A concrete member defect identification and labeling system

Abstract

The present application relates to the technical field of concrete member detection, and discloses a concrete member defect identification and labeling system, which comprises an image acquisition device matching module, a defect feature analysis module and a real-time labeling and regulation module, and a defect classification priority judgment module can be additionally provided.The image acquisition device matching module matches the optimal device based on the image resolution, device acquisition accuracy, working distance and illumination compensation parameters through adaptive characteristic value calculation; the defect feature analysis module quantitatively analyzes the texture, crack morphology, hole distribution and other features of the divided image by using multi-scale image segmentation, frequency domain transformation and other algorithms; the real-time labeling and regulation module dynamically adjusts the labeling position according to the defect position offset, size change rate and illumination fluctuation parameters; and the defect classification priority judgment module divides the defect grades according to the crack width, hole density and the like.The system improves the automation level and accuracy of concrete member defect detection and is suitable for building engineering member quality detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of concrete member detection, in particular to a concrete member defect identification and labeling system. BACKGROUND

[0002] In the field of construction engineering, concrete members serve as the main load-bearing structure, and their quality directly relates to the safety and durability of the building. However, during the production, transportation, construction, and use of concrete members, defects such as cracks, holes, honeycomb surfaces, and surface damage may occur due to factors such as material performance, construction technology, and environmental loads. These defects not only affect the appearance quality of the members but also may lead to a decline in structural performance and even cause safety accidents. Therefore, accurately identifying and labeling the defects of concrete members is a key step in realizing member quality assessment, defect early warning, and repair decision-making.

[0003] Traditional concrete member defect detection mainly relies on manual inspection methods. Inspectors identify defects through visual observation and knocking and listen to sound, and manually mark and record them. This method has significant limitations: on the one hand, manual detection is greatly influenced by subjective factors, and the experience and fatigue level of the inspectors can lead to poor accuracy and consistency in defect identification, and small or hidden defects may be easily missed; on the other hand, manual operation is inefficient, and for large and complex members or large-scale engineering detection tasks, it is time-consuming and labor-intensive, making it difficult to meet the needs of modern engineering rapid detection.

[0004] With the development of computer vision technology and artificial intelligence, image recognition-based concrete defect detection technology has gradually been applied. In existing technologies, some systems use fixed cameras to collect member images and then use image processing algorithms to identify defects. However, such systems generally have the following problems: first, the selection of image collection devices lacks specificity and does not fully consider the matching relationship between member surface features (such as reflective properties and size) and device parameters (such as collection accuracy, working distance, and light compensation), resulting in inconsistent image quality and affecting the accuracy of subsequent defect analysis; second, the defect feature analysis method is relatively simple, and most can only achieve simple identification of cracks, holes, and other defects, lacking in-depth analysis of defect morphology (such as crack extension direction and width variation), texture features, and distribution patterns, making it difficult to meet the needs of defect development trend prediction and severity assessment; third, the defect labeling process lacks dynamic regulation mechanisms, and when environmental light changes or the surface state of the member changes, the labeling position may deviate, making it difficult to accurately reflect the actual position and state of the defect in real time.

[0005] Furthermore, existing systems typically fail to categorize defects, making it difficult to develop differentiated treatment strategies based on defect severity, leading to irrational allocation of maintenance resources. Therefore, developing a concrete component defect identification and annotation system that can intelligently match image acquisition devices, perform multidimensional analysis of defect characteristics, and dynamically adjust annotation positions has important practical significance and engineering application value for improving the automation, accuracy, and efficiency of concrete component inspection. Summary of the Invention

[0006] The purpose of the present invention is to provide a concrete component defect identification and marking system to solve the problems raised in the above background technology.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a concrete component defect identification and marking system, the system comprising:

[0008] Image acquisition device matching module, defect feature analysis module and real-time annotation control module, including:

[0009] An image acquisition device matching module is used to obtain a concrete component surface image dataset and parameter datasets of each image acquisition device, and based on the obtained concrete component surface image dataset and parameter datasets of each image acquisition device, match the most suitable image acquisition device to perform image acquisition on the concrete component surface;

[0010] A defect feature analysis module is used to obtain a concrete component surface image dataset acquired by the most suitable image acquisition device, and extract and analyze the surface defect features of the concrete component based on the acquired concrete component surface image dataset;

[0011] The real-time annotation control module is used to obtain a real-time detection data set of surface defect characteristics of concrete components, and dynamically adjust the defect annotation position based on the obtained real-time detection data set of surface defect characteristics of concrete components.

[0012] Preferably, the concrete component surface image dataset and the parameter datasets of each image acquisition device specifically include the concrete component surface image resolution, the acquisition accuracy of each image acquisition device, the working distance of each image acquisition device, and the illumination compensation parameters of each image acquisition device.

[0013] Preferably, the step of matching the most suitable image acquisition device to acquire images of the concrete component surface based on the acquired concrete component surface image dataset and parameter datasets of each image acquisition device comprises:

[0014] Based on the acquired concrete component surface image dataset and the parameter dataset of each image acquisition device, a comprehensive analysis is performed to obtain the adaptive characteristic values ​​of each image acquisition device. The adaptive characteristic values ​​of each image acquisition device are used as the analysis basis for matching the most suitable image acquisition device.

[0015] Comparing the image acquisition device adaptation characteristic values, a maximum value in the image acquisition device adaptation characteristic values is obtained, and an image acquisition device corresponding to the maximum value in the image acquisition device adaptation characteristic values is a most adapted image acquisition device;

[0016] The most adapted image acquisition device is used to collect images of the concrete member surface.

[0017] Preferably, the image acquisition device adaptation characteristic values are obtained based on matching degrees of image resolution of the concrete member surface and image acquisition device acquisition accuracy, adaptation degrees of image acquisition device working distance and concrete member surface size, and correlation degrees of image acquisition device illumination compensation parameters and concrete member surface reflection characteristics, and the image acquisition device adaptation characteristic values are calculated in combination with preset acquisition device resolution weights, working distance weights, and illumination compensation weights.

[0018] Preferably, the concrete member surface image dataset collected by the most adapted image acquisition device includes: based on a multi-scale image segmentation algorithm and a gray scale distribution statistical method, the concrete member surface image is divided into a plurality of detection regions, and texture features, crack morphologies, and hole distributions of each detection region are quantitatively analyzed to generate a dataset containing defect positions and feature categories.

[0019] The multi-scale image segmentation algorithm includes a local region enhancement mechanism based on illumination self-adaptive adjustment, and the specific process is as follows: according to the illumination uniformity of the real-time collected image, high-contrast regions and low-contrast regions are dynamically segmented, and differentiated gray scale equalization strategies are respectively used, and an enhanced multi-scale segmentation image is generated after fusion.

[0020] Preferably, based on the obtained concrete member surface image dataset, concrete member surface defect features are extracted and analyzed, including:

[0021] The texture features of each detection region are subjected to frequency domain transformation to identify abnormal frequency domain components.

[0022] The crack morphology is subjected to skeleton extraction and curvature analysis to determine the crack extension direction and width change.

[0023] The hole distribution is subjected to density clustering calculation to distinguish isolated holes from continuous hole groups.

[0024] Preferably, the analysis process of the crack extension direction and width change includes: based on a coordinate sequence of crack skeleton points, a direction angle change rate between adjacent points is calculated, and a crack morphology dynamic change atlas is generated by sliding window statistics of crack width fluctuation range.

[0025] Preferably, the real-time detection data set based on the obtained concrete member surface defect feature is used to dynamically adjust the defect labeling position, comprising:

[0026] According to the defect position offset, the defect size change rate and the illumination condition fluctuation parameter in the real-time detection data set, the deviation degree of the current labeling position and the theoretical labeling position is calculated.

[0027] If the deviation degree exceeds the preset threshold, the labeling position compensation mechanism is started, and according to the correlation between the defect size change rate and the illumination compensation parameter, a labeling position correction vector is generated.

[0028] After detecting that the defect morphology is stable, the labeling position is restored to the theoretical labeling position.

[0029] Preferably, the generation process of the labeling position correction vector comprises: fitting a defect movement trend curve based on the historical trajectory of the defect center point coordinates, and combining the current illumination compensation parameter to weight and correct the curve slope, to generate the dynamically adjusted labeling coordinates.

[0030] Preferably, the system further comprises a defect classification priority determination module, which is used to divide the defects into emergency repair type, routine maintenance type and observation monitoring type according to the crack width, hole density and texture abnormality degree extracted by the defect feature analysis module, and dynamically update the classification results based on the preset defect level threshold to generate the defect priority labeling label.

[0031] Compared with the prior art, the beneficial effects of the present application are:

[0032] The image acquisition device matching module realizes intelligent selection of the image acquisition device by comprehensively analyzing multi-dimensional data such as the concrete member surface image resolution, device acquisition accuracy, working distance and illumination compensation parameter, and calculating the device adaptation feature value based on the matching degree, the adaptation degree, the correlation degree and the preset weight. This process fully considers the matching relationship between the member surface characteristics and the device parameters, can ensure that the acquired image has the best quality, provides a reliable data basis for subsequent defect analysis, avoids problems such as image blur and uneven illumination caused by improper device selection, and improves the detection accuracy from the source.

[0033] The defect feature analysis module adopts a multi-scale image segmentation algorithm and a gray distribution statistical method, and combines a local region enhancement mechanism with adaptive adjustment of illumination, so that the component surface image can be divided into fine detection regions, and the texture features, crack morphology and hole distribution of each region are quantitatively analyzed. Through frequency domain transformation, the texture abnormalities are identified, the skeleton is extracted, and the curvature analysis determines the crack morphology, so that the hole distribution is distinguished, and multi-dimensional feature extraction and in-depth analysis of the defects are realized. For example, based on the sequence of crack skeleton point coordinates, the directional angle change rate and the width fluctuation range are calculated, so that the crack morphology dynamic change atlas can be generated, key data for defect development trend prediction are provided, and the structure safety hidden danger can be warned in advance.

[0034] The real-time labeling regulation module dynamically calculates the labeling deviation degree based on the defect position offset, size change rate and illumination fluctuation parameters, and starts the compensation mechanism. By fitting the defect movement trend curve and combining the illumination parameter weighted correction, the labeling position correction vector is generated, so that when the environment changes or the defect morphology fluctuates, the labeling position can accurately reflect the actual state of the defect. When the defect morphology is stable, the labeling position is automatically restored to the theoretical labeling position, which not only ensures the dynamic adaptability of the labeling, but also ensures the accuracy of the final labeling result, avoiding the deviation problem caused by environmental interference in the traditional static labeling.

[0035] The defect classification priority determination module divides the defects into three categories of emergency repair, regular maintenance and observation monitoring according to the crack width, hole density and texture abnormality degree, and dynamically updates the classification results based on the preset threshold. This function realizes the fine management of defects, helps engineers to reasonably allocate maintenance resources according to the severity of defects, improves the maintenance efficiency, and reduces the maintenance cost. For example, the emergency repair type defect can trigger the repair mechanism in time to avoid safety accidents; the regular maintenance type defect can be included in the regular maintenance plan to realize preventive maintenance; and the observation monitoring type defect can be tracked to evaluate its development trend, avoiding over-maintenance. BRIEF DESCRIPTION OF DRAWINGS

[0036] Figure 1 A working principle diagram of the concrete component defect identification and labeling system is provided.

[0037] Figure 2 A working flowchart of the defect feature analysis module is provided.

[0038] Figure 3 A principle diagram of dynamic adjustment of the real-time labeling regulation module is provided.

[0039] Figure 4 A design diagram of the concrete component surface defect feature analysis is provided.

[0040] Figure 5 A working flowchart of the defect classification priority determination module is provided. DETAILED DESCRIPTION

[0041] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0042] Please refer to Figures 1-5 The present application relates to a concrete member defect identification and labeling system, which comprises an image acquisition device matching module, a defect feature analysis module and a real-time labeling regulation module. The modules work together to realize accurate identification and dynamic labeling of concrete member defects. The specific implementation scheme is as follows:

[0043] Image acquisition device matching module: Obtain the concrete member surface image dataset and the image acquisition device parameter dataset. The concrete member surface image dataset contains information such as image resolution, and the image acquisition device parameter dataset includes parameters such as acquisition accuracy, working distance and light compensation. Based on the above two types of datasets, the adaptive characteristic values of each image acquisition device are analyzed comprehensively. Specifically, by calculating the matching degree of the concrete member surface image resolution and the image acquisition device acquisition accuracy, the adaptation degree of the image acquisition device working distance and the concrete member surface size, and the correlation degree of the image acquisition device light compensation parameter and the concrete member surface reflection characteristic, and combining the preset acquisition device resolution weight, working distance weight and light compensation weight, the comprehensive adaptive characteristic values of each image acquisition device are obtained. Then, the adaptive characteristic values of each device are compared, and the device with the largest characteristic value is selected as the most adaptive image acquisition device, which is used for image acquisition of the concrete member surface.

[0044] Defect feature analysis module: after obtaining the concrete component surface image dataset collected by the most suitable image acquisition device, the image is divided into several detection regions based on a multi-scale image segmentation algorithm and a gray scale distribution statistical method. The multi-scale image segmentation algorithm includes a local region enhancement mechanism with adaptive light adjustment. The specific process is as follows: according to the uniformity of the light of the real-time collected image, the high-contrast region and the low-contrast region are dynamically segmented, and different gray scale equalization strategies are adopted for the two types of regions respectively, and the enhanced multi-scale segmentation image is generated after fusion. For each detection region, the texture feature, crack morphology and hole distribution are quantitatively analyzed to generate a dataset containing the defect position and feature category. Further, the texture feature is subjected to frequency domain transformation to identify abnormal frequency domain components; the crack morphology is subjected to skeleton extraction and curvature analysis, the directional angle change rate between adjacent points is calculated based on the coordinate sequence of the crack skeleton points, and the crack width fluctuation range is calculated through sliding window statistics to generate a crack morphology dynamic change atlas, so as to determine the crack extension direction and width change; the density clustering calculation is performed on the hole distribution to distinguish isolated holes from continuous hole groups.

[0045] Real-time labeling regulation module: after obtaining the real-time detection dataset of the concrete component surface defect features, the deviation degree of the current labeling position and the theoretical labeling position is calculated according to the defect position offset, defect size change rate and light condition fluctuation parameters in the dataset. If the deviation degree exceeds the preset threshold, the labeling position compensation mechanism is started, the defect movement trend curve is fitted based on the historical trajectory of the defect center point coordinates, and the curve slope is weighted and corrected combined with the current light compensation parameter to generate a labeling position correction vector, so as to realize the dynamic adjustment of the defect labeling position. When the defect morphology is detected to be stable, the labeling position is restored to the theoretical labeling position.

[0046] The application will be further described below in combination with Examples 1 to 5:

[0047] Example 1:

[0048] In the specific implementation of the image acquisition device matching module, the system completes the matching and selection of the most suitable image acquisition device by obtaining the concrete component surface image dataset and the parameter dataset of each image acquisition device. The specific composition of the two types of datasets includes: the concrete component surface image dataset contains the key parameter of image resolution, which is used to represent the density of pixel points in the image and the detail presentation ability; the parameter dataset of each image acquisition device includes acquisition accuracy, working distance and light compensation parameter, wherein the acquisition accuracy reflects the ability of the device to capture image details, the working distance is the effective distance between the device lens and the concrete component surface, and the light compensation parameter is used to adjust the imaging effect of the device under different light conditions.

[0049] When the system performs the adaptive feature value analysis based on the above data set, a correlation model of multi-dimensional parameters is first established. For the matching degree calculation of the image resolution of the concrete member surface and the collection accuracy of the image collection device, the resolution overlap rate algorithm is adopted: the pixel density of the image resolution (such as the number of pixels per inch DPI) and the pixel density of the device collection accuracy are normalized, and the overlapping proportion of the two is calculated through the formula:

[0050]

[0051] The value ranges from 0 to 1, and the larger the value, the higher the matching degree. For example, if the image resolution is 300 DPI, the collection accuracy of device A is 280 DPI, and the collection accuracy of device B is 320 DPI, then the matching degree of device A is 280 / 300≈0.933, and the matching degree of device B is 300 / 320≈0.937, both of which are close to the ideal matching state.

[0052] In the matching degree analysis of the working distance of the image collection device and the surface size of the concrete member, the actual size range of the concrete member surface is first determined through image processing algorithms (such as edge detection and contour extraction), and the length, width and other geometric parameters of the member are obtained. Then, the working distance of the device and the member size are analyzed for spatial correlation: if the member surface size is large (such as more than 10 square meters), the device needs to maintain imaging integrity at a long working distance. At this time, the matching degree calculation is based on the proportional relationship between the working distance and the member size, and the formula is:

[0053]

[0054] Wherein is a preset proportion coefficient (usually determined according to the field of view angle of the device and the imaging range of the member, for example, take 0.5). If the characteristic size of the member is 5 meters and the working distance of the device is 2.5 meters, the matching degree is 1-|2.5 / 5-0.5|=1, indicating complete matching; if the working distance is 3 meters, the matching degree is 1-|0.6-0.5|=0.9, which has a certain deviation but is still within an acceptable range.

[0055] For the correlation degree analysis of the image collection device light compensation parameters and the surface light reflection characteristics of the concrete member, the light reflection characteristic parameters of the member surface are first obtained through image gray statistical method, including average gray value, gray variance, etc. The light compensation parameters usually include automatic gain control (AGC) coefficient, exposure time, white balance parameter, etc. The system establishes the mapping relationship between the light compensation parameters and the light reflection characteristics: when the surface light reflection of the member is strong (such as the average gray value is higher than the threshold value), the device needs to reduce the AGC coefficient and shorten the exposure time to avoid overexposure; when the light reflection is weak, the AGC coefficient is increased and the exposure time is prolonged to enhance the brightness. The correlation degree calculation adopts the Pearson correlation coefficient, and the formula is:

[0056]

[0057] wherein is a light compensation parameter normalized value, is a light reflection characteristic parameter normalized value, , is a corresponding mean value. The closer the absolute value of the coefficient is to 1, the higher the correlation degree is, and the stronger the adaptability of the device light compensation parameter to the component light reflection characteristic is.

[0058] In the calculation process of the comprehensive adaptation characteristic value, a preset weight system is introduced to reflect the importance of each dimension parameter. The resolution weight, working distance weight and light compensation weight of the collection device can be adjusted according to the actual application scene. For example, in a high-precision detection scene, the resolution weight can be set to 0.5, the working distance weight to 0.3, and the light compensation weight to 0.2; in a complex light environment, the light compensation weight can be increased to 0.4, the resolution weight to 0.4, and the working distance weight to 0.2. The calculation formula of the comprehensive adaptation characteristic value is:

[0059] wherein . Taking device A as an example, if its resolution matching degree is 0.933, the working distance adaptation degree is 0.9, and the light compensation correlation degree is 0.85, the weights are 0.5, 0.3 and 0.2 respectively, then the comprehensive adaptation characteristic value is 0.933x0.5+0.9x0.3+0.85x0.2=0.9115; if the resolution matching degree of device B is 0.937, the working distance adaptation degree is 0.8 (assuming that the working distance exceeds the ideal range), and the light compensation correlation degree is 0.9, and the weights are the same, then the comprehensive value is 0.937x0.5+0.8x0.3+0.9x0.2=0.8985. By comparison, the comprehensive adaptation characteristic value of device A is higher, so it is selected as the most adaptive image collection device.

[0060] After completing the matching of the most adaptive device, the system triggers the device control instruction to make the device collect images on the surface of the concrete component. During the collection process, the device collects data according to the preset collection parameters (such as resolution, frame rate, exposure time, etc.), and transmits the collected image data to the defect feature analysis module in real time through wired or wireless transmission. In order to ensure the comprehensiveness of image collection, the system can control the device to move and shoot at multiple angles and multiple positions to form an image data set covering the entire area of the component surface. For example, for a large concrete column, the device can be driven by a mechanical arm to move at a constant speed along the axis of the column and collect images synchronously. A certain overlap area is reserved between each collected image to facilitate subsequent image stitching and panoramic analysis.

[0061] During the image acquisition process, the system continuously monitors the working state of the equipment and environmental parameters. If a significant change in lighting conditions is detected (such as sudden overcast weather causing a decrease in ambient light intensity), the system will automatically adjust the AGC coefficient and exposure time based on the current lighting intensity to ensure that the brightness and contrast of the acquired image remain within an appropriate range, avoiding the degradation of image quality caused by fluctuations in lighting conditions and affecting the subsequent extraction and analysis of defect features.

[0062] In addition, the system also has a device parameter database that stores historical adaptation data and performance indicators of various image acquisition devices. When a new concrete component detection task is assigned, the system can first retrieve similar historical records from the database based on the current component type, size, and detection environment, and obtain the device parameters and adaptation characteristics of the matched devices as reference data for the current matching process, thereby improving the matching efficiency. If there are no similar records in the database, the system will start the full device parameter analysis process and complete the dynamic matching of the most suitable device through the above multi-dimensional calculation and weight distribution.

[0063] Example 2:

[0064] In the defect feature analysis module, after obtaining the image data set of the concrete component surface collected by the most suitable image acquisition device, the system preprocesses and divides the image into regions using multi-scale image segmentation algorithms and gray scale distribution statistical methods to achieve fine analysis of defect features. The multi-scale image segmentation algorithm integrates a local region enhancement mechanism with adaptive illumination adjustment. This mechanism processes the image through the following process: First, the system detects the uniformity of the illumination of the acquired image in real time by calculating the gray mean value difference and variance distribution of different regions of the image to determine whether the illumination is uniform. If high-contrast regions (such as highlight regions under strong light) and low-contrast regions (such as dark regions covered by shadows) are detected in the image, the dynamic segmentation mechanism is activated to divide the image into two regions based on the gray value distribution range.

[0065] For high-contrast regions, the system uses an adaptive histogram equalization strategy that limits contrast. This strategy sets a contrast limit threshold to avoid excessive amplification of noise caused by histogram equalization while enhancing the details and textures within the region. In specific operations, the high-contrast region is divided into several sub-blocks, and histogram equalization is performed on each sub-block. By truncating the pixel value distribution that exceeds the threshold and redistributing it to the entire gray scale range, the system preserves edge information while suppressing noise. For low-contrast regions, due to the low overall gray value and blurred details, the system uses a global histogram equalization strategy to expand the gray scale dynamic range of the image, improving the brightness and contrast of the dark regions, and making potential defect features (such as fine cracks and hole edges) visible.

[0066] After the differentiation of the two types of regions is completed, the system merges the processing results of the high-contrast region and the low-contrast region through an image fusion algorithm. In the fusion process, based on pixel-level gray value matching and edge smoothing processing, the visual difference of the region segmentation boundary is eliminated, and a multi-scale segmentation image with enhanced uniformity of illumination and prominent detail features is generated. This image not only retains the texture details of the strong light region, but also improves the visibility of the shadow region, laying a foundation for subsequent detection region division.

[0067] Based on the enhanced multi-scale segmentation image, the system divides the concrete member surface image into a plurality of detection regions using a multi-scale image segmentation algorithm. In the segmentation process, combined with the spatial scale features of the image (such as different sizes of defects corresponding to different image resolution units), the multi-resolution levels of the image are constructed through a Gaussian pyramid, and the segmentation boundary is gradually determined from coarse to fine. At the lowest resolution level, the overall contour and main structural features of the member are identified through a contour detection algorithm; at the high-resolution level, for the regions within the contour, a watershed algorithm or region growing algorithm is used to segment the image into detection regions of different sizes according to the pixel gray similarity and spatial continuity. Each detection region corresponds to a specific physical region on the surface of the member, and the division granularity is determined according to the defect detection accuracy requirement, for example, for areas with high crack detection requirements, the segmentation granularity is finer (such as pixel-level segmentation), and for large-area texture analysis areas, larger segmentation units can be used.

[0068] After segmentation, the system quantitatively analyzes the texture features, crack morphology, and hole distribution of each detection region. In the texture feature analysis, by calculating the gray mean, gray variance, and texture roughness of each detection region, a texture feature vector is constructed. The gray mean reflects the overall brightness level of the region, the gray variance reflects the dispersion degree of the pixel value, and the texture roughness is obtained by calculating the mean and variance of the pixel gradient in the region, which is used to describe the fine or coarse degree of the texture. For example, the texture feature vector of a normal concrete surface presents a stable gray mean and a low variance, while a region with honeycomb pitting defects shows an increased gray variance and a higher roughness value.

[0069] For crack morphology analysis, the system first identifies potential crack edges within the detection region using edge detection algorithms (e.g., Canny operator), generating a binary edge image. Subsequently, skeleton extraction is performed on the edge image, simplifying cracks to single-pixel width skeleton lines to facilitate subsequent geometric feature analysis. During skeleton extraction, morphological erosion and dilation operations are employed to progressively remove redundant edge pixels, preserving the core structure of the cracks. Based on the skeleton lines, the system calculates basic parameters such as crack length and strike angle, and determines the degree of curvature and local feature points (e.g., inflection points, branch points) through curvature analysis. For example, straight-line cracks have lower curvature values, while meandering cracks have higher curvature values, indicating more complex stress states.

[0070] In the analysis of hole distribution, the system first separates the hole regions from the background in the detection area through a threshold segmentation algorithm, generating a binary hole image. Then, a connected region labeling algorithm is used to identify the outline of each hole and calculate geometric feature parameters such as area, perimeter, and circularity. The circularity parameter is calculated by the formula (4 × area) / (perimeter²), with a value range of 0 to 1. The closer the value is to 1, the closer the hole shape is to a circle, and the smaller the value, the more irregular the hole shape. Through a density clustering algorithm (such as a density-based spatial clustering algorithm), the system analyzes the spatial distribution of the holes, and according to the distance and density threshold between the holes, it distinguishes between isolated holes and continuous hole groups. Isolated holes are usually caused by local pouring defects, while continuous hole groups may indicate more serious structural defects and need to be paid special attention to.

[0071] In generating a dataset containing defect location and feature category, the system assigns a unique identifier to each detection region and stores the texture feature parameters, crack morphology parameters, hole distribution parameters, and defect location coordinates (such as the upper left corner coordinates, width, and height) of the region in a structured manner. The data format uses standard formats such as JSON or XML, facilitating data interaction with subsequent modules. For example, the data set entry of a certain detection region can be expressed as:

[0072] {

[0073] "region_id":"R001",

[0074] "position":{"x":100,"y":200,"width":50,"height":50},

[0075] "texture_features":{"mean":120,"variance":15,"roughness":8.5},

[0076] "crack_features": {"length": 200, "angle": 45°, "curvature": 0.02},

[0077] "hole_features": {"count": 3, "max_area": 50, "cluster_type": "isolated"}

[0078] }

[0079] During the multi-scale image segmentation and feature analysis process, the system supports interactive parameter adjustment. The operator can set the segmentation scale, gray equalization threshold, clustering density parameter, etc. through the graphical user interface, and real-time view the processing results and optimize them. For example, when it is detected that the crack edge of a certain area cannot be effectively identified due to light reflection, the light compensation parameter or edge detection threshold of the area can be manually adjusted to regenerate the edge image until the detection requirements are met.

[0080] In addition, the system has a historical data caching and comparison function, which can store image data sets and feature analysis results of the same component at different times. Through time series comparison, the trend of defect feature change (such as crack length growth, hole number increase, etc.) is analyzed, which provides a basis for component durability evaluation and maintenance decision. For example, a crack of 100 mm in length is found in a certain detection area during the first detection, and the crack length increases to 150 mm during the second detection three months later. The system automatically marks the change and generates a trend analysis report.

[0081] Example 3:

[0082] In the defect feature analysis module, for the detection area of the concrete component surface image, the system realizes the deep analysis of texture features, crack morphology and hole distribution through multi-step algorithms. Among them, the frequency domain transformation processing of texture features, the skeleton extraction and curvature analysis of crack morphology, and the density clustering calculation of hole distribution constitute a complete defect feature extraction process, and the specific implementation is as follows:

[0083] In the texture feature analysis link, the system performs frequency domain transformation on the two-dimensional texture image of each detection area to identify abnormal frequency domain components. The frequency domain transformation uses Discrete Fourier Transform (DFT) to convert the image from the spatial domain to the frequency domain, so that the periodicity, directionality and other characteristics of the texture are presented in the form of frequency components. In specific operation, first, the gray-scale image of the detection area is normalized to eliminate the influence of uneven light on pixel values; then the two-dimensional Fourier transform formula is used:

[0084]

[0085] Calculate the frequency domain matrix, where Represents coordinates in the spatial domain The grayscale value of the pixel at is the corresponding frequency domain coordinate The frequency component at and are the width and height pixels of the detection area image respectively, and is the frequency domain variable, is an imaginary unit. After the transformation, the central area of ​​the frequency domain matrix corresponds to the low-frequency components of the image (representing the overall brightness and slowly changing background), and the edge area corresponds to the high-frequency components (representing texture details, edges, and noise).

[0086] The system identifies abnormal frequency components by analyzing the amplitude spectrum distribution of the frequency matrix. While the texture of a normal concrete surface has a relatively uniform low-frequency distribution, defective areas (such as honeycombs and pitting) can significantly increase the high-frequency components. For example, if energy concentration appears in a high-frequency region in a certain direction within the amplitude spectrum, this may indicate a directional texture anomaly (such as formwork splicing marks or mechanical damage). The system automatically identifies abnormal regions that exceed this threshold by setting amplitude thresholds for frequency components, providing a basis for subsequent defect classification.

[0087] In crack morphology analysis, the system first performs skeleton extraction on the edge detection results of the detection area, simplifying the two-dimensional outline of the crack into a skeleton line with a single pixel width to highlight the geometric structure of the crack. Skeleton extraction uses a morphological erosion algorithm combined with the principle of topology preservation. It removes redundant pixels at the edge of the crack through iterative erosion operations until the remaining pixels form a connected and branchless skeleton structure. Each pixel point of the skeleton line (i.e., skeleton point) corresponds to the center path of the crack, and its coordinate sequence is ( ) records the extension trajectory of the crack.

[0088] Based on the skeleton point coordinate sequence, the system performs curvature analysis to determine the extension direction and width change of the crack. The curvature calculation uses the adjacent three-point difference method: for the first point, take the previous point , current point And the latter point , forming a vector and , the curvature is reflected by the rate of change of the vector angle. Specifically, the rate of change of the direction angle between adjacent points is expressed by the formula Calculate, where is the four-quadrant inverse tangent function, which is used to determine the direction angle of a vector (in the range of to ). The greater the absolute value of the derivative, the higher the degree of bending of the crack at that point, which may correspond to a stress concentration area.

[0089] To analyze the crack width variation, the system uses a sliding window technique to sample the edges on both sides of the skeleton line. With the skeleton point as the center, the distance between the edge pixels on both sides is detected in the window perpendicular to the tangent direction of the skeleton line. This distance is the crack width. By moving the window along the entire skeleton line (the window size is set according to the image resolution, such as 5-10 pixels), the width values at each position are counted to generate crack width fluctuation range data. For example, if the width of a certain section of the crack fluctuates between 1-2 pixels, it indicates that the width variation is small; if the width suddenly increases to 5 pixels at a certain point, it is marked as a width mutation point, which may indicate a risk of crack propagation.

[0090] In the analysis of hole distribution, the system first separates the holes in the detection area from the background by threshold segmentation algorithm (such as Otsu algorithm) to generate a binary image (hole area is black and background is white). Then, the connected region labeling algorithm is used to identify the contour of each hole and calculate its geometric feature parameters such as area, perimeter, and centroid coordinates. To distinguish between isolated holes and continuous hole groups, the system uses a density clustering algorithm with the hole centroid as the sample point, sets the neighborhood radius and the minimum number of samples . If a hole contains at least other holes in its neighborhood, it is determined to belong to a continuous hole group; otherwise, it is an isolated hole.

[0091] The specific clustering process is as follows: randomly select an unmarked hole as the core point, search for all adjacent holes in its neighborhood, if the number of adjacent holes is ≥ , then mark these holes as the same cluster, and continue to expand the clustering range with an unmarked point in the adjacent hole as the new core point; if the number of holes in the core point's neighborhood is < , then mark it as a noise point (isolated hole). Through this method, the system can effectively identify densely distributed hole groups. For example, when is set to 50 pixels (corresponding to an actual distance of 5 mm, converted according to the image resolution), and is set to 3, if the centroid distance between three holes is less than 50 pixels, it is determined to be a continuous hole group, indicating a possible large area of concrete pouring defects.

[0092] In the interactive implementation of feature analysis, the system provides a visual interface for the operator to adjust the algorithm parameters. For example, in the frequency domain transformation, the filtering threshold of high-frequency components can be manually set to exclude noise interference; in the skeleton extraction, the number of erosion iterations can be adjusted to avoid excessive erosion leading to the fracture of the crack skeleton; in the density clustering, the neighborhood radius and the minimum number of samples can be modified to adapt to the detection needs of holes of different scales. The operator adjusts the parameters in real time by viewing the feature analysis results (such as frequency domain amplitude spectrum, crack skeleton diagram, and hole clustering distribution diagram), and fine-tunes the parameters to ensure accurate extraction of defect features.

[0093] In addition, the system establishes a defect feature database to store historical data such as texture frequency domain components, crack curvature sequences, and hole clustering results of each detection area. By comparing the feature data of the same component at different times, the defect development trend can be analyzed. For example, the rate of change of the direction angle of a crack gradually increases in multiple detections, indicating that the degree of crack bending is increasing, which may be related to changes in the stress state of the structure; the hole clustering results of a certain area change from isolated holes to continuous hole groups, indicating that the defect range is expanding and needs to be monitored.

[0094] In the data output link, the system associates the feature analysis results of each detection area with the image position information to generate a labeled defect feature map. In the map, the texture abnormal area is marked with pseudo-color (such as red for high-frequency abnormalities), the crack skeleton line is superimposed on the original image and displayed as a blue line, the hole group is marked with a green polygon frame, and the isolated hole is marked with a yellow dot. This map can be transmitted to a remote terminal through a network for the inspector to view the distribution of defects in the component in real time, providing a visual reference for on-site investigation.

[0095] Example 4:

[0096] In the real-time labeling and regulation module, the system dynamically adjusts the defect labeling position based on the real-time detection data set of the surface defect features of the concrete component to adapt to possible position shifts, size changes, and environmental disturbances during the detection process. This process is achieved through deviation calculation, labeling position compensation mechanism triggering, and correction vector generation, and the specific implementation is described in detail in the following examples:

[0097] Suppose a crack is detected on the surface of a concrete beam component, and the initial theoretical labeling position is image coordinates (500, 300), corresponding to the actual physical position of 2 meters from the left end and 1.5 meters high on the beam body. The real-time detection data set contains the following parameters: defect position offset (unit: pixels, reflecting the coordinate difference between the current detection position and the theoretical position), defect size change rate (unit: %, reflecting the real-time change proportion of crack length or width), and illumination condition fluctuation parameter (measured by image gray value change, unit: gray value). The system first calculates the deviation degree of the current labeling position from the theoretical labeling position according to the three parameters. This deviation degree is a quantitative indicator that comprehensively reflects the accuracy of the labeling position. The larger the value, the more significant the deviation.

[0098] For example, at a certain moment, the defect position offset is detected as (+15, -8) pixels (i.e., a rightward offset of 15 pixels and a downward offset of 8 pixels), the defect size change rate is +5% (crack length increases by 5%), and the illumination condition fluctuation parameter is -20 gray values (overall image brightness decreases). The system performs weighted summation of each parameter according to the pre-set weight distribution rule (such as offset weight 0.5, size change rate weight 0.3, and illumination fluctuation weight 0.2) to obtain the deviation degree value. If the deviation degree exceeds the pre-set threshold (such as 10 pixel equivalents, converted by the ratio of pixels to actual physical size), the current labeling position needs to be adjusted, and the labeling position compensation mechanism is started.

[0099] The core of the labeling position compensation mechanism is to generate a labeling position correction vector, which is used to indicate the coordinate change from the current labeling position to the corrected position. When generating the correction vector, the system first analyzes the historical trajectory of the defect center point coordinates. Taking a crack as an example, suppose that in the past 5 detections, the crack center point coordinates are (490, 305), (502, 302), (510, 298), (518, 295), and (525, 290), respectively. By fitting these coordinate points, a defect movement trend curve is formed, which can be fitted using a polynomial (such as a quadratic polynomial) or a linear polynomial, reflecting the movement direction and speed of the defect over time. For example, the linear fitting trend equation is: x = 500 + 3t, y = 300 - 2t (where t is the detection number), indicating that the crack center point moves 3 pixels to the right and 2 pixels downward with each detection.

[0100] Meanwhile, the system combines the current light condition fluctuation parameter to weight and correct the slope of the trend curve. Light changes can cause the edge detection result of defects in the image to shift, for example, when the brightness decreases, the crack edge may shrink inward due to the decrease in contrast, causing the detection position to shift inward. Assuming that the current light fluctuation parameter is -20 gray value, according to the correlation between the light compensation parameter and the position shift established based on historical data (for example, every 10 gray value decrease, the crack detection position shifts 1 pixel inward), the y-axis slope of the trend curve is corrected. The original slope is -2 pixels / time, and after correction, it is -2-(20 / 10)*1=-4 pixels / time, that is, after considering the effect of light, the prediction of the y-coordinate shift speed is accelerated.

[0101] Based on the corrected trend curve, the system calculates the predicted coordinates of the defect center point at the next time, for example, the predicted coordinates of the 6th detection are x=500+3x6=518 pixels, y=300-4x6=276 pixels. Compare the predicted coordinates with the current theoretical labeling position (500, 300) to generate a labeling position correction vector (+18, -24) pixels, that is, the current labeling position needs to move 18 pixels to the right and 24 pixels downward to match the predicted position. The system applies the correction vector to the real-time labeling interface, so that the labeling box (such as a rectangular box or an arrow mark) adjusts the position dynamically according to the correction vector, ensuring that it always accurately covers the actual area of the defect.

[0102] When the defect morphology is stable (such as the position shift of continuous 3 detections is less than 1 pixel, and the size change rate is less than 0.5%), the system determines that it is not necessary to continue dynamic adjustment, and restores the labeling position to the theoretical labeling position. The theoretical labeling position is determined based on the initial detection result of the defect feature analysis module, and represents the true position of the defect under ideal light conditions. For example, after the above-mentioned crack is stable in morphology, the labeling box returns from the dynamically adjusted position (518, 276) to (500, 300), avoiding the accumulation of labeling deviation caused by short-term fluctuations.

[0103] In practical applications, different types of defects may have different dynamic characteristics. For example, due to the thermal expansion and contraction of cracks caused by temperature changes, the position shift and size change rate present periodic fluctuations, and the system can recognize this periodic law and optimize the fitting parameters of the trend curve by continuously recording historical trajectories; for cracks caused by structural settlement, the trend curve presents a one-way growth characteristic, and the system automatically extends the dynamic adjustment period until the settlement is stable.

[0104] The influence of light condition fluctuation can also be reflected in concrete surfaces of different materials. For example, components with smooth surfaces are prone to reflect light under strong light, causing large-area deviation in defect edge detection results. In this case, the weight of the light compensation parameter needs to be dynamically increased (e.g., from the preset 0.2 to 0.4) to highlight the impact of light changes on the labeled position. Components with rough surfaces are less affected by light fluctuations, so the weight can remain unchanged. The system automatically adjusts the weight of each parameter based on the material type of the component through machine learning algorithms, improving the adaptability of dynamic adjustment.

[0105] The dynamic adjustment process of the labeled position is synchronized with the image acquisition frequency. For example, when the system acquires images at a frequency of 5 frames per second, the real-time labeling regulation module updates the detection dataset every 0.2 seconds and triggers the calculation of deviation degree and the generation of correction vectors once, ensuring that the adjustment speed of the labeled position matches the image acquisition speed, achieving closed-loop control of "real-time detection-real-time analysis-real-time labeling".

[0106] In addition, the system has a labeled position adjustment log function that records the generation time of each correction vector, correction parameters (such as offset, change rate, light fluctuation value), and coordinate data before and after adjustment. This log can be used to trace the accuracy of the labeling process. For example, in subsequent reviews, if a defect labeling position is found to be controversial, the log can be used to analyze the detection environment and algorithm parameters at the time to determine whether the adjustment was reasonable.

[0107] For scenarios where multiple defects exist simultaneously, the system uses a multi-thread processing mechanism to generate a labeled position correction vector for each defect independently. For example, there are 3 cracks and 2 hole groups on the surface of a concrete column. The detection dataset of each defect is independently input into the real-time labeling regulation module, and each module thread calculates the deviation degree and generates a correction vector in parallel, avoiding labeling errors caused by interference between defects.

[0108] Example 5:

[0109] The defect classification priority determination module in the system automatically classifies concrete component defects based on the three core parameters of crack width, hole density, and texture abnormality degree output by the defect feature analysis module, and generates dynamically updated priority labeling labels. The implementation process of this module is closely related to actual engineering needs, and through preset level thresholds and logical rules, it achieves accurate grading of defects. The following will be explained in detail with specific examples:

[0110] Example scenario: In the detection of a precast concrete box girder of a bridge, the feature parameters of three typical defects obtained by the defect feature analysis module are as follows:

[0111] Defect A: Located in the middle of the box girder web, crack width 0.35mm, hole density 0 / m2 (no holes), texture abnormality degree performance as local honeycomb roughness caused by roughness value increased by 15%;

[0112] Defect B: Located at the edge of the bottom plate, crack width 0.12mm, hole density 8 / m2 (continuous hole group distribution), texture abnormality degree slight (roughness value increased by 5%);

[0113] Defect C: Located on the top surface of the top plate, crack width 0.08mm, hole density 2 / m2 (isolated holes), texture abnormality degree no significant change (roughness value increased by 2%).

[0114] Classification rules and threshold settings:

[0115] The defect level threshold preset by the module is divided into three levels, the specific standards are as follows:

[0116] Emergency repair type: crack width ≥0.3mm and showing expansion trend, or hole density ≥5 / m2 and forming continuous group, or texture abnormality leading to significant decrease of structure bearing capacity (such as large area peeling);

[0117] Routine maintenance type: crack width between 0.1mm-0.3mm, or hole density 1-5 / m2 and isolated distribution, or texture abnormality not affecting structure safety but need to prevent deterioration;

[0118] Observation and monitoring type: crack width <0.1mm and no expansion, or hole density <1 / m2, or texture abnormality only as surface imperfections (such as slight sand spots).

[0119] Classification process of defect A: The crack width of defect A is 0.35mm, exceeding the threshold of emergency repair type (0.3mm), and through the crack morphology dynamic change atlas analysis, it is found that the extension direction is consistent with the principal stress direction, and there is a risk of continuous expansion. Although the defect has no holes, a single parameter (crack width) has triggered the emergency repair condition. The module automatically classifies it as emergency repair type, generates a red label tag, and adds the warning information "crack over limit, need to review within 24 hours". At the same time, according to the comparison of historical detection data, if it is found that the crack width has increased by 0.05mm in the past week, the classification result will be further confirmed to avoid misjudgment due to detection error.

[0120] Classification process of defect B: The crack width of defect B is 0.12 mm, which is within the threshold range of regular maintenance class (0.1 mm-0.3 mm), but its hole density is 8 / m² and is in a continuous group distribution, exceeding the hole density threshold of emergency repair class (5 / m²). According to the principle of "not low but high", the module prioritizes the hole density parameter to classify it as an emergency repair class, generating a red label tag. In addition, through the density clustering results of hole distribution, it is found that the hole group occupies an area of 1.5 square meters at the edge of the bottom plate, which may affect the waterproof performance and reinforcement cover thickness of the structure, further supporting the classification decision of emergency repair.

[0121] Classification process of defect C: The crack width of defect C is 0.08 mm, which is less than the lower limit of regular maintenance class (0.1 mm), the hole density is 2 / m² and is in isolated distribution, which does not meet the hole density threshold of regular maintenance class (1-5 / m² and needs to be continuously distributed), and the texture abnormality degree is only slight roughness increase. The module determines that it belongs to the observation monitoring class, generates a blue label tag, and records the note information "suggesting to review once every quarter". At the same time, the system automatically associates the historical detection data of the area, if it is found that the same defect exists in the previous detection and there is no development trend, the observation monitoring classification is maintained; if the hole density increases to 3 / m² and shows clustering trend in subsequent detection, the classification update mechanism is triggered, and the classification is adjusted to regular maintenance class.

[0122] Dynamic updating mechanism: The defect classification priority determination module is not completed in one time, but is dynamically adjusted based on real-time detection data and historical trend analysis. For example:

[0123] A certain defect is initially classified as a regular maintenance class (crack width 0.2 mm), and in subsequent detection, it is found that its width increases to 0.32 mm and the hole density increases to 4 / m², the module automatically upgrades it to an emergency repair class, the label tag color changes from yellow to red, and a warning notice is sent to the operation and maintenance platform;

[0124] A certain emergency repair class defect is repaired by grouting, and the recheck detection shows that the crack width decreases to 0.09 mm and the hole density is 0 / m², the module reduces it to an observation monitoring class according to the repaired data, the label tag color changes from red to blue, and the defect state is updated to "repaired".

[0125] Multi-parameter coupling analysis: When a defect involves multiple parameters exceeding the threshold, the module uses coupling analysis logic. For example, a defect with a crack width of 0.25 mm (close to the emergency repair threshold), a hole density of 4 / m² (close to the emergency repair threshold), and a texture abnormality that causes roughness to increase by 25% due to exposed reinforcement. At this time, the module evaluates the overall risk by weighting and superimposing: crack width weight 0.5, hole density weight 0.3, texture abnormality weight 0.2, calculate the overall risk value 0.25 x 0.5 + 4 x 0.3 / 5 (standardized processing) + 0.25 x 0.2 = 0.125 + 0.24 + 0.05 = 0.415, exceeding the emergency repair threshold (0.4), so it is classified as an emergency repair.

[0126] Engineering application of labeling tags: The generated defect priority labeling tags are deeply bound with image acquisition location information to form a defect distribution map with geographic coordinates. For example, in the bridge BIM model, the red label defect corresponds to a specific location in the box girder web, and clicking the label can view detailed feature parameters, classification basis, and historical evolution process; the yellow label defect corresponds to the edge area of the bottom plate, with regular maintenance suggestions (such as surface sealing treatment); the blue label defect corresponds to the top surface of the top plate, showing negligible surface flaws. This distribution map can be synchronized to the field detection personnel through the mobile APP to guide them to carry out repair work according to priority, avoiding resource mismatch.

[0127] Manual intervention and review: Although the module realizes automatic classification, the system still supports manual intervention functions. For example, the detection personnel find that a defect does not reach the emergency repair threshold, but is located in a key stress part of the structure (such as near the support), which can be manually classified as an emergency repair class, and the label tag is marked with "manual intervention" identification; for defects caused by detection angle leading to parameter misjudgment (such as crack width measurement value being larger due to reflection interference), the classification result can be corrected through manual review to ensure the engineering rationality of the classification.

[0128] Data storage and traceability: Defect classification results and dynamic update records are stored in the system database to form a complete defect life cycle file. The file content includes: first detection time, historical classification results, key parameters triggering classification change, manual intervention record, repair measures and effect evaluation, etc. For example, in the process of a defect being downgraded from the emergency repair class to the observation and monitoring class, the database records the crack width changes before and after repair (0.35 mm→0.1 mm→0.08 mm), hole density changes (8 / m²→2 / m²→0 / m²), providing data support for structure durability evaluation.

[0129] It is to be understood that the terminology used herein such as first and second, and the like, is only used to distinguish one entity or action from another entity or action, and does not necessarily require or imply any such actual relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0130] While embodiments of the present application have been shown and described with reference to particular embodiments thereof, it will be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the application. The scope of the application is defined by the appended claims and their equivalents.

Claims

1. A concrete member defect recognition labeling system, characterized by, The system comprises an image acquisition device matching module, a defect feature analysis module and a real-time labeling regulation module. The image acquisition device matching module is used to obtain concrete member surface image data sets and image acquisition device parameter data sets, and based on the obtained concrete member surface image data sets and image acquisition device parameter data sets, the most suitable image acquisition device is matched to collect images of the concrete member surface. The defect feature analysis module is used to obtain concrete member surface image data sets collected by the most suitable image acquisition device, and based on the obtained concrete member surface image data sets, the defect features of the concrete member surface are extracted and analyzed. The real-time labeling regulation module is used to obtain real-time detection data sets of the defect features of the concrete member surface, and based on the obtained real-time detection data sets of the defect features of the concrete member surface, the defect labeling position is dynamically adjusted. The concrete member surface image data sets and the image acquisition device parameter data sets specifically include concrete member surface image resolution, image acquisition device collection accuracy, image acquisition device working distance and image acquisition device illumination compensation parameters. Based on the obtained concrete member surface image data sets and the image acquisition device parameter data sets, the most suitable image acquisition device is matched to collect images of the concrete member surface, which includes: Based on the obtained concrete member surface image data sets and the image acquisition device parameter data sets, comprehensive analysis is performed to obtain image acquisition device adaptation characteristic values, which serve as the basis for matching the most suitable image acquisition device. The image acquisition device adaptation characteristic values are compared to obtain the maximum value among the image acquisition device adaptation characteristic values, and the image acquisition device corresponding to the maximum value is the most suitable image acquisition device. The most suitable image acquisition device is used to collect images of the concrete member surface. The image acquisition device adaptation characteristic values are obtained based on the matching degree of the concrete member surface image resolution and the image acquisition device collection accuracy, the adaptation degree of the image acquisition device working distance and the concrete member surface size, and the correlation degree of the image acquisition device illumination compensation parameters and the concrete member surface reflection characteristics, combined with the preset matching degree weight, adaptation degree weight and correlation degree weight of the collection device, to calculate the comprehensive adaptation characteristic values of the image acquisition devices. The concrete member surface image data sets collected by the most suitable image acquisition device are obtained based on a multi-scale image segmentation algorithm and a gray scale distribution statistical method, the concrete member surface image is divided into a plurality of detection regions, and the texture features, crack morphology and hole distribution of each detection region are quantitatively analyzed to generate a data set containing defect positions and feature categories. The multi-scale image segmentation algorithm includes a local region enhancement mechanism based on illumination self-adaptive adjustment, and the specific process is as follows: according to the illumination uniformity of the real-time collected image, the high-contrast region and the low-contrast region are dynamically segmented, and different gray scale equalization strategies are used respectively, and an enhanced multi-scale segmentation image is generated after fusion.

2. The concrete member defect recognition labeling system of claim 1, wherein: The concrete member surface image data set is acquired, and the concrete member surface defect features are extracted and analyzed, including: Performing frequency domain transformation on the texture features of each detection area to identify abnormal frequency domain components; Performing skeleton extraction and curvature analysis on the crack morphology to determine the crack extension direction and width variation; Performing density clustering calculation on the hole distribution to distinguish isolated holes from continuous hole groups.

3. The concrete member defect recognition labeling system according to claim 2, characterized by: The analysis process of the crack extension direction and width variation includes: based on the coordinate sequence of the crack skeleton points, calculating the direction angle change rate between adjacent points, and through the sliding window statistics crack width fluctuation range, generating crack morphology dynamic change atlas.

4. The concrete member defect recognition labeling system of claim 1, wherein: The real-time detection data set based on the acquired concrete member surface defect features is used to dynamically adjust the defect labeling position, including: According to the defect position offset, defect size change rate and illumination condition fluctuation parameters in the real-time detection data set, the deviation degree of the current labeling position and the theoretical labeling position is calculated; If the deviation degree exceeds the preset threshold, the labeling position compensation mechanism is started, and according to the correlation between the defect size change rate and the illumination compensation parameter, a labeling position correction vector is generated; After detecting the stable defect morphology, the labeling position is restored to the theoretical labeling position.

5. The concrete member defect recognition labeling system according to claim 4, characterized by: The generation process of the labeling position correction vector includes: based on the historical trajectory of the defect center point coordinates, fitting the defect movement trend curve, and combining the current illumination compensation parameter to weight and correct the curve slope, generating the dynamically adjusted labeling coordinates.

6. The concrete member defect recognition labeling system of claim 2, wherein: It also includes a defect classification priority determination module for dividing defects into emergency repair type, routine maintenance type and observation monitoring type according to the crack width, hole density and texture abnormality degree extracted by the defect feature analysis module, and dynamically updating the classification results based on the preset defect level threshold to generate defect priority labeling labels.

Images