; The enhancement processing module is configured to perform enhancement processing on the preprocessed image data to obtain an enhanced feature map; The recognition module is configured to use an improved semantic segmentation model to identify crack regions in the enhanced feature map, thereby obtaining a crack segmentation image; The verification module is configured to calculate parameters on the crack segmentation image and verify the results by combining them with associated data to obtain the crack detection results. The processing module is configured to process the crack detection results and generate a bridge crack detection report.

[0061] In other embodiments, a terminal is provided, including a processor and a memory storing program instructions, the processor being configured to, when executing the program instructions, perform a bridge crack identification method for feature enhancement and semantic segmentation as described above.

[0062] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for identifying bridge cracks using feature enhancement and semantic segmentation, characterized in that, include: Obtain raw data related to the bridge; The original data is preprocessed to obtain preprocessed image data; The preprocessed image data is then enhanced to obtain an enhanced feature map; An improved semantic segmentation model is used to identify crack regions in the enhanced feature map, resulting in a crack segmentation image. The parameters of the crack segmentation image are calculated, and the results are verified by combining the associated data to obtain the crack detection results; The crack detection results are compiled to obtain a bridge crack detection report.

2. The bridge crack identification method based on feature enhancement and semantic segmentation according to claim 1, characterized in that, The acquisition of the original bridge-related data includes: The original image of the target area of ​​the bridge is obtained through image acquisition equipment, and the corresponding environmental parameters are obtained through the built-in sensor of the acquisition equipment or the external environmental monitoring module. The original image and environmental parameters are associated and stored in the bridge detection database according to the correspondence. The bridge detection database stores historical detection data and bridge structural parameters.

3. The bridge crack identification method based on feature enhancement and semantic segmentation according to claim 1, characterized in that, The preprocessing of the original data to obtain preprocessed image data includes: Based on the intrinsic and extrinsic parameters of the image acquisition device, the original image is subjected to distortion correction to obtain the corrected image; An adaptive median filtering algorithm is used to suppress noise in the corrected image, resulting in preprocessed image data.

4. The bridge crack identification method based on feature enhancement and semantic segmentation according to claim 1, characterized in that, The enhancement process performed on the preprocessed image data to obtain an enhanced feature map includes: Based on the illumination-related information in the environmental parameters, illumination correction is performed on the preprocessed image data to obtain an illumination-equalized image. The illumination equalization image is sampled at multiple scales and edge enhancement is performed to obtain a multi-scale edge-enhanced image; The multi-scale edge enhancement image is fused across scales using a pyramid pooling module to obtain an enhanced feature map.

5. The bridge crack identification method based on feature enhancement and semantic segmentation according to claim 1, characterized in that, The step of using an improved semantic segmentation model to identify crack regions in the enhanced feature map to obtain a crack segmentation image includes: Construct an improved semantic segmentation model; The enhanced feature map is encoded and decoded using the improved semantic segmentation model to obtain a crack region probability map. The probability map is thresholded to convert it into a binary image, and then morphological processing is used to fill holes and connect broken segments to obtain a crack segmentation image.

6. The bridge crack identification method based on feature enhancement and semantic segmentation according to claim 1, characterized in that, The parameter calculation for the segmented crack image includes: The crack centerline was obtained using a skeleton extraction algorithm; Sampling points are selected along the center line of the crack, and the vertical width of the crack at each sampling point is calculated to determine the average width and maximum width of the crack. The number of pixels in the crack area is counted, and the actual area of ​​the crack is calculated by combining the conversion relationship between pixel area and actual area.

7. The bridge crack identification method based on feature enhancement and semantic segmentation according to claim 2, characterized in that, The result verification by combining associated data includes: The calculated crack parameters are compensated and corrected based on the environmental parameters. Retrieve historical inspection data for the same area from the bridge inspection database, calculate the deviation between the current inspection results and the historical results, and mark areas where the deviation exceeds the preset range; The reliability of the segmentation results is evaluated using accuracy, recall, and F1 score. If the reliability does not meet the preset standard, the feature enhancement and semantic segmentation process is re-executed.

8. The bridge crack identification method based on feature enhancement and semantic segmentation according to claim 1, characterized in that, The process of organizing the crack detection results to obtain a bridge crack detection report includes: The crack segmentation images, crack parameters, and reliability assessment results are compiled to generate a standardized inspection report that includes basic bridge information, inspection-related information, crack parameters, and maintenance recommendations.

9. A bridge crack identification device based on feature enhancement and semantic segmentation, characterized in that, A bridge crack identification method based on feature enhancement and semantic segmentation according to any one of claims 1 to 8 includes: The acquisition module is configured to acquire raw data related to the bridge. The preprocessing module is configured to preprocess the original data to obtain preprocessed image data; The enhancement processing module is configured to perform enhancement processing on the preprocessed image data to obtain an enhanced feature map; The recognition module is configured to use an improved semantic segmentation model to identify crack regions in the enhanced feature map, thereby obtaining a crack segmentation image; The verification module is configured to calculate parameters on the crack segmentation image and verify the results by combining them with associated data to obtain the crack detection results. The processing module is configured to process the crack detection results to obtain a bridge crack detection report.

10. A terminal, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to, when running the program instructions, execute a bridge crack identification method based on feature enhancement and semantic segmentation as described in any one of claims 1 to 8.

Images