Systematic cable defect detection and durability improvement method and system based on humidity control

Abstract

The present application relates to the technical field of image analysis, in particular to a system for detecting defects of cable-stayed cables and improving durability based on humidity control, comprising the following steps: collecting images of cable-stayed cables, constructing regional division, analyzing gray boundary and brightness direction, judging consistency of interruption area, distinguishing decayed morphology, adjusting parameters and collecting again, aligning and comparing images, adjusting decayed contour, identifying decayed area, and generating surface decayed defect detection results.In the present application, the correspondence between gray boundary, brightness direction and pixel connection direction is constructed to enhance the clear expression of regional structure, the consistency of boundary morphology and brightness of interruption area is analyzed to realize accurate distinction of decayed types, the image parameter adjustment and regional alignment are used to improve the degree of detail restoration of target area, the dynamic correction of decayed contour is completed through the analysis of boundary position changes of images before and after, and the accuracy and integrity of decayed area identification are improved in combination with boundary closure and texture structure distribution.

Description

Technical Field

[0001] This invention relates to the field of image analysis technology, and in particular to a systematic method and system for detecting defects and improving the durability of cable-stayed bridges based on humidity control. Background Technology

[0002] Image analysis technology encompasses methods for acquiring, processing, and understanding image data. It aims to assist in tasks such as classification, recognition, detection, and tracking by identifying and extracting targets, features, and structures from images. This technology includes core aspects such as image acquisition, image preprocessing, feature extraction, target recognition, image segmentation, and pattern recognition, and is widely applied in industrial inspection, security monitoring, medical diagnosis, traffic management, and intelligent manufacturing. Image analysis enhances the understanding of image information through computer vision and artificial intelligence, transforming images from perceptual data into structured information, thereby supporting decision-making and control. Visual information-based defect detection is a typical application, playing a crucial role, especially in infrastructure and equipment maintenance. Traditional intelligent detection methods for surface corrosion defects in cable-stayed bridges refer to the identification and judgment of defects caused by corrosion, wear, and spalling on the surface of cable-stayed bridges or large structures. These methods typically involve manual visual inspection or close-range photography for analysis. These methods rely on the experience of the inspectors, and images are often acquired using ordinary cameras. Defect analysis often relies on personnel comparing reference images and marking abnormal areas, resulting in low detection efficiency, high subjectivity, and poor repeatability.

[0003] Existing technologies lack a systematic division of image information organization, making it difficult to accurately express the features of eroded areas. Grayscale and brightness changes are not clearly correlated, making it difficult to identify feature differences within the area. The image acquisition process fails to adjust parameters for specific areas, resulting in insufficient capture of image details, which affects the completeness and clarity of subsequent feature extraction. Corroded areas cannot be compared in terms of positional changes, making it difficult to grasp the evolution trend of defects. Type judgment relies heavily on the overall contour shape of the image, lacking effective texture separation and closure confirmation methods, which affects the accuracy and comprehensiveness of identification. Corrosion detection lacks a traceable multi-stage analysis mechanism, failing to meet the systematic processing requirements for structural surface defect identification. Summary of the Invention

[0004] To achieve the above objectives, the present invention adopts the following technical solution: a systematic method for detecting defects and improving the durability of cable-stayed bridges based on humidity control, comprising the following steps: S1: Collect surface images of the cable-stayed bridge, construct regular region division relationships, analyze the gray-scale change boundaries, brightness change directions and pixel connection directions within the regions, and generate surface region distribution images based on the relationship between pixel connection directions and brightness change directions. S2: Determine the consistency between the grayscale change boundary and the brightness change direction of the pixel connection interruption area in the surface region distribution image, distinguish between spot distribution pattern, block distribution pattern and strip distribution pattern, and generate a surface corrosion type identification image; S3: Based on the surface corrosion type identification image, adjust the brightness channel, gain channel and exposure channel of the image acquisition device, re-acquire the corrosion area image marked in the surface corrosion type identification image, perform area alignment, generate the corrosion area and then acquire the image again. S4: Compare the image of the corrupted region with the image of the surface region distribution, calculate the change in the boundary position of the corrupted region, adjust the outline of the corrupted region according to the continuity and transition distribution, and generate a corrupted region outline image. S5: Based on the outline image of the eroded area, determine the relationship between the boundary closure of the eroded area and the distribution of texture blocks, identify the overall eroded area, and generate the surface erosion defect detection result.

[0005] As a further aspect of the present invention, the surface region distribution image includes grayscale boundary distribution information, brightness direction information, and pixel connection structure; the surface corrosion type identification image includes spot type identification, block type identification, and strip type identification; the corrosion region re-acquisition image includes brightness adjustment image, contrast adjustment image, and exposure compensation image; the corrosion region contour image includes boundary continuity features, boundary transition features, and position change features; and the surface corrosion defect detection result includes closed boundary defects, texture block distribution defects, and overall corrosion region defects.

[0006] As a further aspect of the present invention, adjusting the contour of the eroded region based on continuity and transition distribution refers to dynamically optimizing the boundary path and correcting the initially extracted eroded contour lines based on the gray-level continuity and local transition direction distribution of the boundary of the eroded region in the pixel image.

[0007] As a further aspect of the present invention, the execution region alignment refers to aligning the re-acquired corrupted region image with the initial region distribution image in spatial coordinates through feature point matching and image registration algorithms, thereby verifying the consistency and comparability of the corrupted information.

[0008] As a further aspect of the present invention, the specific steps of S1 are as follows: S101: Based on the acquired images of the cable-stayed bridge surface, construct an image pixel array and divide it into regular grid regions, extract the gray value distribution within the regions, calculate the gray value difference between adjacent pixels and compare it with a preset gray value change threshold, filter gray value change regions, and generate a gray value change boundary matrix. S102: Based on the pixel positions in the grayscale change boundary matrix, extract the grayscale change direction and pixel connection direction of adjacent regions, calculate the angle between them and compare it with a preset continuous change direction threshold, filter pixel paths with the same direction, and obtain a set of brightness change path sequences. S103: Call the pixel paths in the brightness change path sequence set, extract the corresponding region blocks and aggregate and connect them, reconstruct the regional spatial relationship in the image, and obtain the surface region distribution image.

[0009] As a further aspect of the present invention, the specific steps of S2 are as follows: S201: Based on the pixel connection interruption region in the surface region distribution image, extract the corresponding gray-scale change boundary direction and brightness change direction vector, calculate the angle between them and compare it with a preset direction consistency threshold, filter the pixel index set whose angle meets the threshold condition, and generate a boundary direction consistency marker set. S202: Call the pixel index in the boundary direction consistency mark set, perform structural scale calculation and pixel arrangement statistics on the region corresponding to the index, and judge based on the region size parameter and connectivity extension parameter to divide the region structure into three categories: spots, blocks and stripes, and obtain the erosion morphology classification label group. S203: Based on the pixel regions corresponding to the labels in the corrosion morphology classification label group, perform region mapping and label encoding configuration in the surface region distribution image, integrate image pixel annotations, and establish a surface corrosion type label image.

[0010] As a further aspect of the present invention, the specific steps of S3 are as follows: S301: Based on the pixel position of the corroded area in the surface corrosion type identifier image, retrieve the image acquisition configuration parameters, adjust the current values ​​of the brightness channel, gain channel and exposure channel, and set the corresponding parameter set according to the area coordinates to generate the image acquisition parameter configuration group. S302: Call the configuration parameters in the image acquisition parameter configuration group to re-acquire the image of the marked eroded area, extract the pixel data frame of the corresponding coordinate area in the newly acquired image, establish the image frame set under the current channel parameters, and obtain the area re-acquired image frame group. S303: Based on the spatial coordinates of the image frames in the reacquired image frame group and the original image position, perform regional boundary mapping and pixel matrix alignment operations, perform spatial correction processing between the acquired image and the original region, and obtain the reacquired image of the eroded region.

[0011] As a further aspect of the present invention, the specific steps of S4 are as follows: S401: Based on the corresponding pixel positions in the re-acquired image of the eroded region and the distribution image of the surface region, extract the set of pixel coordinates of the boundary of the eroded region, and map them according to a unified coordinate system. Then, perform pixel pairing and comparison of the boundary contours of the same region in the two images to obtain the set of boundary position correspondences. S402: Call the coordinate offset values ​​in the boundary position correspondence set, calculate the spatial change trend of adjacent boundary points, identify continuous boundary segments and turning point regions based on the angle between adjacent offset vectors, mark the location and type of the turning structure, and obtain the boundary change structure information set. S403: Based on the location and structure label in the boundary change structure information set, perform node position replacement and path connectivity adjustment on the boundary contour in the re-acquired image of the eroded region, perform boundary reconstruction and closure processing of the eroded region, and obtain the contour image of the eroded region.

[0012] As a further aspect of the present invention, the specific steps of S5 are as follows: S501: Based on each closed contour structure in the eroded region contour image, calculate the boundary closure index and pixel connectivity parameter, determine whether the spatial distance between the boundary start point and the end point is lower than the preset closure judgment threshold, and perform closure recognition in combination with the pixel difference inside and outside the boundary to obtain the eroded boundary closure mark group. S502: Call the boundary closure region in the erosion boundary closure marker group, perform pixel texture fitting degree evaluation on the internal region, determine whether the texture block is independent of the erosion structure based on the texture block direction distribution parameters and area ratio, mark the position of the separable texture block, and generate a erosion texture block separation identifier set. S503: Based on the position index information in the corrosion texture block separation identifier set and the corrosion boundary closure marker group, perform corrosion structure aggregation processing and separable region removal operation, reconstruct the corrosion structure range layer, and establish surface corrosion defect detection results.

[0013] A systematic cable-stayed bridge defect detection and durability improvement system based on humidity control includes: The region segmentation module is used to implement S1: acquiring surface images of the cable-stayed bridge, constructing regular region segmentation relationships, analyzing grayscale change boundaries, brightness change directions and pixel connection directions within the regions, and generating surface region distribution images based on the relationship between pixel connection directions and brightness change directions. The type recognition module is used to implement S2: determine the consistency between the grayscale change boundary and the brightness change direction of the pixel connection interruption area in the surface region distribution image, distinguish between spot distribution pattern, block distribution pattern and strip distribution pattern, and generate a surface corrosion type identification image; The image re-acquisition module is used to implement S3: based on the surface corrosion type identification image, adjust the brightness channel, gain channel and exposure channel of the image acquisition device, re-acquire the corrosion area image marked in the surface corrosion type identification image, perform area alignment, and generate a corrosion area re-acquisition image; The contour extraction module is used to implement S4: compare the re-acquired image of the eroded region with the distribution image of the surface region, calculate the change in the boundary position of the eroded region, adjust the contour of the eroded region according to the continuity and transition distribution, and generate a contour image of the eroded region. The defect detection module is used to implement S5: based on the outline image of the eroded area, determine the relationship between the boundary closure of the eroded area and the distribution of texture blocks, identify the overall eroded area, and generate surface erosion defect detection results.

[0014] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, by constructing the correspondence between grayscale boundaries, brightness direction, and pixel connection direction, the clear expression of regional structure is enhanced. By analyzing the boundary morphology and brightness consistency of the interrupted area, the accurate differentiation of corruption types is achieved. By adjusting image parameters and aligning regions, the detail reproduction of the target region is improved. By analyzing the changes in the boundary position of the preceding and following images, the dynamic correction of the corruption contour is completed. By combining boundary closure and texture structure distribution, the accuracy and completeness of corruption region recognition are improved. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the steps of the present invention; Figure 2 This is a detailed schematic diagram of S1 of the present invention; Figure 3 This is a detailed schematic diagram of S2 of the present invention; Figure 4 This is a detailed schematic diagram of S3 of the present invention; Figure 5 This is a detailed schematic diagram of S4 of the present invention; Figure 6 This is a detailed schematic diagram of S5 of the present invention; Figure 7 This is a system module diagram of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will now be described with reference to the accompanying drawings.

[0018] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0019] Please see Figure 1 This invention provides a systematic method for detecting defects and improving the durability of stay cables based on humidity control, comprising the following steps: S1: By acquiring surface images of the cable-stayed bridge and constructing regular region division relationships in the images, we analyze the grayscale change boundaries, brightness change directions, and pixel connection directions in each region. Based on the correspondence between pixel connection directions and brightness change directions, we construct a surface region distribution image. S2: For areas in the surface region distribution image that show interrupted pixel connections, the distribution pattern of grayscale change boundaries is judged based on the consistency of the direction of brightness change, and the distribution patterns of spots, blocks, and stripes are distinguished to generate a surface corrosion type identification image. S3: Based on the corrosion areas marked in the surface corrosion type identification image, adjust the brightness channel, gain channel and exposure channel of the image acquisition device accordingly, re-acquire images of the marked areas, and perform area alignment processing on the re-acquired images to generate images of the corrosion areas for re-acquisition. S4: Compare the pixel correspondence between the re-acquired image of the corrupted region and the surface region distribution image, calculate the positional change relationship of the corrupted region boundary in the two images, and adjust the contour of the corrupted region based on the continuity and turning distribution of the boundary positional change to generate a corrupted region contour image. S5: Based on the contour image of the eroded region, perform boundary closure judgment and texture block separation judgment on the eroded region, and identify the overall eroded region according to the spatial distribution relationship between the closed boundary and the texture block to generate surface erosion defect detection results; To overcome the interference of high humidity outdoor environments on image quality, an environmental adaptability control module is specially configured. This module mainly includes a high-precision humidity sensor and a dehumidification actuator, which can be a ceramic heating element or a miniature fan. During operation, the humidity sensor collects the ambient relative humidity value in real time and transmits this value to the central controller for comparison with a preset humidity safety threshold (e.g., 85% relative humidity). Once the real-time ambient relative humidity value is detected to be higher than the preset humidity safety threshold, the central controller will immediately instruct the dehumidification actuator to start, physically dehumidifying the image acquisition window until the ambient relative humidity value drops below the stop threshold. This control logic ensures that image data is effectively acquired and saved only when the ambient humidity meets the requirements, thereby preventing image blurring caused by lens fogging and ensuring the data input quality of subsequent defect recognition algorithms.

[0020] The surface region distribution image includes grayscale boundary distribution information, brightness direction information, and pixel connection structure. The surface corrosion type identification image includes spot type identification, block type identification, and strip type identification. The corrosion region re-acquired image includes brightness adjustment image, contrast adjustment image, and exposure compensation image. The corrosion region contour image includes boundary continuity features, boundary transition features, and position change features. The surface corrosion defect detection results include closed boundary defects, texture block distribution defects, and overall corrosion region defects.

[0021] Please see Figure 2 The specific steps of S1 are as follows: S101: Based on the acquired images of the cable-stayed bridge surface, construct an image pixel array and divide it into regular grid regions, extract the gray value distribution within the regions, calculate the gray value difference between adjacent pixels and compare it with a preset gray value change threshold, filter gray value change regions, and generate a gray value change boundary matrix. The raw, high-resolution image data of the cable-stayed bridge surface was acquired using an image acquisition device equipped with a high-resolution charge-coupled device (CCD) sensor. The resolution of this image data was set to 4096 pixels multiplied by 2048 pixels, with a color depth of 24 bits. The process first converts the acquired raw color image data into a single-channel grayscale image matrix. The conversion logic employs a weighted average method, extracting the red, green, and blue channel values ​​for each pixel and assigning them weights of 0.3, 0.59, and 0.11 respectively. The sum of the products of these three values ​​is then used as the grayscale value of that pixel. For example, when a pixel has a red channel value of 100, a green channel value of 150, and a blue channel value of 200, the calculated grayscale value is 140.5, which is rounded down to 141. Subsequently, an image pixel array with the same size as the original image is constructed, and this array is divided into regular grid regions of 16×16 pixels, generating a total of 32768 independent grids. For each independent grid region, the grayscale distribution data of all 256 pixels within it are extracted, and the grayscale mean and standard deviation of that region are calculated. Next, the grayscale difference calculation operation between adjacent pixels is performed. For any coordinate point in the pixel array, its right-hand adjacent pixel and its lower-hand adjacent pixel are selected as comparison objects. The absolute value of the difference between the current pixel's grayscale value and the grayscale value of the pixel to its right, and the absolute value of the difference between the current pixel's grayscale value and the grayscale value of the pixel below, are calculated. The maximum of the two absolute differences is selected as the grayscale gradient intensity at that location. The process of obtaining the preset grayscale change amplitude threshold is as follows: A defect-free background region in the image is selected as the reference sample. The grayscale gradient intensity of all pixels in this sample region is statistically analyzed, and its arithmetic mean is calculated to be 10, and its standard deviation is 2. The sum of the arithmetic mean and three times the standard deviation is taken as the threshold, i.e., the threshold is set to 16. The calculated grayscale gradient intensity of each pixel is compared with the threshold of 16. If the gradient intensity is greater than 16, the pixel is determined to be in a region of drastic grayscale change and is marked as 1 in the newly created binary matrix; otherwise, it is marked as 0, thus generating a grayscale change boundary matrix corresponding to the original image size. Table 1 shows the grayscale statistics and threshold determination results for some grid areas.

[0022] Table 1. Gray-level statistics and threshold determination table for grid area. The units of gray-level mean, gray-level standard deviation, maximum calculated gradient and determination threshold are all image gray-level or gray-level difference. The gray-level is calculated based on the gray-level value of the image pixels.

[0023] As shown in Table 1, grid 002 was filtered as a grayscale variation area because its maximum gradient value exceeded the preset threshold.

[0024] S102: Based on the pixel positions in the grayscale change boundary matrix, extract the grayscale change direction and pixel connection direction of adjacent regions, calculate the angle between them and compare it with the preset continuous change direction threshold, filter the pixel paths with the same direction, and obtain the brightness change path sequence set. The grayscale change boundary matrix is ​​invoked, and the pixel positions marked as 1 are traversed. For each target pixel, a 3-pixel multiplied neighborhood window centered on it is established, and the grayscale change states of the eight neighboring pixels within the neighborhood are extracted. The grayscale change direction calculation for the adjacent regions is performed, calculating the horizontal and vertical gradient components separately. Specifically, the sum of the grayscale values ​​of the right column of pixels in the neighborhood is subtracted from the sum of the grayscale values ​​of the left column of pixels to obtain the horizontal grayscale change component; the sum of the grayscale values ​​of the lower row of pixels in the neighborhood is subtracted from the sum of the grayscale values ​​of the upper row of pixels to obtain the vertical grayscale change component. The arctangent function value of the ratio of the vertical to the horizontal grayscale change component is calculated to obtain the grayscale change direction angle of the pixel. Simultaneously, the pixel connection direction is extracted, i.e., the angle of the line connecting the current pixel to its predecessor connected pixels already determined in the boundary matrix. The absolute value of the difference between the grayscale change direction angle and the pixel connection direction angle is calculated to obtain the included angle value. A preset threshold for continuous change direction is set at 25 degrees. This threshold is based on statistical analysis of the smoothness of real cracks in numerous experiments, meaning that 95% of the directional abrupt changes between adjacent points at the edge of a real crack do not exceed 25 degrees. The calculated angle value is compared with 25 degrees. If the angle value is less than or equal to 25 degrees, the direction is considered consistent, and the pixel connection is retained. If the angle value is greater than 25 degrees, it is considered noise or irrelevant texture and is discarded. This logic is used to filter out all pixel connection paths with consistent directions, and the coordinates of these consecutive pixels are stored sequentially in a list to obtain a set of brightness change path sequences. For example, if a pixel's grayscale change direction is 45 degrees, and its connection direction with the preceding pixel is 40 degrees, the angle between them is 5 degrees, which is less than the threshold of 25 degrees, so this point is included in the path sequence. If another pixel's grayscale change direction is 90 degrees, and its connection direction is 45 degrees, the angle of 45 degrees is greater than the threshold, then the path is broken at this point.

[0025] S103: Call the pixel paths in the brightness change path sequence set, extract the corresponding region blocks and aggregate and connect them, reconstruct the regional spatial relationship in the image, and obtain the surface region distribution image; The system reads the individual pixel path data stored in the brightness change path sequence set. For each path, it identifies the pixel coordinate range it covers and maps it back to the original image array, extracting the corresponding rectangular region block. The size of this rectangular region block is set to the smallest bounding rectangle that can completely enclose the path, with an outward buffer distance of 2 pixels. An aggregation connection operation is performed using morphological dilation logic, setting the structuring element to a 3×3 pixel solid square. A binarized mask dilation operation is applied to all extracted region blocks, merging and connecting broken regions with a spatial distance of less than 3 pixels. Subsequently, connected component analysis is performed, merging all contacting or overlapping region blocks into a single connected object. During the reconstruction of the spatial relationships of regions in the image, a new blank layer is created, and the merged connected object is mapped to this layer, preserving its absolute coordinate position and geometry in the original image. This process not only restores the spatial continuity of discontinuous edges but also removes isolated noise regions with an area less than 10 pixels. The final obtained surface region distribution image is a binary mask image, where white areas represent the distribution range of potential defects or texture structures, and black areas represent the background. This image accurately reflects the spatial topology of gray-scale anomaly areas on the cable-stayed bridge surface.

[0026] Please see Figure 3 The specific steps of S2 are as follows: S201: Based on the pixel connection interruption region in the surface region distribution image, extract the corresponding gray-scale change boundary direction and brightness change direction vector, calculate the angle between the two and compare it with the preset direction consistency threshold, filter the pixel index set that meets the threshold condition, and generate the boundary direction consistency mark set. The surface region distribution image is scanned to identify the endpoints of all pixel connection breaks, i.e., coordinates of points where a pixel value exists but the number of connected pixels in its 8-neighborhood is less than 2. For each pair of interruption endpoints with a spatial distance of less than 50 pixels, the gray-level change boundary direction vector of its connected region is extracted. This vector is obtained through least squares linear fitting, selecting the coordinates of 10 consecutive boundary pixels near the endpoint and fitting a straight line; the direction of this line is the boundary direction vector. Simultaneously, the connection vector between two interruption endpoints, i.e., the brightness change direction vector, is calculated. The angle between the gray-level change boundary direction vector and the brightness change direction vector is calculated. A direction consistency threshold of 20 degrees is set, based on the extension characteristics of linear defects. The calculated angle value is compared with 20 degrees. If the calculated angle value is less than 20 degrees, the two interruption regions are considered to belong to the fracture part of the same potential defect, satisfying the merging condition, and the pixel indices of these two endpoints and their respective regions are recorded in the filtering results; if the angle is greater than 20 degrees, they are considered irrelevant independent features. For example, if the boundary orientation of endpoint A is 30 degrees, and the azimuth angle of endpoint B relative to A is 35 degrees, the included angle between them is 5 degrees, which is less than 20 degrees. Therefore, the indices of A and B are added to the set. The final generated boundary orientation consistency tag set contains a list of pixel indices for all identified as homologous fracture structures, providing a complete structural basis for subsequent morphological classification.

[0027] S202: Call the pixel index in the boundary direction consistency tag set, perform structural scale calculation and pixel arrangement statistics on the region corresponding to the index, and judge based on the region size parameter and connectivity extension parameter to divide the region structure into three categories: spots, blocks and stripes, and obtain the erosion morphology classification label group. The pixel indices from the boundary direction consistency marker set are used to extract the corresponding connected regions in the surface region distribution image. For each connected region, structural scale calculations are performed, specifically calculating the aspect ratio of the region's bounding rectangle, the region's solidity (the ratio of pixel area to bounding rectangle area), and the region's maximum Freette diameter. Classification is performed based on the region size parameters and connectivity extension parameters, with the following logic: if the region's aspect ratio is less than 1.5 and its solidity is greater than 0.8, it is classified as blobular erosion; if the region's aspect ratio is greater than 3.0, it is classified as striped erosion; if neither of these conditions is met, i.e., the aspect ratio is between 1.5 and 3.0, it is classified as blocky erosion. For example, a region with a bounding rectangle of 100 pixels long and 20 pixels wide has an aspect ratio of 5, which is greater than 3.0, and is directly classified as striped; another region with a length of 30 pixels and a width of 25 pixels has an aspect ratio of 1.2, a pixel area of ​​600, a bounding rectangle area of ​​750, and a solidity of 0.8, and is classified as blobular. The classification process directly generates corresponding corruption morphology classification label groups. Each label contains a region ID and a corresponding type identifier, such as "Type_Strip", "Type_Spot", or "Type_Block". Table 2 shows the calculation results of geometric parameters and classification conclusions for different regions.

[0028] Table 2 Classification parameters of morphological classification of decay areas

[0029] As shown in Table 2, the corrosion types were accurately classified based on geometric parameters.

[0030] S203: Based on the pixel regions corresponding to the labels in the corrosion morphology classification label group, perform region mapping and label encoding configuration in the surface region distribution image, integrate image pixel annotations, and establish a surface corrosion type labeling image. Based on the type information determined in the corrosion morphology classification label group, region mapping and label encoding configuration are performed on the surface area distribution image. For regions classified as "stripes," their pixel values ​​are assigned the red encoding value 255, 0, 0; for "spots," the green encoding value 0, 255, 0; and for "blocks," the blue encoding value 0, 0, 255. During image pixel annotation integration, all classified regions are traversed, and their corresponding color codes are written into a new three-channel label image with a black background. Simultaneously, a metadata file for the surface corrosion type label image is created, recording the center coordinates, area size, and category of each labeled region. This step completes the transformation from abstract geometric features to an intuitive visual layer. The generated surface corrosion type label image not only visually distinguishes different types of defects but also directly carries classification information through pixel values, facilitating subsequent steps to call differentiated acquisition parameters for different types of defects.

[0031] Please see Figure 4 The specific steps of S3 are as follows: S301: Based on the surface corrosion type, identify the pixel position of the corrosion area in the image, retrieve the image acquisition configuration parameters, adjust the current values ​​of the brightness channel, gain channel and exposure channel, and set the corresponding parameter set according to the area coordinates to generate the image acquisition parameter configuration group. The center coordinates and type codes of all non-background areas are extracted, and the preset image acquisition configuration parameter database is retrieved based on the type code. For "strip-type" defects, which typically have deep texture depth, the gain parameter in the image acquisition parameter configuration group is increased and the exposure time is reduced to enhance edge detail. Specifically, the gain parameter is set to 1.5 times its original value, and the exposure time is reduced by 20% to enhance edge sharpness. For "block-type" defects, to capture surface roughness details, the exposure time is appropriately increased in conjunction with gain parameter adjustments, with the exposure time increased by 10%. For "spot-type" defects, the default acquisition parameters are maintained, or the exposure time is appropriately reduced to prevent local overexposure. The corresponding parameter set is set according to the region coordinates. For example, for strip-type regions with coordinates ranging from 100,100 to 200,200, the exposure time is set to 12000 microseconds and the gain parameter to 6 dB; while for block-type regions at another coordinate, the exposure time is set to 15000 microseconds and the gain parameter to 4 dB. The generated image acquisition parameter configuration group is a list containing multiple "coordinate-parameter packages" used to guide the image acquisition device to perform targeted secondary image acquisition. An example of the actual parameter adjustment logic is as follows: Assume the initial exposure time is 10000 microseconds and the initial gain is 1.0; for strip-like regions, calculate the new exposure time as 10000 × 0.8, resulting in 8000 microseconds, and simultaneously calculate the new gain as 1.0 × 1.5, resulting in 1.5; for block-like regions, calculate the new exposure time as 10000 × 1.1, resulting in 11000 microseconds.

[0032] S302: Call the configuration parameters in the image acquisition parameter configuration group to re-acquire the image of the labeled eroded area, extract the pixel data frame of the corresponding coordinate area in the newly acquired image, establish the image frame set under the current channel parameters, and obtain the area re-acquired image frame group. The system executes actions according to the sequence in the image acquisition parameter configuration group. When the acquisition device moves to the specified coordinate area, it reads the specific parameter package corresponding to that area and dynamically adjusts the camera's shutter speed, ISO sensitivity, and light source brightness. After adjustment, it triggers an acquisition command to obtain a high-quality image of that local area. Pixel data frames of the corresponding coordinate area are extracted from the newly acquired image. Since secondary acquisition is usually only for local regions of interest, it is necessary to crop a sub-image from the full-frame image that matches the size of the original defect area. An image frame set under the current channel parameters is established, ensuring that each frame is associated with its specific parameter metadata from the time of acquisition. A group of reacquired image frames for the region is obtained. This group of data contains a sequence of optically optimized local images enhanced for specific defect types. Compared to the global image acquired in the first acquisition, its signal-to-noise ratio and feature saliency are improved by more than 30% through targeted parameter settings.

[0033] S303: Based on the spatial coordinates of the image frames in the reacquired image frame group and the original image position, perform regional boundary mapping and pixel matrix alignment operations, perform spatial correction processing between the acquired image and the original region, and obtain the reacquired image of the eroded region. The absolute spatial coordinates of each frame in the re-acquired image frame group are compared with the corresponding positions in the original surface region distribution image. Region boundary mapping is performed, and affine transformation logic is used to calculate the translation, rotation angle, and scaling ratio of the re-acquired image relative to the original image. Since minor positioning errors may occur due to device movement, a feature point matching algorithm is used to identify identical texture feature points in both images and calculate the transformation matrix. Pixel matrix alignment is performed, projecting the re-acquired local high-resolution image back to the original coordinate system through inverse transformation, covering the original low-quality areas. Spatial correction processing is performed between the acquired image and the original region, using bilinear interpolation to fill pixel gaps generated during the transformation process, ensuring smooth transitions at the edges of overlapping areas. The final re-acquired image of the eroded region is a mixed-resolution image, where the background region retains its original resolution, while the eroded defect region is replaced with parameter-optimized, more detailed high-resolution image data.

[0034] Please see Figure 5 The specific steps of S4 are as follows: S401: Based on the corresponding pixel positions in the re-acquired image of the eroded region and the surface region distribution image, extract the set of pixel coordinates of the boundary of the eroded region, and map them according to a unified coordinate system. Then, perform pixel pairing and comparison of the boundary contours of the same region in the two images to obtain the set of boundary position correspondences. The re-acquired image of the corrupted region is overlaid with the surface region distribution image, and the fine boundary of the corrupted region is extracted based on the corresponding pixel positions. Under a unified coordinate system, the set of edge pixel coordinates of the same defect region in both images is extracted. The boundary contours of the same region in the two images are pixel-paired and compared, and the Euclidean distance between corresponding boundary points is calculated. Specifically, for a point P1 on the boundary of the original image, the nearest point P2 on the boundary of the re-acquired image is found, and the coordinate offsets dx and dy between P1 and P2 are calculated. A boundary position correspondence set is obtained, which records the displacement vector field corrected from the original coarse boundary to the fine boundary. For example, if the original boundary point coordinates are (50, 50), and the corresponding high-contrast edge point coordinates in the re-acquired image are (51, 50.5), then the offset vector is (1, 0.5). By traversing the entire contour, a complete offset mapping relationship is established. This step eliminates the boundary blurring error caused by insufficient lighting during the initial acquisition.

[0035] S402: Call the coordinate offset values ​​in the boundary position correspondence set, calculate the spatial change trend of adjacent boundary points, identify continuous boundary segments and turning point regions based on the angle between adjacent offset vectors, mark the location and type of turning structure, and obtain the boundary change structure information set. The system retrieves coordinate offset values ​​from the boundary position correspondence set and arranges the boundary points in contour order. It calculates the spatial variation trend of adjacent boundary points, i.e., calculates the difference vector between two adjacent offset vectors. Based on the angle between adjacent offset vectors, it identifies continuous boundary segments and inflection point regions. A threshold of 45 degrees is set for inflection point determination. If the angle of change of the tangent direction between two adjacent points exceeds 45 degrees, it is marked as an inflection structure (such as a corner point); if the angle change is less than 10 degrees, it is marked as a continuous smooth segment. The location and type of the inflection structure are marked, resulting in a boundary variation structure information set. This logic aims to distinguish between genuine defect shape inflections and jagged noise caused by pixel discretization. For example, if vector V1 is at 0 degrees and vector V2 is at 10 degrees, with an angle of 10 degrees, it is determined to be smooth; if vector V3 is at 0 degrees and vector V4 is at 90 degrees, with an angle of 90 degrees, it is determined to be a right-angle inflection point.

[0036] S403: Based on the location and structure label in the boundary change structure information set, perform node position replacement and path connectivity adjustment on the boundary contour in the re-acquired image of the eroded region, reconstruct and close the boundary of the eroded region, and obtain the contour image of the eroded region. Based on the boundary change structure information set, the boundary contours in the re-acquired images of the eroded region are reconstructed. For regions marked as "continuous smooth segments," a cubic spline interpolation algorithm is used to smoothly connect adjacent nodes, eliminating the jagged effect. For regions marked as "turning structures," their sharp features are preserved, and no smoothing is performed. Path connectivity adjustments are made; if a small gap (less than 3 pixels) is found in the contour, a straight line segment is used to connect the tangent points at both ends of the gap, performing closure processing. Finally, a contour image of the eroded region is obtained. The defect contour in this image is a closed curve with well-defined geometric features, corrected by high-precision acquired data, accurately depicting the true physical boundary of the eroded region.

[0037] Please see Figure 6 The specific steps of S5 are as follows: S501: Based on each closed contour structure in the eroded region contour image, calculate the boundary closure index and pixel connectivity parameters, determine whether the spatial distance between the boundary start point and the end point is lower than the preset closure judgment threshold, and combine the pixel differences inside and outside the boundary to perform closure recognition and obtain the eroded boundary closure marker group. For each closed contour structure in the eroded region contour image, a boundary closure index is calculated. This index is defined as the Euclidean distance between the start and end points of the contour divided by the total perimeter of the contour. A closure threshold of 0.05 is set. The spatial distance between the start and end points of the boundary is checked against the preset closure threshold. If the calculated ratio is less than 0.05, it is considered geometrically closed. Simultaneously, closure recognition is performed by combining the pixel differences inside and outside the boundary. The average gray value of the region inside the contour is calculated, along with the average gray value of the 5-pixel-wide annular region outside the contour. If the difference is greater than 30 (gray level), the region is confirmed to have significant entity features. A eroded boundary closure marker group is obtained, where regions marked "True" represent confirmed closed and significant defect structures. Table 3 shows the calculation and judgment results of boundary closure and the gray-level difference between the inside and outside.

[0038] Table 3. Boundary Closure Determination Parameters; Note: In the table, "Start-End Distance / Perimeter" and "Closing Threshold" are dimensionless normalized ratios. "Internal / External Gray Level Difference" and "Difference Threshold" are in image gray levels, with a value range of 0–255.

[0039] As shown in Table 3, C01 and C03 meet the dual conditions and are confirmed as effective closed structures.

[0040] S502: Call the boundary closure region in the erosion boundary closure marker group, perform pixel texture fitting evaluation on the internal region, determine whether the texture block is independent of the erosion structure based on the texture block direction distribution parameters and area ratio, mark the location of separable texture blocks, and generate a erosion texture block separation identifier set. The process calls upon the closed regions identified in the erosion boundary closure marker group to perform pixel texture fitting evaluation on the internal regions. This process extracts texture features based on the Gray-Level Co-occurrence Matrix (GLCM) and evaluates the texture distribution characteristics by calculating the texture energy and contrast parameters within the regions. Based on the texture block orientation distribution parameters, i.e., the principal direction angle of the GLCM, and the area ratio (the proportion of the area with significant texture features to the total area), it is determined whether the texture blocks are independent of the erosion structure. If the texture orientation in a certain region is disordered (variance greater than 0.5) and the area ratio of high-frequency textures is less than 20%, it is identified as an attached surface stain or a separable texture block; if the texture orientation is consistent and runs through the entire region, it is identified as the erosion body texture. The locations of separable texture blocks are marked, generating a set of erosion texture block separation identifiers. For example, if the calculated texture orientation variance of a certain region is 0.8 and the area ratio is 15%, it is identified as a separable stain; another region has a variance of 0.1 and a ratio of 90%, and is identified as the erosion body.

[0041] S503: Based on the position index information in the corrosion texture block separation identifier set and the corrosion boundary closed marker group, perform corrosion structure aggregation processing and separable region removal operation, reconstruct the corrosion structure range layer, and establish the surface corrosion defect detection results. Based on the position index information in the corrupted texture block separation identifier set and the corrupted boundary closure marker group, corrupted structure aggregation processing is performed. All contours marked as "closed and salient" are retained, and pixels in areas marked as "separable texture blocks" are subtracted, performing a culling operation. This step effectively removes surface oil or water stains that are mistakenly identified as corrupted. The corrupted structure range layer is reconstructed, mapping the processed net defect area onto the final result layer, and assigning corresponding color labels based on the classification results in step S2. A surface corruption defect detection result is established, outputting a high-confidence defect distribution map after layers of screening, parameter optimization and re-sampling, geometric correction, and texture verification. Test results show that under the adopted multi-stage processing mechanism, the accuracy of the detection results shows a significant improvement compared to the traditional single-acquisition method, effectively reducing the false alarm rate.

[0042] Please see Figure 7 A systematic cable-stayed bridge defect detection and durability improvement system based on humidity control, including: The region segmentation module is used to implement S1: acquiring surface images of the cable-stayed bridge, constructing regular region segmentation relationships, analyzing grayscale change boundaries, brightness change directions and pixel connection directions within the regions, and generating surface region distribution images based on the relationship between pixel connection directions and brightness change directions. The type recognition module is used to implement S2: determine the consistency between the grayscale change boundary and the brightness change direction of the pixel connection interruption area in the surface region distribution image, distinguish between spot distribution pattern, block distribution pattern and strip distribution pattern, and generate a surface corrosion type identification image; The image reacquisition module is used to implement S3: based on the surface corrosion type identification image, adjust the brightness channel, gain channel and exposure channel of the image acquisition device, reacquire the corrosion area image marked in the surface corrosion type identification image, perform area alignment, and generate a reacquisition image of the corrosion area. The contour extraction module is used to implement S4: compare the re-acquired image of the corrupted region with the surface region distribution image, calculate the change in the boundary position of the corrupted region, adjust the contour of the corrupted region according to the continuity and transition distribution, and generate the contour image of the corrupted region. The defect detection module is used to implement S5: based on the contour image of the eroded area, determine the relationship between the boundary closure of the eroded area and the distribution of texture blocks, identify the overall eroded area, and generate surface erosion defect detection results.

[0043] 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 protection of the technical solution.

Claims

1. A systematic method for detecting defects and improving the durability of cable-stayed bridges based on humidity control, characterized in that, Includes the following steps: S1: Collect surface images of the cable-stayed bridge, construct regular region division relationships, analyze the gray-scale change boundaries, brightness change directions and pixel connection directions within the regions, and generate surface region distribution images based on the relationship between pixel connection directions and brightness change directions. S2: Determine the consistency between the grayscale change boundary and the brightness change direction of the pixel connection interruption area in the surface region distribution image, distinguish between spot distribution pattern, block distribution pattern and strip distribution pattern, and generate a surface corrosion type identification image; S3: Based on the surface corrosion type identification image, adjust the brightness channel, gain channel and exposure channel of the image acquisition device, re-acquire the corrosion area image marked in the surface corrosion type identification image, perform area alignment, generate the corrosion area and then acquire the image again. S4: Compare the image of the corrupted region with the image of the surface region distribution, calculate the change in the boundary position of the corrupted region, adjust the outline of the corrupted region according to the continuity and transition distribution, and generate a corrupted region outline image. S5: Based on the outline image of the eroded area, determine the relationship between the boundary closure of the eroded area and the distribution of texture blocks, identify the overall eroded area, and generate surface erosion defect detection results; During image acquisition, an environmental adaptive control step is performed: the relative humidity value of the environment is collected in real time and compared with a preset humidity safety threshold; when the real-time relative humidity value is detected to be greater than the preset humidity safety threshold, the central controller automatically triggers the dehumidification actuators configured around the image acquisition area to start, heating and purging the image acquisition window; humidity changes are continuously monitored until the relative humidity value drops below a stop threshold, the currently acquired image data is saved and subsequent image acquisition operations continue, wherein the value of the stop threshold is less than the value of the preset humidity safety threshold.

2. The systematic method for detecting and improving the durability of cable-stayed bridge defects based on humidity control according to claim 1, characterized in that, The surface region distribution image includes grayscale boundary distribution information, brightness direction information, and pixel connection structure. The surface corrosion type identification image includes spot type identification, block type identification, and strip type identification. The corrosion region re-acquired image includes brightness adjustment image, contrast adjustment image, and exposure compensation image. The corrosion region contour image includes boundary continuity features, boundary transition features, and position change features. The surface corrosion defect detection results include closed boundary defects, texture block distribution defects, and overall corrosion region defects.

3. The systematic method for detecting defects and improving durability of cable-stayed bridges based on humidity control according to claim 1, characterized in that, The adjustment of the erosion region contour based on continuity and transition distribution refers to dynamically optimizing the boundary path and correcting the initially extracted erosion contour lines based on the gray-level continuity and local transition direction distribution of the erosion region boundary in the pixel image.

4. The systematic method for detecting defects and improving durability of cable-stayed bridges based on humidity control according to claim 1, characterized in that, The execution region alignment refers to aligning the newly acquired corrupted region image with the initial region distribution image in spatial coordinates using feature point matching and image registration algorithms, thereby verifying the consistency and comparability of the corruption information.

5. The systematic method for detecting defects and improving durability of cable-stayed bridges based on humidity control according to claim 1, characterized in that, The specific steps of S1 are as follows: S101: Based on the acquired images of the cable-stayed bridge surface, construct an image pixel array and divide it into regular grid regions, extract the gray value distribution within the regions, calculate the gray value difference between adjacent pixels and compare it with a preset gray value change threshold, filter gray value change regions, and generate a gray value change boundary matrix. S102: Based on the pixel positions in the grayscale change boundary matrix, extract the grayscale change direction and pixel connection direction of adjacent regions, calculate the angle between them and compare it with a preset continuous change direction threshold, filter pixel paths with the same direction, and obtain a set of brightness change path sequences. S103: Call the pixel paths in the brightness change path sequence set, extract the corresponding region blocks and aggregate and connect them, reconstruct the regional spatial relationship in the image, and obtain the surface region distribution image.

6. The systematic method for detecting defects and improving durability of cable-stayed bridges based on humidity control according to claim 1, characterized in that, The specific steps of S2 are as follows: S201: Based on the pixel connection interruption region in the surface region distribution image, extract the corresponding gray-scale change boundary direction and brightness change direction vector, calculate the angle between them and compare it with a preset direction consistency threshold, filter the pixel index set whose angle meets the threshold condition, and generate a boundary direction consistency marker set. S202: Call the pixel index in the boundary direction consistency mark set, perform structural scale calculation and pixel arrangement statistics on the region corresponding to the index, and judge based on the region size parameter and connectivity extension parameter to divide the region structure into three categories: spots, blocks and stripes, and obtain the erosion morphology classification label group. S203: Based on the pixel regions corresponding to the labels in the corrosion morphology classification label group, perform region mapping and label encoding configuration in the surface region distribution image, integrate image pixel annotations, and establish a surface corrosion type label image.

7. The systematic method for detecting defects and improving durability of cable-stayed bridges based on humidity control according to claim 1, characterized in that, The specific steps for S3 are as follows: S301: Based on the pixel position of the corroded area in the surface corrosion type identifier image, retrieve the image acquisition configuration parameters, adjust the current values ​​of the brightness channel, gain channel and exposure channel, and set the corresponding parameter set according to the area coordinates to generate the image acquisition parameter configuration group. S302: Call the configuration parameters in the image acquisition parameter configuration group to re-acquire the image of the marked eroded area, extract the pixel data frame of the corresponding coordinate area in the newly acquired image, establish the image frame set under the current channel parameters, and obtain the area re-acquired image frame group. S303: Based on the spatial coordinates of the image frames in the reacquired image frame group and the original image position, perform regional boundary mapping and pixel matrix alignment operations, perform spatial correction processing between the acquired image and the original region, and obtain the reacquired image of the eroded region.

8. The systematic method for detecting defects and improving durability of cable-stayed bridges based on humidity control according to claim 1, characterized in that, The specific steps of S4 are as follows: S401: Based on the corresponding pixel positions in the re-acquired image of the eroded region and the distribution image of the surface region, extract the set of pixel coordinates of the boundary of the eroded region, and map them according to a unified coordinate system. Then, perform pixel pairing and comparison of the boundary contours of the same region in the two images to obtain the set of boundary position correspondences. S402: Call the coordinate offset values ​​in the boundary position correspondence set, calculate the spatial change trend of adjacent boundary points, identify continuous boundary segments and turning point regions based on the angle between adjacent offset vectors, mark the location and type of the turning structure, and obtain the boundary change structure information set. S403: Based on the location and structure label in the boundary change structure information set, perform node position replacement and path connectivity adjustment on the boundary contour in the re-acquired image of the eroded region, perform boundary reconstruction and closure processing of the eroded region, and obtain the contour image of the eroded region.

9. The systematic method for detecting defects and improving durability of cable-stayed bridges based on humidity control according to claim 1, characterized in that, The specific steps of S5 are as follows: S501: Based on each closed contour structure in the eroded region contour image, calculate the boundary closure index and pixel connectivity parameter, determine whether the spatial distance between the boundary start point and the end point is lower than the preset closure judgment threshold, and perform closure recognition in combination with the pixel difference inside and outside the boundary to obtain the eroded boundary closure mark group. S502: Call the boundary closure region in the erosion boundary closure marker group, perform pixel texture fitting degree evaluation on the internal region, determine whether the texture block is independent of the erosion structure based on the texture block direction distribution parameters and area ratio, mark the position of the separable texture block, and generate a erosion texture block separation identifier set. S503: Based on the position index information in the corrosion texture block separation identifier set and the corrosion boundary closure marker group, perform corrosion structure aggregation processing and separable region removal operation, reconstruct the corrosion structure range layer, and establish surface corrosion defect detection results.

10. A systematic cable-stayed bridge defect detection and durability improvement system based on humidity control, characterized in that, The system is used to implement the systematic cable-stayed bridge defect detection and durability improvement method based on humidity control as described in any one of claims 1-9, and the system includes: The region segmentation module is used to implement S1: acquiring surface images of the cable-stayed bridge, constructing regular region segmentation relationships, analyzing grayscale change boundaries, brightness change directions and pixel connection directions within the regions, and generating surface region distribution images based on the relationship between pixel connection directions and brightness change directions. The type recognition module is used to implement S2: determine the consistency between the grayscale change boundary and the brightness change direction of the pixel connection interruption area in the surface region distribution image, distinguish between spot distribution pattern, block distribution pattern and strip distribution pattern, and generate a surface corrosion type identification image; The image re-acquisition module is used to implement S3: based on the surface corrosion type identification image, adjust the brightness channel, gain channel and exposure channel of the image acquisition device, re-acquire the corrosion area image marked in the surface corrosion type identification image, perform area alignment, and generate a corrosion area re-acquisition image; The contour extraction module is used to implement S4: compare the re-acquired image of the eroded region with the distribution image of the surface region, calculate the change in the boundary position of the eroded region, adjust the contour of the eroded region according to the continuity and transition distribution, and generate a contour image of the eroded region. The defect detection module is used to implement S5: based on the outline image of the eroded area, determine the relationship between the boundary closure of the eroded area and the distribution of texture blocks, identify the overall eroded area, and generate surface erosion defect detection results.

Images