Iron tower base crack measurement evaluation method and device based on instance segmentation

Abstract

The application discloses a kind of tower base crack measurement evaluation method and device based on instance segmentation.The method comprises: obtaining tower base surface image and its image identification, shooting distance, shooting angle and equivalent calibration coefficient, using pre-trained crack instance segmentation model to extract the instance mask, circumscribed frame and confidence of each crack target;According to instance mask, crack center point, pixel length, pixel width estimated value and pixel area are measured, and actual length, actual average width and actual area are converted in combination with shooting condition;Actual size is matched with pre-set crack danger rule base, dangerous star level, trigger reason and processing suggestion are determined, associated identification, measurement and grading result is formed structured crack file.The technical scheme of the application can improve the accuracy of crack identification and measurement under complex background, and can determine the dangerous star level and generate a structured file, which facilitates the sorting, review and tracing of inspection results.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and in particular to a method and apparatus for measuring and evaluating cracks in tower bases based on instance segmentation. Background Technology

[0002] The base of a transmission line tower is a crucial load-bearing component between the tower and the ground. Exposed to the elements for extended periods, it is susceptible to defects such as micro-cracks, through-cracks, spalling, and water seepage due to factors including temperature differences, freeze-thaw cycles, moisture intrusion, material aging, ground disturbance, and maintenance loads. Early-stage cracks typically exhibit low contrast, curved shapes, and discontinuous boundaries. Manual inspections are easily affected by personnel experience, lighting conditions, camera angles, and operator fatigue. Existing crack identification technologies primarily include image preprocessing, machine learning classification, and deep learning instance segmentation. Traditional methods typically involve grayscale conversion, filtering, contrast enhancement, morphological enhancement, threshold segmentation, and connected component screening to identify suspected crack areas. Then, geometric, grayscale, texture, edge density, and principal direction features are extracted, and support vector machines are used for classification.

[0003] Traditional candidate region identification methods are characterized by low computational cost and easy-to-interpret processing, but their results rely on thresholds and morphological parameters. Concrete textures, holes, stains, and shadow edges are easily misclassified as cracks. Support vector machines only classify local candidate regions, making it difficult to identify crack-free backgrounds using the semantic information of the entire image. Deep learning instance segmentation methods can output crack bounding boxes, pixel contours, and confidence scores, demonstrating strong expressive power for bending cracks, multi-crack targets, and complex backgrounds. However, their output usually remains at the image annotation level, lacking geometric measurement, shooting condition correction, actual size estimation, hazard rating, and structured archiving, making it difficult to form sortable, verifiable, and traceable inspection records. Different shooting distances and angles also change the correspondence between pixel size and actual size, affecting the assessment of hazard level.

[0004] One drawback of existing technologies is the potential information gap between the generated candidate crack regions and the actual crack targets. Traditional crack recognition methods mainly extract candidate crack regions based on grayscale, edge, and morphological features. When the tower base surface has rough concrete textures, holes, stain edges, brush marks, or shadow edges, it easily generates a large number of candidate crack regions that deviate from the actual crack targets. Even when using a support vector machine classifier to filter candidate crack regions, it can only compress some false positives and is difficult to recover the complete pixel contour of the actual crack target, thus making it difficult to consistently obtain crack targets that can be geometrically measured. Test results on crack-free negative samples show that open-source computer vision libraries and support vector machines lack the ability to determine the absence of cracks based on the entire tower base surface image. When using black-hat morphological enhancement, threshold segmentation, and morphological closing operations, mortar particles, holes, stain edges, brush marks, and shadow edges are easily magnified into candidate crack regions. Support vector machine (SVM) classifiers only judge local suspected crack candidate regions, making it difficult to form a semantic judgment that no crack targets exist in the entire image of the tower base surface. Therefore, this paper uses an open-source computer vision library and a comparison route with SVM as a means of analyzing false detection causes, and employs the YOLO11m-seg instance segmentation model as the main detection route. This helps reduce false detections in crack-free scenes while preserving instance masks of crack targets.

[0005] The second drawback of existing technologies is that the crack instance segmentation results output by the crack instance segmentation model lack a complete explanatory chain for inspection applications. Although existing deep learning segmentation methods can output instance masks, bounding boxes, and confidence scores for crack targets, the crack instance segmentation results mainly represent the position and pixel contours of crack targets in images of the tower base surface. They cannot directly reflect the center point, pixel length, estimated pixel width, pixel area, and corresponding actual size of the crack target. If the instance mask is not uniformly geometrically measured, if the pixel geometric quantities are not converted into estimated actual length, estimated actual average width, and estimated actual area, and if the hazard level, triggering cause, and handling suggestions are not determined based on the actual size estimation results, then the crack instance segmentation results remain at the image annotation level, making it difficult to form a structured crack archive that can be reviewed, sorted, and traced by inspection personnel.

[0006] The third drawback of existing technologies is the lack of comparability in crack geometry measurement results obtained under different shooting conditions. Images of the tower base surface may be acquired by drones, handheld cameras, or fixed cameras at different shooting distances and angles. The estimated pixel length and pixel width of a crack target of the same actual size will differ in different tower base surface images. If existing methods only output estimated pixel length and pixel width values ​​or bounding box dimensions without considering the shooting distance, shooting angle, and equivalent calibration coefficients to determine the current image scale, they cannot accurately convert pixel geometry into actual size estimates. This also makes it difficult to perform lateral comparisons of crack targets in different tower base surface images, thus affecting the accuracy of hazard rating determination.

[0007] The fourth drawback of existing technologies is the lack of effective feedback between manual verification information and subsequent crack identification processes. Current crack identification methods typically treat false detections as single identification errors, failing to record the corresponding image identifier, target location, and reason for the false detection for each identified crack target. They also fail to convert the surface image and labeled data of the tower base corresponding to the falsely detected crack target into negative or difficult samples. Because the manual verification information is not used to adjust the training samples, confidence thresholds, or preset crack hazard rule base of the crack instance segmentation model, similar interference features such as concrete textures, holes, stain edges, brush marks, and shadow edges are prone to repeatedly causing false detections during subsequent inspections, hindering the continuous improvement of the accuracy of crack identification, hazard classification, and structured archiving.

[0008] How to properly solve the above problems has become an urgent issue for the industry. Summary of the Invention

[0009] This invention provides a method and apparatus for measuring and evaluating cracks in tower bases based on instance segmentation, which improves the accuracy of crack identification and size measurement in complex backgrounds, and enables hazard classification and structured archiving, facilitating inspection review, sorting, and traceability.

[0010] According to a first aspect of the present invention, a method for measuring and evaluating cracks in a steel tower base based on instance segmentation is provided, the method comprising: Acquire the surface image of the iron tower base and the image identifier, shooting distance, shooting angle and equivalent calibration coefficient associated with the surface image of the iron tower base. Input the surface image of the iron tower base into a pre-trained crack instance segmentation model to obtain the instance mask, bounding box and confidence score corresponding to the crack target. Assign target number to each crack target according to the order of the crack targets in the model output results to form crack instance segmentation results. Based on the crack instance segmentation results, geometric measurements are performed on the foreground pixels in each instance mask to obtain the center point, pixel length, pixel width estimate and pixel area corresponding to each crack target; The current image scale is determined based on the shooting distance, shooting angle and equivalent calibration coefficient, and the pixel length, pixel width estimate and pixel area are scaled based on the current image scale to obtain the actual length estimate, actual average width estimate and actual area estimate corresponding to each crack target. The estimated actual length and the estimated actual average width are matched with a preset crack hazard rule base to obtain the hazard star rating, triggering cause, and handling suggestions for each crack target. Based on the image identifier, the segmentation results of the crack instance, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard level, triggering cause, and processing suggestions are associated to generate a structured crack profile.

[0011] According to a second aspect of the present invention, a method for measuring and evaluating cracks in a steel tower base based on instance segmentation is provided, comprising: The acquisition module is used to acquire the surface image of the iron tower base and the image identifier, shooting distance, shooting angle and equivalent calibration coefficient associated with the surface image of the iron tower base. The surface image of the iron tower base is input into a pre-trained crack instance segmentation model to obtain the instance mask, bounding box and confidence score corresponding to the crack target. According to the order of the crack targets in the model output results, each crack target is assigned a target number to form the crack instance segmentation result. The measurement module is used to perform geometric measurements on the foreground pixels in each instance mask based on the crack instance segmentation results, and to obtain the center point, pixel length, pixel width estimate and pixel area corresponding to each crack target; The conversion module is used to determine the current image scale based on the shooting distance, shooting angle and equivalent calibration coefficient, and to perform scale conversion on the estimated values ​​of pixel length, pixel width and pixel area based on the current image scale to obtain the estimated values ​​of actual length, actual average width and actual area corresponding to each crack target. The matching module is used to match the estimated actual length and the estimated actual average width with a preset crack hazard rule base to obtain the hazard star rating, triggering cause and handling suggestions for each crack target; The generation module is used to associate the crack instance segmentation results, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard star rating, triggering cause, and processing suggestions based on the image identifier to generate a structured crack profile.

[0012] According to a third aspect of the present invention, an electronic device is provided, comprising: a communication interface, a processor, and a memory; The memory is used to store program instructions, which, when executed by the processor that is connected to the memory via the communication interface, implement any of the above-described methods for measuring and evaluating cracks in the iron tower base based on instance segmentation.

[0013] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored, which, when executed by a computer (e.g., a processor in the computer), implement any of the above-described methods for measuring and evaluating cracks in tower bases based on instance segmentation.

[0014] In summary, this invention provides a method and apparatus for measuring and evaluating cracks in a steel tower base based on instance segmentation. The method includes: acquiring an image of the steel tower base surface and associated image identifiers, shooting distance, shooting angle, and equivalent calibration coefficients; inputting the steel tower base surface image into a pre-trained crack instance segmentation model to obtain instance masks, bounding boxes, and confidence scores corresponding to crack targets; assigning target numbers to each crack target according to their order in the model output to form crack instance segmentation results; performing geometric measurements on the foreground pixels in each instance mask based on the crack instance segmentation results to obtain the estimated center point, pixel length, pixel width, and pixel area corresponding to each crack target; and based on the shooting distance... The current image scale is determined by distance, shooting angle, and equivalent calibration coefficient. Based on the current image scale, the estimated pixel length, pixel width, and pixel area are scaled to obtain the estimated actual length, estimated actual average width, and estimated actual area for each crack target. The estimated actual length and estimated actual average width are matched with a preset crack hazard rule base to obtain the hazard star rating, triggering cause, and handling suggestion for each crack target. Based on the image identifier, the crack instance segmentation result, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard star rating, triggering cause, and handling suggestion are associated to generate a structured crack file. The technical solution of this invention extracts crack contours through instance segmentation and converts actual dimensions by combining shooting distance, shooting angle, and equivalent calibration coefficient, improving the accuracy of crack identification and measurement in complex backgrounds. The hazard star rating is determined based on the crack size, and a structured file is generated, which facilitates the sorting, review, and traceability of inspection results.

[0015] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and drawings.

[0016] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0017] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 A flowchart of a method for measuring and evaluating cracks in a tower base based on instance segmentation, provided as an embodiment of the present invention; Figure 2 A flowchart of another method for measuring and evaluating cracks in a tower base based on instance segmentation, provided as an embodiment of the present invention; Figure 3 A flowchart of a method for measuring and evaluating cracks in a tower base based on instance segmentation, provided as an embodiment of the present invention; Figure 4 A flowchart of another method for measuring and evaluating cracks in a tower base based on instance segmentation, provided as an embodiment of the present invention; Figure 5 A flowchart of a method for measuring and evaluating cracks in a tower base based on instance segmentation, provided as an embodiment of the present invention; Figure 6 A flowchart of a method for measuring and evaluating cracks in a tower base based on instance segmentation, provided as an embodiment of the present invention; Figure 7 A structural diagram of a tower base crack measurement and evaluation device based on instance segmentation provided for an embodiment of the present invention; Figure 8 A structural diagram of an electronic device provided as an embodiment of the present invention; Figure 9 This is a schematic diagram illustrating the crack identification, measurement, and risk output effects provided in an embodiment of the present invention. Detailed Implementation

[0019] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely intended to explain the present invention and not to limit the present invention. For those skilled in the art, the present invention can be practiced without some of these specific details. The following description of the embodiments is merely to provide a better understanding of the present invention by illustrating examples of the invention.

[0020] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0021] Definitions: YOLO-seg: An instance segmentation model based on the YOLO architecture, used to identify crack targets from images of the base of a steel tower, and output the bounding box, instance mask, target number and confidence score of each crack target, providing input for subsequent geometric measurement, scale conversion and hazard classification.

[0022] Instance mask: A binary or polygonal outline used to represent the specific pixel area occupied by a single crack target in an image of the tower base surface. It is closer to the true shape of the crack target than the bounding box and can be used as the basis for estimating pixel length, pixel width and measuring pixel area.

[0023] Tower base cracks: Linear cracks that form on the surface of concrete or mortar in the tower base. Their occurrence may be related to factors such as temperature changes, material aging, foundation disturbance, moisture intrusion, or external loads.

[0024] Instance segmentation: Simultaneously identify the type, location, and pixel contour of crack targets in the surface image of the iron tower base, and output instance mask, bounding box, and confidence score for each crack target.

[0025] Candidate region: refers to the suspected crack region extracted from the crack enhancement image after threshold segmentation, dark line response, morphological processing and connected component screening. It needs to be further screened by a classifier or preset rules to determine whether it belongs to the crack target.

[0026] Support Vector Machine (SVM) is a supervised machine learning classification model used to determine whether a suspected crack candidate region belongs to a crack target based on the forty-three-dimensional artificial features formed by the geometric features, gray-level statistical features, texture features, edge density, Hu moments, and principal directions of the suspected crack candidate region.

[0027] Black hat morphology: refers to an image processing method that uses the difference between the morphological closing operation result and the original image to highlight thin, dark cracks or dark area structures in a light-colored background. It is used to enhance crack targets and assist in extracting suspected crack candidate regions.

[0028] The bounding box (bbox) is the smallest rectangular box that encloses the target crack region. It is used to characterize the position, width, height, and center point of the crack target in the image of the tower base surface, and provides basic data for crack target localization, result display, and generation of structured crack archives.

[0029] Equivalent calibration coefficient (k0): refers to the actual size calibration parameter corresponding to each pixel in the image of the tower base surface at a shooting distance of one meter. Its value is related to the camera, image resolution, shooting settings and calibration method, and is used to determine the current image scale by combining the shooting distance and shooting angle.

[0030] Scale factor: refers to the actual size of each pixel in the image of the tower base surface under the current shooting conditions. It is calculated by the equivalent calibration factor, shooting distance and shooting angle, and is used to convert the estimated values ​​of pixel length, pixel width and pixel area into estimated values ​​of actual size.

[0031] Current image scale k: refers to the actual size of each pixel in the image of the tower base surface at the current shooting distance and shooting angle. It is calculated by the equivalent calibration coefficient k0, shooting distance d and shooting angle θ according to k=k0·d / cosθ, and is used to convert the pixel geometry into an estimated actual size value.

[0032] like Figure 1 As shown, the present invention provides a method for measuring and evaluating cracks in tower bases based on instance segmentation. This method includes steps S1-S5: In step S1, an image of the tower base surface and the image identifier, shooting distance, shooting angle and equivalent calibration coefficient associated with the image of the tower base surface are obtained. The image of the tower base surface is input into a pre-trained crack instance segmentation model to obtain an instance mask, bounding box and confidence score corresponding to the crack target. According to the order of the crack targets in the model output results, a target number is assigned to each crack target to form a crack instance segmentation result. In step S2, based on the crack instance segmentation results, geometric measurements are performed on the foreground pixels in each instance mask to obtain the center point, pixel length, pixel width estimate and pixel area corresponding to each crack target; In step S3, the current image scale is determined based on the shooting distance, shooting angle and equivalent calibration coefficient, and the pixel length, pixel width estimate and pixel area are scaled based on the current image scale to obtain the actual length estimate, actual average width estimate and actual area estimate corresponding to each crack target. In step S4, the estimated actual length and the estimated actual average width are matched with a preset crack hazard rule base to obtain the hazard star rating, triggering cause, and handling suggestions for each crack target. In step S5, based on the image identifier, the crack instance segmentation result, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard star rating, triggering cause, and processing suggestions are associated to generate a structured crack profile.

[0033] In one embodiment, images of the tower base surface are acquired, and a unique image identifier is assigned to each image. Simultaneously, the corresponding shooting distance, shooting angle, and equivalent calibration coefficients are read. The equivalent calibration coefficients are pre-determined using a calibration plate, a reference object of known size on-site, or a fixed shooting position. The tower base surface images are input into a pre-trained YOLO11m-seg instance segmentation model. The crack instance segmentation model identifies each crack target and outputs an instance mask, bounding box, and confidence score for each crack target. Based on the order of the crack targets in the model output, target numbers are assigned to each crack target, forming the crack instance segmentation result.

[0034] For each instance mask in the crack instance segmentation result, the corresponding foreground pixel coordinate set is extracted. The bounding box is determined based on the foreground pixel coordinate set. The coordinates of all foreground pixels in the instance mask are extracted. The average of the foreground pixel x-coordinates is used as the x-coordinate of the center point, and the average of the foreground pixel y-coordinates is used as the y-coordinate of the center point, thus determining the center point of the crack target. When the instance mask is empty and the bounding box is used as the backtracking measurement object, the geometric center of the bounding box is determined as the center point. Principal component analysis is performed based on the foreground pixel coordinate set to determine the principal direction and secondary direction. The centered foreground pixel coordinates are projected onto the principal direction, and the projection span in the principal direction is calculated. The width and height of the bounding box are obtained. The maximum value among the projection span in the principal direction, the width and height of the bounding box, and a single pixel is determined as the pixel length. The number of foreground pixels in the instance mask is counted, and the result is used as the pixel area. The ratio of the pixel area to the pixel length is calculated, and this ratio is used as the estimated pixel width. This geometric measurement method makes the measurement results correspond to the pixel contour of the crack target, reducing the deviation caused by measuring only based on the bounding box.

[0035] Read the shooting distance d, shooting angle θ, and equivalent calibration coefficient k0 corresponding to the image of the tower base surface. Here, the equivalent calibration coefficient k0 represents the actual size of each pixel at a shooting distance of one meter. Calculate the current image scale k according to k=k0·d / cosθ, and then calculate the scale according to L=l px ·k、W= w px ·k and A= ·k², for pixel length l px Pixel width estimate w px and pixel area Scale conversion is performed to obtain the estimated actual length L, the estimated actual average width W, and the estimated actual area A. The equivalent calibration coefficients are related to the camera, image resolution, shooting settings, and calibration method. If significant perspective distortion exists, the estimated actual dimensions are verified on-site.

[0036] The estimated actual length and the estimated actual average width are matched with the preset crack hazard rule library, as shown in Table 1 below.

[0037] Table 1

[0038] When the estimated actual average width is not less than 1.0 mm or the estimated actual length is not less than 300 mm, the hazard rating is four stars; if the four-star condition is not met, but the estimated actual average width is not less than 0.3 mm or the estimated actual length is not less than 150 mm, the hazard rating is three stars; if neither the four-star nor the three-star condition is met, but the estimated actual average width is not less than 0.2 mm or the estimated actual length is not less than 50 mm, the hazard rating is two stars; all other crack targets have a hazard rating of one star. Triggering causes are generated based on the triggered width and length conditions, and processing recommendations are generated based on the hazard rating. Based on image identification, the crack instance segmentation results, geometric measurement results, actual size estimation results, hazard rating, triggering causes, and processing recommendations are correlated to generate a structured crack profile for crack target sorting, review, and tracing. This hazard rating is used for review priority ranking and does not replace the final structural safety assessment. The crack identification, measurement, and risk output effects in a preferred embodiment are shown in the attached figure. Figure 9As shown in the figure, the identification, measurement, and hazard classification results of two crack targets in an image of the tower base surface are illustrated. Each crack target is labeled with the instance mask outline, bounding box, center point, target number, pixel length, estimated pixel width, estimated actual length, estimated actual average width, actual size, pixel size, and hazard star rating. Target 1 (the upper crack) has an estimated actual length of approximately 11.6 mm and an estimated actual average width of approximately 0.90 mm, and is classified as a three-star target. Target 2 (the lower crack) has an estimated actual length of approximately 12.5 mm and an estimated actual average width of approximately 1.21 mm, and is classified as a four-star target because its width reaches the four-star threshold. Corresponding triggering reasons and handling suggestions are generated. (Attached) Figure 9 The instance mask outlines are marked with lines; the bounding boxes are marked with square outlines; the center points are marked with "+" signs in the image; the symbol # represents the target number, such as #1 representing crack target number 1; the symbol L represents the estimated actual pixel length, such as L=11.6mm representing the estimated actual pixel length of 11.6 mm; the symbol W represents the estimated actual pixel width, such as =0.90mm representing the estimated actual pixel width of 0.90 mm; px represents pixels, such as L=433.9 W=33.61 representing a pixel length of 433.9 and an estimated pixel width of 33.61; the symbol pos represents the area, such as pos=(254,192) representing an actual size of 254 and a pixel size of 192; the symbol risk represents the hazard level, such as risk= The danger level is represented by 3 stars.

[0039] In other embodiments, the crack instance segmentation model is not limited to the YOLO11m-seg instance segmentation model. Mask R-CNN, DeepLab, U-Net, SegFormer, or other segmentation networks capable of outputting instance masks, bounding boxes, and confidence scores for crack targets can also be used. The resulting crack instance segmentation results can then be used for geometric measurements, scale conversion, hazard classification, and the generation of structured crack profiles. The identification route is not limited to support vector machine classifiers; random forests, gradient boosting trees, K-nearest neighbors classifiers, lightweight neural networks, or other binary classification models can also be used. Artificial features of suspected crack candidate regions can also include gray-level co-occurrence matrix features, directional gradient histogram features, or skeleton features. Equivalent calibration coefficients can be determined based on the camera field of view, calibration board, on-site known size reference objects, fixed camera pose, structured light measurement results, or 3D reconstruction results. When acquiring images of the tower base surface at a fixed workstation, calibration tables can be established for different workstations. The width thresholds, length thresholds, and treatment suggestions in the preset crack hazard rule base can be adjusted based on the company's operation and maintenance procedures, concrete environment category, tower importance level, regional climate conditions, and on-site verification results. These thresholds are not limited to 0.2mm, 0.3mm, 1.0mm, and their corresponding length thresholds. Structured crack files are not limited to comma-separated value files; they can also be saved or exported using spreadsheets, database tables, inspection management interfaces, web reports, or mobile verification task sheets. However, at least the image identifier, target number, target location, actual size estimation result, hazard star rating, and treatment suggestions should be retained.

[0040] The main implementation files in this embodiment include the training weights, evaluation results, and crack instance segmentation results of the YOLO11m-seg instance segmentation model; the size measurement program tools / crack_measurement.py; the preset crack risk rule file crack_risk_standard.csv; the image metadata file capture_metadata.csv; and the structured crack archive file crack_archive.csv. The image preprocessing, suspected crack candidate region extraction, manual feature extraction, and support vector machine classifier training programs corresponding to the open-source computer vision library and support vector machine comparison route are retained as comparison data for traditional methods. These are used to analyze the sources of false detections in traditional candidate region recognition methods and to verify the technical effectiveness of the main detection route of this invention. Experimental results show that after 100 training epochs, the YOLO11m-seg instance segmentation model achieves an optimal average precision of 0.7921 for bounding boxes with an intersection-union (IU) threshold of 0.5, 0.6616 for instance masks with an IU threshold of 0.5, and 0.2219 for instance masks with IU thresholds ranging from 0.5 to 0.95. The model outputs 30 crack targets from 26 evaluation images, generating 30 structured crack profiles and corresponding processing suggestions. In the control experiment, the support vector machine (SVM) classifier compressed 292 initial suspected crack candidate regions into 28 final suspected crack candidate regions. Furthermore, the open-source computer vision library and SVM control route showed significantly more false detections than the YOLO11m-seg main detection route in negative samples without cracks.

[0041] The technical solution in this embodiment provides a visual recognition method for tower base cracks based on YOLO11m-seg instance segmentation, and uses the traditional machine vision / SVM route as a comparison and verification route. After the YOLO11m-seg instance segmentation model outputs the instance mask, pixel geometric measurement, shooting distance and angle correction, hazard star rules and traceable file output are added, thereby forming a crack inspection result that can be verified, sorted and iteratively optimized.

[0042] In one embodiment, such as Figure 2 As shown, in step S1, the crack instance segmentation result is obtained, including the following steps S11-S14: In step S11, the image of the tower base surface is subjected to grayscale conversion, noise reduction, contrast enhancement, sharpening, and multi-scale black hat morphological enhancement to obtain a crack-enhanced image. In step S12, the image of the tower base surface is input into the YOLO11m-seg instance segmentation model. The YOLO11m-seg instance segmentation model outputs the instance mask, bounding box and confidence score of each crack target. Target numbers are assigned to each crack target according to the output order to obtain the main detection result. In step S13, the enhanced crack image is subjected to threshold segmentation, dark line response processing, morphological closing operation, connected component filtering, and candidate box filtering to obtain suspected crack candidate regions. The geometric features, gray-level statistical features, texture features, edge density, Hu moments, and principal directions of the suspected crack candidate regions are extracted to form forty-three-dimensional artificial features. The forty-three-dimensional artificial features are then input into a support vector machine classifier to obtain the comparison recognition results. In step S14, the candidate boxes or candidate support masks in the comparison identification results are converted into the same geometric measurement format as the main detection results. The comparison identification results are compared with the main detection results according to the target position and the target quantity to obtain comparison information used to characterize false detections caused by concrete texture, holes, stain edges, brush marks or shadow edges.

[0043] In one embodiment, an image of the tower base surface is acquired, and the corresponding image identifier is retained. The tower base surface image is then subjected to grayscale conversion and denoising to reduce interference from color differences and random noise. The denoised tower base surface image is then subjected to contrast enhancement and sharpening to enhance the grayscale difference between the cracks and the concrete background. Multi-scale black hat morphological enhancement is used to highlight the slender, dark cracks in the light-colored background, resulting in an enhanced crack image. This enhanced crack image is used for extracting suspected crack candidate regions and serves as an auxiliary verification basis for the main detection results.

[0044] The surface image of the tower base is input into a pre-trained YOLO11m-seg instance segmentation model. The YOLO11m-seg model identifies crack instances across the entire tower base surface image, outputting the instance mask, bounding box, and confidence score for each crack target. Target numbers are assigned to each crack target according to the output order, forming the main detection result. Thresholding segmentation, dark line response processing, and morphological closing operations are performed on the enhanced crack image. The resulting connected components are then filtered, and candidate boxes are selected to obtain suspected crack candidate regions. For each suspected crack candidate region, geometric features, grayscale statistical features, texture features, edge density, Hu moments, and principal directions are extracted to form forty-three-dimensional artificial features. These forty-three-dimensional artificial features are input into a pre-trained support vector machine classifier to obtain the comparison recognition result.

[0045] The candidate bounding boxes or candidate support masks in the comparison identification results are converted into the same geometric measurement format as the main detection results. A correspondence is established between the comparison identification results and the main detection results according to the target location, and the target locations and number of targets are compared. For suspected crack candidate regions present in the comparison identification results but not confirmed by the main detection results, the crack enhancement image and the tower base surface image are combined to determine whether they originate from concrete textures, holes, stain edges, brush marks, or shadow edges, generating corresponding comparison information. This comparison information is used to explain the reasons for false detections in traditional candidate region extraction and is not used to replace the main detection results.

[0046] In one embodiment, such as Figure 3 As shown, in step S2, the estimated values ​​of the center point, pixel length, pixel width, and pixel area are obtained, including the following steps S21-S23: In step S21, the set of foreground pixel coordinates in each instance mask is extracted, and the bounding box and center point of the corresponding crack target are determined based on the set of foreground pixel coordinates. In step S22, the main direction of the corresponding crack target is determined according to the foreground pixel coordinate set, the foreground pixel coordinate set is projected onto the main direction, and the pixel length is determined according to the projection range of the foreground pixel coordinate set on the main direction; In step S23, the number of foreground pixels in the instance mask is counted to obtain the pixel area; the ratio of pixel area to pixel length is calculated to obtain the average area width; the average area width is constrained by an upper limit based on the projection span in the secondary direction, and a lower limit of one pixel is set to obtain the estimated pixel width value.

[0047] In one embodiment, for each instance mask in the crack instance segmentation result, the pixels within the instance mask used to characterize the crack target are determined as foreground pixels. The coordinates of each foreground pixel in the image coordinate system are extracted to form a foreground pixel coordinate set. The range of values ​​for the horizontal and vertical coordinates in the foreground pixel coordinate set is determined, and the bounding box of the corresponding crack target is defined by the minimum, maximum, minimum, and maximum values ​​of the horizontal and vertical coordinates. The geometric center of the bounding box is determined as the center point. The instance mask can characterize the specific pixel contour of the crack target. Using the instance mask as the geometric measurement object can reduce the influence of the non-crack background within the bounding box on the measurement results.

[0048] The principal direction of the corresponding crack target is determined based on the overall extension direction of the foreground pixel coordinate set. Each foreground pixel coordinate is projected onto the principal direction, and the maximum and minimum values ​​of the projected coordinates are obtained. The difference between the maximum and minimum values ​​is determined as the pixel length. The number of foreground pixels within the instance mask is counted, and the result is determined as the pixel area. The ratio of the pixel area to the pixel length is calculated, and this ratio is determined as the estimated pixel width. For tilted or curved crack targets, using principal direction projection to determine the pixel length and using the ratio of pixel area to pixel length to determine the estimated pixel width avoids measurement deviations caused by directly using the bounding box length and width.

[0049] In one embodiment, such as Figure 4 As shown, in step S3, the estimated values ​​of the actual length, the actual average width, and the actual area are obtained, including the following steps S31-S33: In step S31, the shooting distance d, shooting angle θ, and equivalent calibration coefficient k0 of the actual size of each pixel at a shooting distance of one meter are obtained for the image of the tower base surface. The current image scale k is calculated according to the following formula.

[0050] In step S32, based on the current image scale k and pixel length l... px Pixel width estimate w px and pixel area The calculation is performed according to the following formula:

[0051] In step S33, the estimated actual length L, the estimated actual average width W, and the estimated actual area A are calculated.

[0052] In one embodiment, the shooting distance d, shooting angle θ, and equivalent calibration coefficient k0 corresponding to the image of the tower base surface are obtained. The equivalent calibration coefficient k0 represents the actual size of each pixel at a shooting distance of one meter, and can be predetermined through camera calibration, a calibration plate, a reference object with known dimensions on site, or a fixed camera pose. The shooting angle θ is used to characterize the overall oblique shooting degree of the tower base surface image. Since the pixel size corresponding to the same actual size decreases with increasing shooting distance, and oblique shooting causes cosine projection compression of the target size, the current image scale k is calculated according to Formula 1. This current image scale k represents the actual size of each pixel under the current shooting conditions.

[0053] Read the pixel length l of the corresponding crack target px Pixel width estimate w px and pixel area Following formula 2, scale conversion is performed to obtain the estimated actual length L, the estimated actual average width W, and the estimated actual area A. The estimated pixel length and pixel width are one-dimensional pixel geometric quantities, which are multiplied by the current image scale k respectively; the pixel area is a two-dimensional pixel geometric quantity, and scale conversion is required in both directions, so it is multiplied by the square of the current image scale k.

[0054] The equivalent calibration coefficient k0 is not a universal constant; its value is related to the camera, image resolution, shooting settings, and calibration method. After changing the shooting device or adjusting the image resolution, the equivalent calibration coefficient k0 should be re-determined. When there is significant lens distortion, perspective distortion, curved surfaces, or local undulations, distortion correction should be performed on the image of the tower base surface, and the estimated actual dimensions should be verified in conjunction with on-site rulers, photogrammetry, or on-site measurement results.

[0055] In one embodiment, such as Figure 5 As shown, in step S4, the danger level, triggering cause, and handling recommendations are obtained, including the following steps S41-S44: In step S41, the width threshold, length threshold and processing suggestions corresponding to four-star, three-star, two-star and one-star are read from the preset crack hazard rule library, and the actual length estimate L and the actual average width estimate W of the crack target to be classified are obtained. In step S42, the cracks are matched according to their degree of danger from high to low. When W ≥ 1.0 mm or L ≥ 300 mm, the crack is designated as a four-star crack. When the four-star condition is not met but W ≥ 0.3 mm or L ≥ 150 mm, the crack is designated as a three-star crack. When the four-star and three-star conditions are not met but W ≥ 0.2 mm or L ≥ 50 mm, the crack is designated as a two-star crack. When the four-star, three-star, and two-star conditions are not met, the crack is designated as a one-star crack. In step S43, the preset triggering reason in the hit danger rule is read, and the preset triggering reason is used as the triggering reason for the crack target to be graded; In step S44, the handling recommendations are generated based on the hazard star rating. Four stars correspond to immediate professional inspection and development of a repair plan; three stars correspond to professional review and inclusion in key tracking; two stars correspond to record and file and periodic retesting; and one star corresponds to routine observation.

[0056] In one embodiment, a crack hazard rule base is pre-established, recording the width threshold, length threshold, and processing suggestions corresponding to four-star, three-star, two-star, and one-star cracks, respectively. The estimated actual length L and the estimated actual average width W of the target crack to be classified are obtained and input into the crack hazard rule base for matching. During the matching process, the width and length thresholds are determined by a trigger mechanism where any condition reaching the corresponding threshold is used. A unique hazard star rating is determined according to the degree of danger from high to low, preventing the same target crack from being repeatedly classified into multiple hazard star ratings.

[0057] When the estimated actual average width W is not less than 1.0 mm, or the estimated actual length L is not less than 300 mm, the crack target to be classified is determined as a four-star crack. If the four-star condition is not met, and the estimated actual average width W is not less than 0.3 mm, or the estimated actual length L is not less than 150 mm, the crack target to be classified is determined as a three-star crack. If neither the four-star nor the three-star condition is met, and the estimated actual average width W is not less than 0.2 mm, or the estimated actual length L is not less than 50 mm, the crack target to be classified is determined as a two-star crack. If neither the four-star, three-star, nor two-star condition is met, the crack target to be classified is determined as a one-star crack. The length threshold is used as an auxiliary condition for batch inspection sorting and hazard star rating upgrades. The hazard star rating is used to indicate the review priority of the crack target to be classified and does not replace the final structural safety assessment.

[0058] Triggering causes are generated based on the threshold conditions met by the target crack to be classified. When only the estimated actual average width reaches the corresponding width threshold, the width condition is recorded as a triggering cause; when only the estimated actual length reaches the corresponding length threshold, the length condition is recorded as a triggering cause; when both the estimated actual average width and the estimated actual length reach their respective thresholds, both the width and length conditions are recorded as triggering causes. Based on the determined hazard level, handling recommendations are generated: four-star corresponds to immediate professional inspection and a repair plan; three-star corresponds to professional review and inclusion in key monitoring; two-star corresponds to filing and regular retesting; and one-star corresponds to routine observation. The thresholds in the crack hazard rule base can be adjusted based on the company's operation and maintenance procedures, concrete environment category, tower importance level, regional climate, and on-site review results.

[0059] In one embodiment, such as Figure 6 As shown, in step S5, the structured crack profile is generated, including the following steps S51-S54: In step S51, a crack file record is established for each crack target, and the image identifier, target number, bounding box, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, shooting distance, shooting angle, confidence level, hazard star rating, triggering cause and processing suggestions are written into the crack file record. In step S52, the crack file records are sorted according to the hazard star rating, the estimated actual length, and the estimated actual average width to obtain the manual review order of the crack targets; In step S53, the manual review information corresponding to the crack file record is obtained, including the image identifier, target location, and reason for false detection for the crack target record that is confirmed as a false detection. In step S54, the images and labeled data corresponding to the crack targets that are identified as falsely detected are converted into negative samples or hard samples, and at least one of the training samples, confidence threshold and preset crack hazard rule base of the crack instance segmentation model is adjusted according to the manual review information.

[0060] In one embodiment, a crack profile record is established for each crack target in the crack instance segmentation results. The image identifier, target number, bounding box, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, shooting distance, shooting angle, confidence level, hazard rating, triggering cause, and processing recommendations are written into the corresponding crack profile record. The crack profile record is stored using structured fields to maintain correlation between the identification results, geometric measurement results, actual size estimation results, and hazard classification results of the same crack target, facilitating the querying of the crack target's location, size, and processing recommendations.

[0061] Crack records are sorted according to their hazard rating, estimated actual length, and estimated actual average width. Crack records with higher hazard ratings are placed first. The estimated actual length and estimated actual average width are used as auxiliary sorting criteria within the same hazard rating to determine the manual review order for crack targets. Inspection personnel review the corresponding tower base surface image, instance mask, and bounding box according to this manual review order, and write the manual review information into the corresponding crack record. For crack targets identified as false positives, their image identifier, target location, and reason for false positive are recorded. Reasons for false positives include those caused by concrete texture, holes, stain edges, brush marks, or shadow edges.

[0062] The tower base surface images and labeled data corresponding to crack targets identified as falsely detected are converted into negative samples or hard samples. The training samples for the crack instance segmentation model are supplemented based on the manually reviewed information, or the confidence threshold of the crack instance segmentation model is adjusted, or the width threshold, length threshold, and processing suggestions in the preset crack hazard rule base are revised. This allows the manually reviewed information to be used for subsequent model training and rule adjustments, reducing the probability of similar false detections recurring in subsequent inspections.

[0063] In one embodiment, Figure 7 This is a block diagram illustrating an instance-segmentation-based tower base crack measurement and evaluation device according to an exemplary embodiment. Figure 7 As shown, the tower base crack measurement and evaluation device based on instance segmentation includes an acquisition module 71, a measurement module 72, a conversion module 73, a matching module 74, and a generation module 75.

[0064] The acquisition module 71 is used to acquire an image of the tower base surface and an image identifier, shooting distance, shooting angle and equivalent calibration coefficient associated with the image of the tower base surface. The image of the tower base surface is input into a pre-trained crack instance segmentation model to obtain an instance mask and bounding box containing the crack target. According to the order of the crack targets in the model output results, a target number is assigned to each crack target to form a crack instance segmentation result with confidence. The measurement module 72 is used to perform geometric measurements on the foreground pixels in each instance mask based on the crack instance segmentation results, and to obtain the center point, pixel length, pixel width estimate and pixel area corresponding to each crack target. The conversion module 73 is used to determine the current image scale based on the shooting distance, shooting angle and equivalent calibration coefficient, and to perform scale conversion on the estimated values ​​of pixel length, pixel width and pixel area based on the current image scale to obtain the estimated values ​​of actual length, actual average width and actual area corresponding to each crack target. The matching module 74 is used to match the estimated actual length and the estimated actual average width with a preset crack hazard rule library to obtain the hazard star rating, triggering cause and handling suggestion for each crack target; The generation module 75 is used to associate the crack instance segmentation result, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard star rating, triggering cause, and processing suggestions based on the image identifier to generate a structured crack profile.

[0065] The block diagram of the tower base crack measurement and evaluation device based on instance segmentation includes the acquisition module 71, the measurement module 72, the conversion module 73, the matching module 74, and the generation module 75, which are controlled to execute the tower base crack measurement and evaluation method based on instance segmentation described in any of the above embodiments.

[0066] like Figure 8 As shown, the present invention provides an electronic device 800, which includes: a communication interface, a processor 801, and a memory 802; The memory 802 stores program instructions. When executed by the processor 801, which is connected to the memory 802 via the communication interface, the program instructions acquire an image of the tower base surface, along with image identifiers, shooting distances, shooting angles, and equivalent calibration coefficients associated with the image. The image is then input into a pre-trained crack instance segmentation model to obtain instance masks, bounding boxes, and confidence scores corresponding to the crack targets. Based on the order of the crack targets in the model output, target numbers are assigned to each crack target, forming a crack instance segmentation result. Based on the crack instance segmentation result, geometric measurements are performed on the foreground pixels in each instance mask to obtain the estimated center point, pixel length, pixel width, and pixel position of each crack target. The area is determined based on the shooting distance, shooting angle, and equivalent calibration coefficient. The estimated pixel length, pixel width, and pixel area are then scaled based on the current image scale to obtain the estimated actual length, estimated actual average width, and estimated actual area for each crack target. The estimated actual length and estimated actual average width are matched with a preset crack hazard rule base to obtain the hazard star rating, triggering cause, and processing suggestion for each crack target. Based on the image identifier, the crack instance segmentation result, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard star rating, triggering cause, and processing suggestion are associated to generate a structured crack profile.

[0067] This invention provides a computer-readable storage medium storing computer program instructions. When executed by a processor, the computer program instructions acquire an image of the surface of a steel tower base, along with image identifiers, shooting distances, shooting angles, and equivalent calibration coefficients associated with the image. The image is then input into a pre-trained crack instance segmentation model to obtain instance masks, bounding boxes, and confidence scores corresponding to the crack targets. Based on the order of the crack targets in the model output, target numbers are assigned to each crack target, forming a crack instance segmentation result. Based on the crack instance segmentation result, geometric measurements are performed on the foreground pixels in each instance mask to obtain the estimated center point, pixel length, pixel width, and pixel area corresponding to each crack target. The current image scale is determined based on the shooting distance, shooting angle, and equivalent calibration coefficient. Based on this current image scale, the estimated pixel length, pixel width, and pixel area are scaled to obtain the estimated actual length, estimated actual average width, and estimated actual area for each crack target. The estimated actual length and estimated actual average width are matched with a preset crack hazard rule base to obtain the hazard star rating, triggering cause, and processing suggestion for each crack target. Based on the image identifier, the crack instance segmentation result, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard star rating, triggering cause, and processing suggestion are associated to generate a structured crack profile.

[0068] It should be understood that the specific features, operations, and details described above regarding the method of the present invention can also be similarly applied to the apparatus and system of the present invention, or vice versa. Furthermore, each step of the method of the present invention described above can be performed by a corresponding component or unit of the apparatus or system of the present invention.

[0069] It should be understood that the various modules / units of the device of the present invention can be implemented wholly or partially through software, hardware, firmware, or a combination thereof. Each module / unit can be embedded in the processor of a computer device in hardware or firmware form or independent of the processor, or it can be stored in the memory of a computer device in software form for the processor to call to execute the operation of each module / unit. Each module / unit can be implemented as an independent component or module, or two or more modules / units can be implemented as a single component or module.

[0070] In one embodiment, a computer device is provided, including a memory and a processor. The memory stores computer instructions executable by the processor, which, when executed by the processor, instruct the processor to perform steps of the methods of embodiments of the present invention. The computer device can be broadly categorized as a server, terminal, or any other electronic device with the necessary computing and / or processing capabilities. In one embodiment, the computer device may include a processor, memory, network interface, communication interface, etc., connected via a system bus. The processor of the computer device can be used to provide the necessary computing, processing, and / or control capabilities. The memory of the computer device may include a non-volatile storage medium and internal memory. The non-volatile storage medium may store an operating system, computer programs, etc. The internal memory can provide an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The network interface and communication interface of the computer device can be used to connect and communicate with external devices via a network. When the computer program is executed by the processor, it performs the steps of the methods of the present invention.

[0071] This invention can be implemented as a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, causes the steps of the methods of embodiments of the invention to be performed. In one embodiment, the computer program is distributed across multiple network-coupled computer devices or processors, such that the computer program is stored, accessed, and executed in a distributed manner by one or more computer devices or processors. A single method step / operation, or two or more method steps / operations, may be executed by a single computer device or processor or by two or more computer devices or processors. One or more method steps / operations may be executed by one or more computer devices or processors, and one or more other method steps / operations may be executed by one or more other computer devices or processors. One or more computer devices or processors may execute a single method step / operation, or execute two or more method steps / operations.

[0072] It will be understood by those skilled in the art that the method steps of the present invention can be performed by a computer program instructing related hardware, such as a computer device or processor. The computer program may be stored in a non-transitory computer-readable storage medium, and its execution causes the steps of the present invention to be performed. Depending on the context, any references herein to memory, storage, databases, or other media may include non-volatile and / or volatile memory. Examples of non-volatile memory include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), flash memory, magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid-state drive, etc. Examples of volatile memory include random access memory (RAM), external cache memory, etc.

[0073] The technical features described above can be combined arbitrarily. Although not all possible combinations of these technical features are described, any combination of these technical features should be considered to be covered by this specification, provided that such combination does not contain contradictions.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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; and these 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.

Claims

1. A method for measuring and evaluating cracks in a steel tower base based on instance segmentation, characterized in that, include: Acquire the surface image of the iron tower base and the image identifier, shooting distance, shooting angle and equivalent calibration coefficient associated with the surface image of the iron tower base. Input the surface image of the iron tower base into a pre-trained crack instance segmentation model to obtain the instance mask, bounding box and confidence score corresponding to the crack target. Assign target number to each crack target according to the order of the crack targets in the model output results to form crack instance segmentation results. Based on the crack instance segmentation results, geometric measurements are performed on the foreground pixels in each instance mask to obtain the center point, pixel length, pixel width estimate and pixel area corresponding to each crack target; The current image scale is determined based on the shooting distance, shooting angle and equivalent calibration coefficient, and the pixel length, pixel width estimate and pixel area are scaled based on the current image scale to obtain the actual length estimate, actual average width estimate and actual area estimate corresponding to each crack target. The estimated actual length and the estimated actual average width are matched with a preset crack hazard rule base to obtain the hazard star rating, triggering cause, and handling suggestions for each crack target. Based on the image identifier, the segmentation results of the crack instance, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard level, triggering cause, and processing suggestions are associated to generate a structured crack profile.

2. The method for measuring and evaluating cracks in tower bases based on instance segmentation as described in claim 1, characterized in that, The crack instance segmentation result is obtained, including: The image of the iron tower base surface is subjected to grayscale conversion, noise reduction, contrast enhancement, sharpening, and multi-scale black hat morphological enhancement to obtain a crack-enhanced image. The image of the tower base surface is input into the YOLO11m-seg instance segmentation model. The YOLO11m-seg instance segmentation model outputs the instance mask, bounding box and confidence score of each crack target. The target number is assigned to each crack target according to the output order to obtain the main detection result. The enhanced crack image is subjected to threshold segmentation, dark line response processing, morphological closing operation, connected component filtering, and candidate box filtering to obtain suspected crack candidate regions. The geometric features, gray-level statistical features, texture features, edge density, Hu moments, and principal directions of the suspected crack candidate regions are extracted to form forty-three-dimensional artificial features. The forty-three-dimensional artificial features are then input into a support vector machine classifier to obtain the comparison recognition results. The candidate boxes or candidate support masks in the comparison identification results are converted into the same geometric measurement format as the main detection results. The comparison identification results are compared with the main detection results according to the target position and the target quantity to obtain comparison information used to characterize false detections caused by concrete texture, holes, stain edges, brush marks or shadow edges.

3. The method for measuring and evaluating cracks in tower bases based on instance segmentation as described in claim 1, characterized in that, Obtaining the center point, pixel length, estimated pixel width, and pixel area includes: Extract the set of foreground pixel coordinates from each instance mask, and determine the bounding box and center point of the corresponding crack target based on the set of foreground pixel coordinates; The main direction of the corresponding crack target is determined based on the set of foreground pixel coordinates, the set of foreground pixel coordinates is projected onto the main direction, and the pixel length is determined based on the projection range of the set of foreground pixel coordinates on the main direction. The number of foreground pixels in the instance mask is counted to obtain the pixel area, and the ratio of the pixel area to the pixel length is determined as the estimated pixel width.

4. The method for measuring and evaluating cracks in tower bases based on instance segmentation as described in claim 1, characterized in that, Obtaining the estimated actual length, estimated actual average width, and estimated actual area includes: Obtain the shooting distance d, shooting angle θ, and equivalent calibration coefficient k0 of the actual size corresponding to each pixel at a shooting distance of one meter for the image of the iron tower base surface; The current image scale k is calculated according to the following formula; Wherein, the current image scale k represents the actual size corresponding to each pixel under the current shooting conditions; Based on the current image scale k and pixel length l px Pixel width estimate w px and pixel area The calculation is performed according to the following formula: The estimated values ​​of actual length L, actual average width W, and actual area A are calculated.

5. The method for measuring and evaluating cracks in tower bases based on instance segmentation as described in claim 1, characterized in that, The danger level, triggering cause, and handling recommendations are obtained, including: Read the width threshold, length threshold and processing suggestions corresponding to four-star, three-star, two-star and one-star respectively from the preset crack hazard rule base, and obtain the actual length estimate L and the actual average width estimate W of the crack target to be classified. The cracks are matched according to their degree of danger from high to low. When W ≥ 1.0 mm or L ≥ 300 mm, the crack is designated as a four-star crack. When the four-star condition is not met but W ≥ 0.3 mm or L ≥ 150 mm, the crack is designated as a three-star crack. When the four-star and three-star conditions are not met but W ≥ 0.2 mm or L ≥ 50 mm, the crack is designated as a two-star crack. When the four-star, three-star, and two-star conditions are not met, the crack is designated as a one-star crack. Read the preset triggering reasons in the hit danger rules, and use the preset triggering reasons as the triggering reasons for the crack target to be graded; Based on the aforementioned hazard rating, the proposed treatment recommendations are generated. For four-star ratings, professional testing should be arranged immediately and a repair plan should be developed. For three-star ratings, professional review should be arranged and the area should be included in key monitoring. For two-star ratings, records should be filed and regular retests should be conducted. For one-star ratings, routine observation should be carried out.

6. The method for measuring and evaluating cracks in tower bases based on instance segmentation as described in claim 1, characterized in that, Generating the structured crack profile includes: For each crack target, a crack file record is established, which includes the image identifier, target number, bounding box, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, shooting distance, shooting angle, confidence level, hazard level, triggering cause, and handling suggestions. According to the aforementioned hazard star rating, the estimated actual length, and the estimated actual average width, the records of each crack file are sorted to obtain the manual review order of the crack targets; Obtain the manual review information corresponding to the crack file record, including the image identifier, target location, and reason for false detection for the crack target record that is confirmed as a false detection. The images and labeled data corresponding to the crack targets that are identified as falsely detected are converted into negative samples or hard samples, and at least one of the training samples, confidence threshold and preset crack hazard rule base of the crack instance segmentation model is adjusted according to the manual review information.

7. A device for measuring and evaluating cracks in a steel tower base based on instance segmentation, characterized in that, include: The acquisition module is used to acquire the surface image of the iron tower base and the image identifier, shooting distance, shooting angle and equivalent calibration coefficient associated with the surface image of the iron tower base. The surface image of the iron tower base is input into a pre-trained crack instance segmentation model to obtain the instance mask, bounding box and confidence score corresponding to the crack target. According to the order of the crack targets in the model output results, each crack target is assigned a target number to form the crack instance segmentation result. The measurement module is used to perform geometric measurements on the foreground pixels in each instance mask based on the crack instance segmentation results, and to obtain the center point, pixel length, pixel width estimate and pixel area corresponding to each crack target; The conversion module is used to determine the current image scale based on the shooting distance, shooting angle and equivalent calibration coefficient, and to perform scale conversion on the estimated values ​​of pixel length, pixel width and pixel area based on the current image scale to obtain the estimated values ​​of actual length, actual average width and actual area corresponding to each crack target. The matching module is used to match the estimated actual length and the estimated actual average width with a preset crack hazard rule base to obtain the hazard star rating, triggering cause and handling suggestions for each crack target; The generation module is used to associate the crack instance segmentation results, center point, pixel length, estimated pixel width, pixel area, estimated actual length, estimated actual average width, estimated actual area, hazard star rating, triggering cause, and processing suggestions based on the image identifier to generate a structured crack profile.

8. The tower base crack measurement and evaluation device based on instance segmentation as described in claim 7, characterized in that: The acquisition module, the measurement module, the conversion module, the matching module, and the generation module are controlled to execute the method for measuring and evaluating cracks in the iron tower base based on instance segmentation as described in any one of claims 2 to 6.

9. An electronic device, characterized in that, include: Communication interface, processor, memory; The memory is used to store program instructions, which, when executed by the processor connected to the memory via the communication interface, cause the electronic device to implement the instance segmentation-based tower base crack measurement and evaluation method according to any one of claims 1 to 6.

10. A computer-readable storage medium having program instructions stored thereon, characterized in that, When the program instructions are executed by a computer, the computer implements the instance segmentation-based method for measuring and evaluating cracks in a tower base as described in any one of claims 1 to 6.

Images