High-speed railway bridge pier column concrete disease damage detection method based on machine vision

Abstract

This invention relates to the fields of computer vision and civil engineering monitoring technology. It discloses a machine vision-based method for detecting concrete defects in high-speed railway bridge piers, comprising: calculating a global threshold and a local threshold based on a second image; weighting the global and local thresholds to obtain a mixed threshold; and segmenting the second image according to the mixed threshold to obtain a defect image. This invention, through binarization preprocessing of the image and combined with the Faster R-CNN algorithm, can accurately extract concrete defect features of high-speed railway bridge piers, significantly reducing computational resource consumption. Simultaneously, a weighted risk scoring method is designed to effectively assess the degree of concrete defects, ensuring high-precision assessment results. Furthermore, this invention innovatively designs an early warning system that can issue timely warnings when the detected defect degree reaches a certain threshold, avoiding potential safety risks.

Description

Technical Field

[0001] This invention relates to the fields of computer vision and civil engineering monitoring technology, and more specifically, to a method for detecting concrete defects and damage in high-speed railway bridge piers based on machine vision. Background Technology

[0002] With the continuous expansion of my country's railway network and the increase in its service life, a large number of railway and high-speed railway bridge piers and concrete structures face severe environmental erosion and dynamic load coupling effects. Especially in cold, saline-alkali, or humid regions, factors such as freeze-thaw cycles, chloride ion erosion, and acid rain corrosion significantly accelerate the expansion of concrete cracks, surface spalling, and steel reinforcement corrosion, posing a serious threat to railway traffic safety. Many early-built railway bridges have entered a period of high incidence of defects, and concrete deterioration problems are becoming increasingly prominent. Traditional manual inspection methods are no longer sufficient to meet the safety assurance requirements of a modern railway network that requires full coverage and high efficiency. Although there are currently monitoring methods based on fixed sensors, their deployment density is limited and their coverage is insufficient, making it difficult to comprehensively capture sudden or distributed damage.

[0003] In the complex environment of railway bridges, existing technologies still face many challenges: First, background interference (such as ballast, shadows, and vegetation) severely affects the accuracy of defect identification, leading to misjudgments and missed diagnoses; second, many methods focus only on a single defect type, lacking the ability to fuse multiple features and conduct comprehensive evaluation, thus failing to fully assess the health status of the bridge; furthermore, existing monitoring systems are insufficient in terms of real-time performance and engineering applicability, and a complete monitoring system integrating data acquisition, analysis, and early warning functions has not yet been formed. Therefore, there is an urgent need to develop an intelligent defect monitoring method for railway and high-speed railway bridge piers that integrates multi-source visual information and has a hierarchical early warning function. Summary of the Invention

[0004] To overcome the above-mentioned deficiencies of the prior art, embodiments of the present invention provide a method for detecting concrete defects and damage in high-speed railway bridge piers based on machine vision.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A machine vision-based method for detecting concrete defects in high-speed railway bridge piers, the method comprising: A first image of the pier concrete is obtained by taking pictures of the pier surface from multiple angles. The first image is preprocessed and output as a second image. Based on the second image, a global threshold and a local threshold are calculated respectively. The global threshold and the local threshold are weighted to obtain a mixed threshold. The second image is then segmented according to the mixed threshold to obtain a disease image. The improved Faster R-CNN algorithm is used to identify concrete defects in the defect images to obtain different defect areas and their categories. Concrete defect features are extracted from the defect images, including: total crack length, spalling area, and exposed steel reinforcement ratio. The total length of the crack, the spalling area, and the exposed ratio of the reinforcing bars are assigned values ​​respectively, the hazard score of the pier column is calculated, and the warning level is determined based on the hazard score.

[0006] Further, the calculation of the global threshold and local threshold based on the second image includes: A1: Iterate through all possible grayscale values ​​of the second image, and determine the optimal segmentation threshold by maximizing the inter-class variance between the damaged region and the background region. This threshold is the global threshold, and its objective function is:

[0007] In the formula: , These represent the pixel percentages of the damaged area and the background area, respectively. , This indicates its average gray value; A2: For any pixel in the second image, a fixed window is set. The threshold of any pixel is determined by the local mean and local standard deviation within the neighborhood window of that pixel. The local threshold can be obtained by the following expression:

[0008] In the formula: It is a local mean. denoted as the local standard deviation, and k is the adjustment coefficient.

[0009] Further, the weighting of the global threshold and the local threshold to obtain the mixed threshold includes:

[0010] in, The weighting parameter is calculated using the following formula:

[0011] In the formula: The global grayscale standard deviation of the second image is . Let be the texture complexity index of the second image. This is the adjustment constant.

[0012] Furthermore, the step of obtaining the disease image is as follows: The pixel grayscale value is compared with the calculated mixing threshold. If the grayscale value of the pixel is less than the mixing threshold, it is determined to be a diseased area and marked as black. If the grayscale value of the pixel is greater than or equal to the mixing threshold, it is considered a good area and marked as white, thus obtaining the disease image.

[0013] Furthermore, the specific steps for identifying concrete defects in the defect images using the improved Faster R-CNN algorithm are as follows: Convolutional feature maps are extracted from the disease images using a convolutional neural network, and a region proposal network is introduced to generate multiple candidate regions. ROI pooling is used to transform multiple candidate regions into pooled feature maps of a fixed size. Then, based on the pooled feature maps, the disease category probability corresponding to each feature value is calculated, and the multiple candidate regions are classified.

[0014] Furthermore, the specific process for calculating the risk score of the pier is as follows: The total length of the crack, the spalling area, and the exposed rebar ratio are respectively denoted as B1, B2, and B3, and the initial risk score is calculated using the following formula:

[0015] In the formula: These are measured values. Design limits, This is an abnormal indicator. The weights corresponding to the total crack length, spalling area, and exposed rebar ratio; By introducing a disease rate factor, the initial risk score is corrected to obtain the risk score of the pier column.

[0016] Furthermore, the introduction of a disease rate factor to correct the initial risk score involves the following steps: The disease rate factor is calculated using the following formula:

[0017] In the formula: j and i represent the sampling number, which is the time sequence number of the image acquisition; , The peeling area is extracted from the j-th and i-th samples. Indicates a fixed sampling interval; Based on the disease rate factor, determine whether to apply an additive correction to the initial risk score, and the correction rules are as follows: If the initial risk score Then, based on the initial risk score, add N; If the initial risk score Then, based on the initial risk score, add M; In other cases, the risk score of the pier remains consistent with the initial risk score.

[0018] Furthermore, the warning levels include: If the pier is rated as dangerous The warning level is level five, and it will be continuously monitored. If the pier is rated as dangerous The warning level is level four, and monthly manual review will be conducted. If the pier is rated as dangerous The warning level is level three, and weekly inspections and patrol notifications will be conducted accordingly. If the pier is rated as dangerous The warning level is Level II, and emergency testing will be conducted within 48 hours. If the pier is rated as dangerous The warning level is Level 1, and traffic should be closed immediately.

[0019] An electronic device includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the aforementioned machine vision-based method for detecting concrete defects in high-speed railway bridge piers.

[0020] A computer-readable storage medium storing a computer program, which, when executed, implements the aforementioned machine vision-based method for detecting concrete defects in high-speed railway bridge piers.

[0021] Compared with the prior art, the beneficial effects of the present invention are as follows: This application discloses a machine vision-based method for detecting concrete defects in high-speed railway bridge piers, comprising: capturing images of the pier surface from multiple angles to obtain a first image of the pier concrete; preprocessing the first image to output a second image; calculating a global threshold and a local threshold based on the second image; weighting the global threshold and the local threshold to obtain a mixed threshold; and segmenting the second image according to the mixed threshold to obtain a defect image. This invention, through binarization of the preprocessed image and combined with the Faster R-CNN algorithm, can quickly and accurately extract concrete defect features of high-speed railway bridge piers, significantly reducing computational resource consumption. Simultaneously, a weighted risk scoring method is designed to effectively assess the degree of concrete defects, ensuring high-precision assessment results. Furthermore, this invention innovatively designs an early warning system that can issue timely warnings when the detected defect degree reaches a certain threshold, avoiding potential safety risks. It possesses high efficiency, accuracy, and real-time performance, providing strong support for the maintenance and safety management of high-speed railway bridges and ensuring the stability and safety of the engineering structure. Attached Figure Description

[0022] Figure 1 A flowchart of the machine vision-based method for detecting concrete defects and damage in high-speed railway bridge piers provided by the present invention. Figure 2 A schematic diagram of the structure of an electronic device provided by the present invention; Figure 3 A schematic diagram of the structure of a computer-readable storage medium provided by the present invention; Figure 4 A flowchart for segmenting disease images provided by the present invention; Figure 5 This invention provides a flowchart for calculating the risk score of pier columns. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1 Please see Figure 1 As shown in the figure, this embodiment discloses a method for detecting concrete defects in high-speed railway bridge piers based on machine vision. The method includes: S101: Take a first image of the pier concrete by shooting the surface of the pier from multiple angles, preprocess the first image, and output it as a second image.

[0025] Real-time imaging technology ensures continuous and stable image acquisition of high-speed railway bridge piers, which is crucial for monitoring changes in concrete defects. Considering the typically slow process of concrete frost formation and to effectively utilize computing resources, the sampling frequency is set to at least one image every 5 minutes. This frequency guarantees monitoring continuity without being too frequent and wasting resources. Images used for subsequent processing are acquired at fixed time intervals, which are adjusted according to the working environment and condition of the high-speed railway bridge piers to meet actual monitoring needs. Preprocessing is key to improving image quality, including denoising and contrast enhancement, which helps improve the effectiveness of subsequent processing.

[0026] The preprocessing of the first image involves median filtering for denoising and adaptive histogram equalization to enhance contrast. Specifically, the denoising step applies median filtering to the initial grayscale image of the first image to reduce noise interference. Median filtering, as an effective non-linear processing method, can effectively remove salt-and-pepper noise and speckle noise while preserving image edge features. This process involves sorting the grayscale values ​​of neighboring pixels around each pixel and replacing the original pixel value with the median value, thereby reducing the impact of extreme values, smoothing the image, and improving the overall image quality. This step is crucial for subsequent image analysis because it provides a clearer image input.

[0027] Next, the contrast of the first image is enhanced using a contrast enhancement algorithm. This step aims to adjust the grayscale differences between pixels, i.e., adjust the brightness contrast, making the difference between the damaged area and the background more obvious. Contrast enhancement improves the visual effect of the image, making the features of the concrete damaged area more prominent, and providing better input for subsequent binarization processing. This contrast adjustment can be achieved through methods such as histogram equalization or adaptive histogram equalization (CLAHE), which can enhance the details of the concrete damaged area in the image and improve the recognizability of the concrete damaged features. S102: As Figure 4 As shown, a global threshold and a local threshold are calculated based on the second image. The global threshold and the local threshold are weighted to obtain a mixed threshold. The second image is then segmented based on the mixed threshold to obtain a disease image.

[0028] As a specific implementation, the step of calculating the global threshold and the local threshold based on the second image includes: A1: The global threshold T1 of the second image is calculated using the maximum inter-class variance method. By iterating through all possible grayscale values ​​of the second image, the optimal segmentation threshold is determined by maximizing the inter-class variance between the damaged region and the background region; this is the global threshold. Its objective function is:

[0029] In the formula: ， These represent the pixel percentages of the damaged area and the background area, respectively. ， This indicates its average grayscale value. It should be noted that when... When the maximum value is obtained, the corresponding gray value is the global threshold T1. This algorithm does not require preset parameters. It should also be noted that the grayscale values ​​set here are primarily for differentiation, and are not absolutely limited to 0 and 255 respectively. In practice, the setting only needs to distinguish the monitored concrete defect area from the background. This flexibility allows the algorithm to adapt to different image conditions and monitoring needs. This binarization process can significantly improve the accuracy of concrete defect feature identification, providing a solid foundation for subsequent concrete defect damage assessment and early warning. Furthermore, this binarization method has high computational efficiency, can meet the needs of real-time monitoring, and ensures the timeliness and effectiveness of concrete defect monitoring in high-speed railway bridge piers.

[0030] Specifically, the OTSU algorithm, also known as Otsu's method, is an adaptive thresholding binarization method. Its basic principle is to analyze the overall grayscale characteristics of an image and calculate a threshold that maximizes the inter-class variance of the image's grayscale. At this threshold, the segmentation probability between the target and the background reaches its optimal state. Therefore, the core of the OTSU algorithm is to find an optimal threshold, setting pixels with values ​​less than this threshold to 0 and pixels with values ​​greater than this threshold to 255. Typically, the optimal OTSU threshold corresponds to the lowest point between the two peaks of the image's grayscale histogram. Calculating the global threshold T1 for each pixel using the OTSU algorithm includes the following steps: In the second image, foreground pixels, background pixels, and total pixels are distinguished, and the average grayscale value is calculated based on the foreground pixels, the background pixels, and the total pixels. Based on the grayscale mean, a maximum inter-class variance objective function is established, and the global threshold T1 is calculated from the maximum inter-class variance objective function. Assume that the proportion of foreground pixels to the total number of pixels in the image is... The mean is Let the proportion of the background pixels to the total number of pixels in the image be... The mean is The mean gray level of the image is then:

[0031] The maximum inter-class variance objective function for the grayscale image is:

[0032] when When the threshold is at its maximum, the probability of separating the image background and the target is at its maximum, which also means that the separation effect is the best at this time. The threshold t at this time is the optimal threshold T1 of Otsu.

[0033] A2: The local threshold T2 of each pixel in the second image is calculated using the NiBlack algorithm. This algorithm can effectively handle uneven lighting and surface texture changes. For any pixel in the second image, a fixed window is set. The threshold of any pixel is determined by the local mean and local standard deviation within the neighborhood window of that pixel. The expression for the local threshold is as follows:

[0034] In the formula: It is a local mean. The standard deviation is denoted as k, which is an adjustment coefficient and can be fine-tuned according to the actual situation.

[0035] Through extensive experimental verification, this invention has determined the optimal value range for the key parameters of the algorithm: the neighborhood window size b, which ranges from 15×15 to 35×35 pixels; with a preferred window size of 25×25 pixels; this range effectively balances noise suppression and detail preservation. The adjustment coefficient k ranges from 0.3 to -0.1; in this embodiment, a preferred value is k=-0.2. This parameter range can be used to optimize the algorithm's sensitivity to dark-colored disease areas (such as cracks and holes).

[0036] For example, the weighting of the global threshold and the local threshold to obtain the mixed threshold includes:

[0037] in, The weighting parameter is calculated using the following formula:

[0038] In the formula: The global grayscale standard deviation of the second image is denoted as , which reflects the overall contrast and texture undulation of the image. The larger the value, the more significant the overall features of the image; The texture complexity index of the second image quantifies the intensity and heterogeneity of local texture variations in the image. The larger the value, the more complex the local changes in the image and the coarser the texture. The adjustment constant is set to a value ranging from 0.3 to 0.7, with 0.5 typically used as the initial value. This dynamic strategy ensures the algorithm's adaptability to concrete surfaces with different roughness.

[0039] Among them, the Calculated through the following process: Calculate the local standard deviation map S: Using the same window size as in the Niblack algorithm (e.g., 25×25 pixels), traverse each pixel (i,j) in the second image and calculate the standard deviation of the gray values ​​of all pixels within its neighborhood window to obtain the local standard deviation S of that point. (i,j) Its formula is:

[0040] In the formula: N is Total number of pixels within the window The grayscale value of a pixel. The average grayscale value of the pixels within the window; all S (i,j) This forms a local standard deviation mapping.

[0041] Next, calculate the standard deviation of all values ​​in the entire local standard deviation mapping S, that is:

[0042] In the formula: M is the total number of pixels in the mapping image S. Let be the mean of all s(i,j). This value reflects the degree of dispersion of local texture changes in the image; a larger value indicates a rougher surface or more complex disease features.

[0043] For example, to ensure the effectiveness of the blending threshold, it is ultimately necessary to constrain the blending threshold within a reasonable range determined by the image grayscale distribution, i.e.:

[0044] Meanwhile, the step of obtaining the disease image is as follows: The pixel grayscale value is compared with the calculated mixing threshold. If the grayscale value of the pixel is less than the mixing threshold, it is determined to be a diseased area and marked as black (0). If the gray value of the pixel is greater than or equal to the mixing threshold, it is considered an intact area and marked as white (255). Finally, the disease image is obtained. In the final disease image, the disease area is significantly highlighted, laying the foundation for subsequent feature extraction.

[0045] S103: The improved Faster R-CNN algorithm is used to identify concrete defects in the defect images to obtain different defect areas and their categories. Concrete defect features are extracted from the defect images, including: total crack length, spalling area, and exposed steel reinforcement ratio.

[0046] Furthermore, the specific steps for identifying concrete defects in the defect images using the improved Faster R-CNN algorithm are as follows: Convolutional feature maps are extracted from the disease images using a convolutional neural network, and a region proposal network is introduced to generate multiple candidate regions. It should be further explained that Faster R-CNN consists of four main parts: a convolutional neural network, a region proposal network, ROI pooling, and a classifier.

[0047] The convolutional neural network (CNN) comprises 13 convolutional layers, 4 pooling layers, and 13 ReLU activation functions. All convolutional layers have a kernel size of 3 and a stride of 1, with edge expansion (filling with a ring of zeros). All pooling layers have a kernel size of 2 and a stride of 2. The CNN (CONV layer) is used to extract features from the lesion image, with the entire lesion image as input and the extracted features as output. The resulting lesion image exhibits high complexity and high recognizability. Therefore, this invention uses the ImageNet pre-trained VGG-16 model as the base network for the classification task. The input size of VGG is 1280*720*3.

[0048] Using the VGG-16 model, which has been pre-trained on the ImageNet dataset, as the base network, VGG-16 extracts high-level features of images through its multiple convolutional layers and applies these features to image classification tasks. The pre-trained VGG-16 can effectively learn general image features, thereby improving classification accuracy in new tasks and avoiding training the model from scratch. Through transfer learning, the model can achieve good performance on less training data, reducing the demand for computing resources.

[0049] The Region Proposal Network (RPN) in this embodiment abandons the traditional sliding window method and directly uses RPN to generate detection boxes. This is a significant advantage of Faster R-CNN, greatly improving the generation speed of detection boxes. The RPN processes the extracted convolutional feature maps to find regions that may contain a predefined number of objects. It is a convolutional network that takes feature maps from the Conv layer as input and outputs proposed regions. In the RPN, anchor points are fixed-size boxes placed on the image using various sizes and scales as reference boxes for the first prediction of object positions. The output of the binary classification layer is the probability of whether each anchor point is background. The regression layer of the RPN outputs the position information of the anchor boxes that match the predicted objects.

[0050] Furthermore, the concrete defect features are extracted from the defect images using image analysis technology. These features include the total length of cracks, the area of ​​spalling, and the proportion of exposed rebar. This invention primarily identifies the following three types of concrete defects: Cracks: These cracks are caused by material shrinkage, load, or environmental factors. In the defect images, cracks typically appear as long, dark lines with a clear color contrast. Spalling area: This refers to the localized loss of material from the concrete surface, forming recessed areas. These areas are caused by impact, freeze-thaw cycles, or chemical erosion. In the defect images, the spalling area typically appears as an area with a significant difference in color and texture from the surrounding concrete. Exposed rebar: This refers to the proportion of internal rebar exposed on the surface after the concrete cover is damaged. This defect is caused by concrete spalling and other concrete defects or construction defects. In the defect images, the exposed rebar area typically appears as a dark-colored and regularly shaped linear or blocky area.

[0051] Meanwhile, the extraction of the disease features can also be achieved through a combination of one or more of the following methods: Crack features: By analyzing the connected components in the disease image, the pixel length is calculated and converted into actual physical dimensions (e.g., millimeters) according to camera calibration parameters, thereby obtaining the crack length and maximum width. Peeling area: By counting the pixels of the identified peeling areas in the disease image and converting them into actual physical areas (e.g., square meters) according to camera calibration parameters; Peeling depth and 3D morphology: High-precision 3D point cloud data of the pier surface can be obtained through a laser scanner or structured light 3D reconstruction camera, thereby directly measuring the depth and volume of the peeling area and achieving a more accurate quantification of the severity of the disease; Reinforcing steel exposure and corrosion: Based on a machine learning image segmentation model, pixel-level recognition is performed on the preprocessed color image (second image) to distinguish between the concrete background and exposed reinforcing steel and corrosion areas, and the pixel ratio is calculated and then converted into actual area ratios.

[0052] ROI pooling is used to transform multiple candidate regions into pooled feature maps of a fixed size. Then, the disease category probability corresponding to each feature value is calculated based on the pooled feature maps, and the multiple candidate regions are classified accordingly.

[0053] In this embodiment, the ROI pooling converts inputs of different sizes into fixed-length outputs. The inputs are different candidate regions. Interpolation is used to fix the cropped feature map to a size of 14*14*512. After max pooling, the output is a fixed-size pooled feature map of 7*7*512.

[0054] The classifier uses the obtained feature maps to calculate the probability that each feature value belongs to a specific category. Its output is the category of the candidate region and the exact location of the candidate region in the image. In Faster R-CNN, its role is similar to the fully connected layer of a conventional convolutional neural network. The two output layers contain n+1 and 4n neurons respectively; one outputs a score for each candidate object, and the other outputs the location information of the n-class predicted boxes. This network is consistent with the RPN function, so weight sharing between the two networks can greatly improve computational speed.

[0055] S104: As Figure 5 As shown, values ​​are assigned to the total length of the crack, the spalling area, and the exposed ratio of the reinforcing bars, respectively, and the hazard score of the pier is calculated. Based on the hazard score, the warning level is determined, and corresponding maintenance measures are taken.

[0056] The specific process for calculating the risk score of the pier is as follows: The total length of the crack, the spalling area, and the exposed rebar ratio are respectively denoted as B1, B2, and B3, and the initial risk score is calculated using the following formula:

[0057] In the formula: These are measured values. For design limits, This is an abnormal indicator. , The weights corresponding to the total crack length, spalling area, and exposed rebar ratio; It should be further clarified that the design limits are not arbitrarily set, but are strictly determined according to the limits for durability and safety requirements of concrete structures in the current technical standards and specifications issued by authoritative departments such as the China National Railway Administration and the Ministry of Transport. Specific values ​​are as follows: For the design limit of the total crack length Lc: its core control indicator is the crack width. According to the "Code for Durability Design of Railway Concrete Structures" (TB10005-2010) and the "Code for Inspection of Railway Bridges" (Tieyun

[2017] No. 9), under normal atmospheric conditions, the limit for the width of stress-induced cracks in reinforced concrete bridge piers is generally 0.20 mm. This invention uses this as its core, and through calibration and conversion of the pixel size of the crack image with the actual physical size, derives the allowable limit of the total crack length. For the design limit of the spalling area Sd: it mainly refers to the evaluation standard for concrete component defects in the "Standard for Technical Condition Assessment of Highway Bridges" (JTG / TH21-2011). When the cumulative area of ​​spalling and corner chipping exceeds 5% of the component area, or the area of ​​a single location is greater than 0.5 m², the limit is set at 0.20 mm. 2 When this occurs, the component is considered to have serious defects. This invention, combining the above standards, sets it as a specific percentage (e.g., 3%-5%) or absolute area value (e.g., 0.5m²) of the pier's surface area. 2 The specific values ​​can be adjusted according to the pier dimensions. Regarding the design limit for the exposed rebar ratio (Re): this indicator directly reflects the severity of protective layer failure. According to the aforementioned specifications, any degree of rebar exposure (corroded or uncorroded) is considered a serious defect. Therefore, this invention sets a very small threshold (e.g., >0%), and once rebar exposure is detected, it is considered to have triggered an abnormal indicator (m=1) for this defect. The above upper limits are general guidelines. In practical applications, the system can be fine-tuned within the above standard framework based on the specific design age of the bridge, the type of environmental erosion (carbonation, chloride erosion, etc.), and the importance level, to ensure the accuracy and engineering relevance of the assessment results.

[0058] The weights corresponding to the total crack length, spalling area, and exposed rebar ratio are as follows: In this embodiment, the weighting coefficients are determined using an expert scoring method combined with the Analytic Hierarchy Process (AHP). The specific determination process is as follows: First, a judgment matrix M is constructed regarding the importance of the three types of defects to structural safety. This matrix is ​​obtained by experts through pairwise comparisons based on a 1-9 scale. The specific form of matrix M is shown below:

[0059] Then, the largest eigenvalue of the judgment matrix is ​​calculated. The initial weight coefficients of each indicator are obtained by normalizing the eigenvectors and their corresponding features. To ensure the rationality of the judgment logic, a consistency check is required, and the consistency ratio (CR) must be calculated, which must satisfy CR < 0.1. The calculated CR of this judgment matrix is ​​0.051 < 0.1, passing the consistency check and proving that the expert evaluation opinions have good consistency. Finally, the weight coefficients of each disease feature are determined as follows: cracks... =0.5, peeling area =0.3, exposed rebar =0.2.

[0060] By introducing a disease rate factor, the initial risk score is corrected to obtain the risk score of the pier column.

[0061] The disease rate factor is introduced to correct the initial risk score. The correction steps are as follows: The disease rate factor is calculated using the following formula:

[0062] In the formula: j and i represent the sampling number, which is the time sequence number of the image acquisition; , The peeling area is extracted from the j-th and i-th samples. Indicates a fixed sampling interval; Based on the disease rate factor, determine whether to apply an additive correction to the initial risk score, and the correction rules are as follows: If the initial risk score Then, based on the initial risk score, add N; If the initial risk score Then, based on the initial risk score, add M; In other cases, the risk score of the pier remains consistent with the initial risk score, and in this embodiment, M and N are 20 and 15, respectively.

[0063] It should be noted that the condition for determining whether to add a correction to the initial hazard score is as follows: >0.5m 2 / sky; It should be understood that this threshold is not set arbitrarily, but is based on a comprehensive judgment of statistical analysis of numerous historical bridge defects, mechanical models of concrete material failure, and the engineering experience of experts in the field. Studies have shown that when the concrete spalling rate consistently exceeds this critical value, it often indicates that internal steel reinforcement corrosion has entered an accelerated phase or the structural bearing capacity is significantly declining. The risk of the defect development shifting from quantitative to qualitative change increases dramatically, marking the beginning of a "rapid expansion phase." Therefore, using this threshold as a condition to trigger the dynamic scoring linkage mechanism has sufficient engineering practice and theoretical basis, and can effectively ensure the system's sensitivity to major safety hazards.

[0064] Meanwhile, the warning level and the maintenance measures include: If the pier is rated as dangerous The warning level is Level 5, corresponding to a stable high-speed railway bridge pier structure with extremely low damage, requiring no warning. In this case, the system will continue monitoring but will not issue an alarm to avoid unnecessary intervention; it will only perform continuous monitoring. If the pier is rated as dangerous The warning level is Level 4, indicating minor concrete defects, but the high-speed railway bridge piers remain stable and no immediate warning is required. In this case, the system may record defect data for trend analysis, but will not take emergency measures and will conduct monthly manual reviews. If the pier is rated as dangerous The warning level is Level 3, corresponding to situations where concrete defects cannot be ignored, requiring close monitoring and alerting staff for manual inspection. This level of warning may trigger regular manual inspections to assess the specific condition of the concrete defects, including weekly patrols and alert notifications. If the pier is rated as dangerous The warning level is Level II, corresponding to severe concrete defects. At this stage, there are serious safety hazards, requiring immediate alerting and feedback to staff. This level of warning may necessitate immediate inspection and necessary maintenance measures, including emergency testing within 48 hours. If the pier is rated as dangerous The warning level is Level 1, which corresponds to a very serious situation of concrete defects. In this case, an immediate warning is required to remind staff to come for maintenance as soon as possible and to close the road immediately.

[0065] Through this detailed classification of early warning levels, the early warning system can issue timely and accurate warnings based on the actual condition of the concrete, thereby ensuring the safe operation of high-speed railway bridge piers. This design considers different levels of risk and configures corresponding response measures for each level, ensuring the efficiency and effectiveness of monitoring and maintenance work.

[0066] For a clearer understanding of the warning levels and maintenance measures, please refer to Table 1: Table 1: Early Warning and Maintenance Measures

[0067] In this embodiment, a comparison mechanism can be set before issuing an alert. This comparison mechanism includes: repeating steps S101 to S103 at least twice within a fixed time period to obtain at least two corresponding alert levels; then comparing each alert level; if each alert level is the same, executing the alert in step S104, thereby ensuring the accuracy of the alert information. If the difference between at least two alert levels is large, for example, if steps S101 to S103 are repeated three times to obtain three alert levels, but the three alert levels are all different, then it is necessary to retest.

[0068] To improve the accuracy of monitoring and early warning, an error elimination step can be added to the comparison mechanism: After repeating steps S101 to S103 multiple times, for example, five times, five corresponding early warning levels are obtained. If most of the early warning levels are the same, and only one or two are different, then these one or two different levels are considered errors and eliminated. The remaining majority of the same early warning levels are then used as the final early warning levels. The errors eliminated here may be caused by sudden changes in the environment or external electromagnetic interference. After elimination, the accuracy of monitoring is ensured, and there is no need to re-monitor, saving resources.

[0069] Example 2 Please see Figure 2 As shown, this embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the machine vision-based method for detecting concrete defects in high-speed railway bridge piers provided by the above methods.

[0070] Since the electronic device described in this embodiment is the one used to implement the machine vision-based high-speed railway bridge pier concrete defect damage detection method in this application embodiment, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the machine vision-based high-speed railway bridge pier concrete defect damage detection method described in this application embodiment. Therefore, how the electronic device implements the method in this application embodiment will not be described in detail here. Any electronic device used by those skilled in the art to implement the machine vision-based high-speed railway bridge pier concrete defect damage detection method in this application embodiment falls within the scope of protection of this application.

[0071] Example 3 Please see Figure 3 As shown, this embodiment discloses a computer-readable storage medium, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the computer program, it implements the machine vision-based method for detecting concrete defects in high-speed railway bridge piers provided by the above methods.

[0072] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters, weights, and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0073] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via a wired or wireless network. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0074] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented in electronic hardware or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0075] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0076] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only one method, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0077] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0078] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0079] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0080] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A machine vision-based method for detecting concrete defects and damage in high-speed railway bridge piers, characterized in that, The method includes: A first image of the pier concrete is obtained by taking pictures of the pier surface from multiple angles. The first image is preprocessed and output as a second image. Based on the second image, a global threshold and a local threshold are calculated respectively. The global threshold and the local threshold are weighted to obtain a mixed threshold. The second image is then segmented according to the mixed threshold to obtain a disease image. The improved Faster R-CNN algorithm is used to identify concrete defects in the defect images to obtain different defect areas and their categories. Concrete defect features are extracted from the defect images, including: total crack length, spalling area, and exposed steel reinforcement ratio. The total length of the crack, the spalling area, and the exposed ratio of the reinforcing bars are assigned values ​​respectively, the hazard score of the pier column is calculated, and the warning level is determined based on the hazard score.

2. The method for detecting concrete defects in high-speed railway bridge piers based on machine vision according to claim 1, characterized in that, The calculation of the global threshold and local threshold based on the second image includes: A1: Iterate through all possible grayscale values ​​of the second image, and determine the optimal segmentation threshold by maximizing the inter-class variance between the damaged region and the background region. This threshold is the global threshold, and its objective function is: ； In the formula: , These represent the pixel percentages of the damaged area and the background area, respectively. , This indicates its average gray value; A2: For any pixel in the second image, a fixed window is set. The threshold of any pixel is determined by the local mean and local standard deviation within the neighborhood window of that pixel. The local threshold can be obtained by the following expression: ； In the formula: It is a local mean. denoted as the local standard deviation, and k is the adjustment coefficient.

3. The method for detecting concrete defects in high-speed railway bridge piers based on machine vision according to claim 2, characterized in that, The step of weighting the global threshold and the local threshold to obtain a mixed threshold includes: ； in, The weighting parameter is calculated using the following formula: ； In the formula: The global grayscale standard deviation of the second image is . Let be the texture complexity index of the second image. This is the adjustment constant.

4. The method for detecting concrete defects in high-speed railway bridge piers based on machine vision according to claim 3, characterized in that, The steps for obtaining the disease image are as follows: The pixel grayscale value is compared with the calculated mixing threshold. If the grayscale value of the pixel is less than the mixing threshold, it is determined to be a diseased area and marked as black. If the grayscale value of the pixel is greater than or equal to the mixing threshold, it is considered a good area and marked as white, thus obtaining the disease image.

5. The method for detecting concrete defects in high-speed railway bridge piers based on machine vision according to claim 1, characterized in that, The specific steps for using the improved Faster R-CNN algorithm to identify concrete defects in the defect images are as follows: Convolutional feature maps are extracted from the disease images using a convolutional neural network, and a region proposal network is introduced to generate multiple candidate regions. ROI pooling is used to transform multiple candidate regions into pooled feature maps of a fixed size. Then, based on the pooled feature maps, the disease category probability corresponding to each feature value is calculated, and the multiple candidate regions are classified.

6. The method for detecting concrete defects in high-speed railway bridge piers based on machine vision according to claim 1, characterized in that, The specific process for calculating the risk score of the pier is as follows: The total length of the crack, the spalling area, and the exposed rebar ratio are respectively denoted as B1, B2, and B3, and the initial risk score is calculated using the following formula: ； In the formula: These are measured values. Design limits, This is an abnormal indicator. The weights corresponding to the total crack length, spalling area, and exposed rebar ratio; By introducing a disease rate factor, the initial risk score is corrected to obtain the risk score of the pier column.

7. The method for detecting concrete defects in high-speed railway bridge piers based on machine vision according to claim 6, characterized in that, The introduction of a disease rate factor to correct the initial risk score involves the following steps: The disease rate factor is calculated using the following formula: ； In the formula: j and i represent the sampling number, which is the time sequence number of the image acquisition; , The peeling area is extracted from the j-th and i-th samples. Indicates a fixed sampling interval; Based on the disease rate factor, determine whether to apply an additive correction to the initial risk score, and the correction rules are as follows: If the initial risk score Then, based on the initial risk score, add N; If the initial risk score Then, based on the initial risk score, add M; In other cases, the risk score of the pier remains consistent with the initial risk score.

8. The method for detecting concrete defects in high-speed railway bridge piers based on machine vision according to claim 1, characterized in that, The warning levels include: If the pier is rated as dangerous The warning level is level five, and it will be continuously monitored. If the pier is rated as dangerous The warning level is level four, and monthly manual review is required. If the pier is rated as dangerous The warning level is level three, and weekly inspections and patrol notifications will be conducted accordingly. If the pier is rated as dangerous The warning level is Level II, and emergency testing will be conducted within 48 hours. If the pier is rated as dangerous The warning level is Level 1, and traffic should be closed immediately.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the machine vision-based method for detecting concrete defects in high-speed railway bridge piers as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed, implements the machine vision-based method for detecting concrete defects in high-speed railway bridge piers as described in any one of claims 1 to 8.

Images