Hydraulic engineering crack detection method and system based on intelligent visual identification

Abstract

The invention relates to the technical field of water conservancy projects, in particular to a water conservancy project crack detection method and system based on intelligent visual identification, and the method comprises the following steps: obtaining a crack image, enhancing the brightness and contrast in different regions, extracting and optimizing the crack edge, executing morphological processing, denoising and smoothing, segmenting a crack region, and judging the connection condition. And analyzing the geometrical characteristic detection connectivity, and calibrating the crack position to obtain a positioning result. According to the method, through dynamic adjustment of image brightness and contrast, the crack is more prominent in a complex background, especially under non-uniform illumination, crack identification difficulty caused by illumination inconsistency is reduced, crack edges are optimized through morphological operation, the missing part of the crack can be filled, the crack contour can be smoothed, and noise can be eliminated. And in the crack segmentation stage, through self-adaptive threshold and geometric feature analysis, the fracture and connection conditions of the crack are accurately judged, and positioning of the crack end point and the continuous area of the crack end point is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of water conservancy projects, and particularly to a method and system for detecting cracks in water conservancy projects based on intelligent vision recognition. Background Art

[0002] Water conservancy engineering technology involves all engineering projects related to the development, utilization, and management of water resources, including the design, construction, operation, and maintenance of facilities such as reservoirs, rivers, lakes, irrigation systems, drainage systems, and hydropower stations. In this field, the technologies involved cover multiple aspects such as water flow control, water quality monitoring, facility structure analysis, and irrigation management. With the continuous development of modern technologies, intelligent and automated technical means have gradually been introduced into water conservancy projects, such as remote sensing technology, sensor technology, artificial intelligence, the Internet of Things, and big data analysis.

[0003] Among them, the method for detecting cracks in water conservancy projects based on intelligent vision recognition mainly applies intelligent vision recognition technology to detect crack problems in water conservancy projects. By automatically detecting and analyzing cracks in water conservancy facilities, the accuracy and efficiency of crack detection can be greatly improved. Especially in complex environments where manual detection is difficult or problems cannot be discovered in a timely manner, through this technology, the health status of water conservancy facilities can be monitored in real time, which helps to discover and handle potential structural problems in a timely manner and ensure the safe operation of water conservancy projects.

[0004] The prior art relies on manual inspection or limited sensor technology for crack detection and cannot cope with real-time monitoring in complex environments. Under uneven illumination or special environmental conditions, manual detection is easily interfered by external factors and cannot achieve efficient and large-scale crack identification and monitoring. The prior art lacks adaptability and cannot automatically adjust the influence of changes such as illumination on the detection results. Moreover, in the case of small or hidden cracks, key crack information is easily missed. In addition, in high-humidity environments or places with accumulated water, the detection difficulty increases due to reflection or light spots, making it difficult to achieve fast and accurate crack detection, affecting the detection effect and the accuracy of data, resulting in potential structural problems not being discovered in a timely manner, and increasing the complexity and risk of later maintenance. Summary of the Invention

[0005] The purpose of the present invention is to solve the deficiencies existing in the prior art and propose a method and system for detecting cracks in water conservancy projects based on intelligent vision recognition.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions: A method for detecting cracks in water conservancy projects based on intelligent vision recognition, including the following steps,

[0007] S1: Obtain the crack image of the water conservancy project, divide the image into multiple regions, detect the brightness change of each region, calculate its local contrast, perform brightness adjustment, refine the contrast between the crack and the background by adjusting the regional contrast weight, and obtain the locally enhanced image;

[0008] S2: Based on the locally enhanced image, perform morphological closing operation to fill the missing parts at the crack edge, eliminate the edge distortion caused by noise or uneven illumination, and then remove the background noise through morphological opening operation to retain the effective crack region, and calculate its length, width and curvature to obtain the crack edge optimization information;

[0009] S3: Based on the crack edge optimization information, compare the image pixel values with the background region pixel values to segment the crack region, and judge whether there are fracture or connection problems by analyzing the geometric features of the crack region to obtain the crack region segmentation result;

[0010] S4: Based on the crack region segmentation result, perform connectivity analysis on the crack region, judge whether the crack is a continuous crack or an intermittent crack, and analyze the relative positions of the fracture endpoints of the crack to generate crack connectivity detection data.

[0011] The improvement of the present invention is that the locally enhanced image includes the region image after brightness adjustment, the region image after contrast optimization, and the crack edge refinement image. The crack edge optimization information includes crack edge filling data and crack smoothness data. The crack region segmentation result includes segmentation threshold, crack region boundary data, and crack endpoint positioning result. The crack connectivity detection data includes crack continuity state and crack fracture endpoint data.

[0012] The improvement of the present invention is that the specific steps for obtaining the locally enhanced image are as follows:

[0013] S111: Obtain the crack image of the water conservancy project, divide the image into multiple regions according to the illumination information, monitor the brightness change of each region, calculate its local contrast, and generate the local contrast value;

[0014] S112: According to the local contrast value, perform brightness adjustment according to the contrast difference of each region, and adjust the contrast weight of each region by using the formula:

[0015]

[0016] Obtain the adjusted local contrast CR adjusted , where CR i represents the local contrast value of region i, WR i represents the contrast weight of region i, and N crrepresents the total number of representative regions, ΔLR represents the local luminance difference, and CR max represents the maximum value of local contrast;

[0017] S113: Based on the adjusted local contrast, optimize the crack edge region, refine the contrast between the crack and the background, and obtain the locally enhanced image.

[0018] The improvement of the present invention is that the step of obtaining the crack edge optimization information is specifically as follows:

[0019] S211: Based on the locally enhanced image, extract the gray gradient and structural direction change values of the crack contour region, perform a closing operation to complement the discontinuous edges, and obtain the edge distortion positioning interval;

[0020] S212: Invoke the edge distortion positioning interval, analyze the roughness and contour fitting residuals of the crack boundary pixels, judge the smoothness and consistency, and use the formula:

[0021]

[0022] Calculate the crack contour continuity cooperation value CE r , reconstruct the edge pixel structure, and obtain the smooth crack contour image, where represents the boundary roughness of the k-th pixel, is the mean value of the boundary roughness, is the gray level jump intensity of the k-th pixel, G (avg) is the average value of the gray level jump intensity, n ce represents the total number of pixels;

[0023] S213: According to the smooth crack contour image, remove the background noise through an opening operation, retain the effective crack region, analyze the connectivity and closure of the crack, identify the key geometric features of the crack, and obtain the crack edge optimization information.

[0024] The improvement of the present invention is that the step of obtaining the crack region segmentation result is specifically as follows:

[0025] S311: Based on the crack edge optimization information, compare the image pixel values with the background region pixel values, analyze the gray level difference of the data, and determine the key difference baseline by sorting the gray level differences, so as to obtain the crack pixel difference interval;

[0026] S312: Based on the crack pixel difference interval, analyze the pixel gradient and the background intensity difference, and use the formula:

[0027]

[0028] Obtain the optimal threshold range TB for crack segmentation opt, screen the pixel regions that meet the conditions to obtain the optimized parameters of the crack gray boundary. Among them, GB k represents the local gradient change intensity of the k-th pixel, BE k represents the average background pixel intensity within the k-th pixel, DB k represents the gray difference between the k-th pixel block and its adjacent blocks, n ce represents the total number of pixels;

[0029] S313: Based on the optimized parameters of the crack gray boundary, evaluate the continuity of the edge points in the region, determine whether there are problems of fracture or connection, locate the endpoints and continuous regions of the crack, and obtain the crack region segmentation result.

[0030] The improvement of the present invention is that the step of obtaining the crack connectivity detection data is specifically as follows:

[0031] S411: Based on the crack region segmentation result, analyze the difference between the Euclidean distance between the starting and ending points of the main direction coordinate axis and the crack pixel boundary, identify the spatial span and geometric extension direction of the crack line segment, and obtain the crack extension geometric quantity;

[0032] S412: Invoke the crack extension geometric quantity, combine the width and curvature changes of the boundary points, and use the formula:

[0033]

[0034] Calculate the crack non-linear change degree RU, determine whether the crack is a continuous crack or an intermittent crack, and obtain the crack geometric change trend. Among them, wu z is the width value of the z-th measurement point, wu is the average width, ku z is the curvature value of the z-th measurement point, ku is the average curvature, θu z is the direction change angle of the z-th measurement point, θu max is the maximum angle change value, N ru is the number of key measurement points included in the crack;

[0035] S413: According to the crack geometric change trend, analyze the distance and angle relationship between the crack endpoints, determine whether the adjacent crack segments are in a continuous connection state, and generate the crack connectivity detection data.

[0036] The improvement of the present invention is that the step further includes:

[0037] S5: Based on the crack connectivity detection data, use geometric constraints to locate the misjudged fractured cracks, calibrate the crack position according to the crack geometric characteristics, determine the current position coordinates of the crack, and verify the geometric shape of the crack to obtain the crack location result;

[0038] The crack location result includes crack position coordinates, crack geometric shape calibration data, and crack correction information.

[0039] The improvement of the present invention is that the steps for obtaining the crack location result are specifically as follows:

[0040] S511: Based on the crack connectivity detection data, quantitatively measure the spatial continuity of the fractured cracks, compare with the regional average crack length and the change in trend, determine whether there is a structural fracture in the cracks, and obtain a set of identified fractured cracks;

[0041] S512: According to the set of identified fractured cracks, analyze the central point position and the starting and ending point coordinates of the cracks, calculate the deviation of the average distance between it and adjacent cracks, evaluate the deviation angle of the trend, and obtain the crack coordinate offset interval;

[0042] S513: According to the crack coordinate offset interval, adjust the position of the central point coordinates of the cracks, re-identify their geometric shapes, determine the current position coordinates of the cracks, and obtain the crack location result.

[0043] A water conservancy project crack detection system based on intelligent vision recognition, the system includes:

[0044] The image processing and enhancement module acquires the water conservancy project crack image, divides the image into multiple regions according to the illumination information, detects the brightness change of each region, calculates its local contrast, performs brightness adjustment, and refines the contrast between the cracks and the background by adjusting the regional contrast weight to obtain a locally enhanced image;

[0045] The crack edge optimization module, based on the locally enhanced image, extracts the crack contour, performs a morphological closing operation to fill in the missing parts of the crack edge, eliminates the edge distortion caused by noise or uneven illumination, and then removes the background noise through a morphological opening operation, and analyzes the crack morphological characteristics according to the geometric characteristics of the cracks to obtain crack edge optimization information;

[0046] The crack region segmentation module, based on the crack edge optimization information, compares the image pixel values with the background region pixel values to segment the crack region, analyzes the geometric characteristics of the crack region to determine whether there are fracture or connection problems, and locates the endpoints and continuous regions of the cracks to obtain the crack region segmentation result;

[0047] The crack connectivity analysis module, based on the crack region segmentation result, analyzes the connectivity of the crack region according to the geometric characteristics of the cracks, calculates the fracture endpoints, length, width, and curvature of the cracks, determines whether the cracks are continuous cracks or intermittent cracks, and analyzes the relative positions of the fracture endpoints of the cracks to generate crack connectivity detection data;

[0048] Based on the crack connectivity detection data, the crack positioning and calibration module locates the misjudged fracture cracks using geometric constraints, calibrates the crack positions according to the crack geometric features, determines the current position coordinates of the cracks, and obtains the crack positioning results.

[0049] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0050] In the present invention, through the dynamic adjustment of image brightness and contrast, the cracks are made more prominent in the complex background. Especially under uneven illumination, the difficulty of crack recognition caused by inconsistent illumination is reduced. By optimizing the crack edges through morphological operations, the missing parts of the cracks can be filled, the crack contours can be smoothed, and the noise can be eliminated, thereby ensuring the clarity and integrity of the crack features. In the crack segmentation stage, through adaptive threshold and geometric feature analysis, the fracture and connection conditions of the cracks are accurately judged, the accurate positioning of the crack endpoints and their continuous regions is realized, the probability of false detection and missed detection is reduced. Combining crack connectivity analysis and geometric constraint positioning further improves the accuracy and reliability of the crack positions, maintains high monitoring capabilities under different environmental conditions, can accurately calibrate the spatial positions of the cracks, and provides real-time and accurate detection support for the structural monitoring of water conservancy projects. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 is the main step flow chart of the present invention;

[0052] Figure 2 is the flow chart for obtaining the locally enhanced image in the present invention;

[0053] Figure 3 is the flow chart for obtaining the crack edge optimization information in the present invention;

[0054] Figure 4 is the flow chart for obtaining the crack region segmentation results in the present invention;

[0055] Figure 5 is the flow chart for obtaining the crack connectivity detection data in the present invention;

[0056] Figure 6 is the flow chart for obtaining the crack positioning results in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0057] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0058] In the description of the present invention, it should be understood that the orientation or positional relationship indicated by terms such as "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as a limitation to the present invention. In addition, in the description of the present invention, the meaning of "a plurality of" is two or more, unless otherwise specifically defined.

[0059] Embodiment

[0060] Please refer to Figure 1 , the present invention provides a technical solution: a method for detecting cracks in hydraulic engineering based on intelligent vision recognition, including the following steps:

[0061] S1: Obtain the crack image of the hydraulic engineering, divide the image into multiple regions according to the illumination information, detect the brightness change of each region, calculate its local contrast, and perform brightness adjustment for the contrast difference of each region. By adjusting the regional contrast weight, optimize the crack edge region, refine the contrast between the crack and the background, and obtain the locally enhanced image;

[0062] S2: Based on the locally enhanced image, extract the crack contour, perform morphological closing operation to fill the missing parts at the crack edge, smooth the crack contour, eliminate the edge distortion caused by noise or uneven illumination, and then remove the background noise through morphological opening operation, retain the effective crack region, and calculate its length, width and curvature according to the geometric characteristics of the crack, analyze the crack morphological characteristics, and obtain the crack edge optimization information;

[0063] S3: Based on the crack edge optimization information, compare the image pixel value with the pixel value of the background region to determine the optimal threshold range, segment the crack region, analyze the geometric characteristics of the crack region to judge whether there are fracture or connection problems, locate the end points and continuous regions of the crack, and obtain the crack region segmentation result;

[0064] S4: Based on the crack region segmentation result, perform connectivity analysis on the crack region according to the geometric characteristics of the crack, calculate the fracture end points, length, width and curvature of the crack, judge whether the crack is a continuous crack or an intermittent crack, and analyze the relative position of the fracture end points of the crack to generate crack connectivity detection data;

[0065] S5: Based on the crack connectivity detection data, use geometric constraints to locate the misjudged fracture cracks, calibrate the crack position according to the geometric characteristics of the crack, determine the current position coordinates of the crack, and verify the geometric shape of the crack to obtain the crack location result.

[0066] The locally enhanced image includes the region image after brightness adjustment, the region image after contrast optimization, and the crack edge refined image. The crack edge optimization information includes crack edge filling data and crack smoothness data. The crack region segmentation result includes the segmentation threshold, the crack region boundary data, and the crack endpoint positioning result. The crack connectivity detection data includes the crack continuity state and the crack fracture endpoint data. The crack positioning result includes the crack position coordinates, the crack geometric shape calibration data, and the crack correction information.

[0067] Please refer to Figure 2 , the steps for obtaining the locally enhanced image are specifically as follows:

[0068] S111: Obtain the crack image of the water conservancy project, divide the image into multiple regions according to the illumination information, monitor the brightness change of each region, calculate its local contrast, and generate the local contrast value;

[0069] Obtain the crack image of the water conservancy project, and divide the image into multiple regions based on the illumination information. Each region represents a different part of the crack. The key to this operation is to accurately obtain the brightness information within each region for further analysis of the crack characteristics of the image. Then, by monitoring the brightness change of each divided region, the illumination fluctuation situation within the region can be analyzed. For example, in a crack image, the brightness value of a certain region changes from 100 to 150 with the change of illumination. Such a change will reflect the light and dark contrast of the crack in this region. To obtain the contrast value, it is necessary to calculate the local contrast of each region, that is, to measure the contrast of the region through the brightness value difference within each region of the image. Taking a specific example, assume that the brightness range of a certain region is [80, 130], then the local contrast of this region is calculated as the difference (130 - 80 = 50). The local contrast of this region can be further quantified through a standardized formula. After this process, the local contrast values of each region in the image can be obtained, providing basic data support for subsequent brightness adjustment and image optimization.

[0070] S112: According to the local contrast value, perform brightness adjustment based on the contrast difference of each region. By adjusting the contrast weight of each region, using the formula:

[0071]

[0072] Obtain the adjusted local contrast CR adjusted , where CR i represents the local contrast value of region i, WR i represents the contrast weight of region i, N cr represents the total number of regions, ΔLR represents the local brightness difference, and CR max represents the maximum value of the local contrast;

[0073] Adjust the brightness according to the contrast difference of each region. For this purpose, it is necessary to calculate the contrast weight of each region, which depends on the brightness difference of each region. For example, assume that the contrast of region 1 is 60 and the contrast of region 2 is 40. The brightness change in region 1 is larger, so the contrast weight will be higher. The weight can be calculated by the ratio between the contrast of the region and the maximum contrast value. Assume that the maximum contrast value is 100. The contrast weight of region 1 can be calculated as 60 / 100 = 0.6, and the contrast weight of region 2 can be calculated as 40 / 100 = 0.4. If the brightness difference ΔLR of the region is 10 and the maximum value CR of the local contrast max is 100, substitute it into the following formula for calculation:

[0074]

[0075] Through this process, CR adjusted = 27 is obtained. This value reflects how the contrast of each region is optimized by adjusting the contrast weight and brightness difference, thus laying a foundation for the subsequent optimization of the crack image.

[0076] S113: Based on the adjusted local contrast, optimize the crack edge region, refine the contrast between the crack and the background, and obtain a locally enhanced image;

[0077] By refining the contrast between the crack and the background, the crack becomes more obvious in the image. According to the obtained adjusted contrast value, further pixel enhancement processing can be performed between regions. For example, increase the contrast of the crack edge to increase the distinguishability between it and the surrounding background region. The goal of this operation is to optimize the crack edge, make the crack features more prominent, and finally obtain a locally enhanced image. For example, in the region where the adjusted contrast value is 70, the edge of the crack changes from the original blurred transition region to a clearly distinguishable boundary. The optimized crack image will provide more accurate visual information for subsequent analysis.

[0078] Please refer to Figure 3 , and the specific steps for obtaining the crack edge optimization information are as follows:

[0079] S211: Based on the locally enhanced image, extract the gray-scale gradient and structural direction change values of the crack contour region, perform a closing operation to fill in the discontinuous edges, and obtain the edge distortion positioning interval;

[0080] Extract the enhanced pixel array from the image input module, read the grayscale value of each pixel, and calculate its grayscale gradient sequentially based on a 3×3 neighborhood window. Specifically, with the target pixel as the center, calculate the absolute difference with its adjacent pixels above, below, left, and right. If the grayscale value of a certain central pixel is 128, and the grayscale values of the adjacent pixels above, below, left, and right are 130, 125, 127, and 129 respectively, then the grayscale gradient values are 2, 3, 1, and 1 respectively. The maximum value of the grayscale gradient is taken as 3 to depict the boundary response intensity at this position. Subsequently, based on the edge structure direction of the image, extract the gradient direction angle difference between the pixels in this neighborhood. If the direction angle between adjacent pixels exceeds 45°, it is marked as a structural mutation point, and a preliminary edge structure diagram is established. Then, perform a morphological closing operation on the extracted contour. Use a circular template with a structural element size of 5×5. First, perform a dilation operation to expand the edge structure, and then perform an erosion operation to merge the broken parts. In an example, there is a breakpoint with a length of 3 pixels on the edge of a crack. After the closing operation, the pixel connectivity is improved, and the number of connected domains on one side is reduced from 3 to 1 closed contour area. After the closing operation, call the connected component detection method to calculate the continuity of the boundary section and the change in grayscale structure. By comparing the number of connected components before and after the closing operation, select the regions with a change value greater than 2 as candidate regions. At the same time, calculate the average value of the structural direction angles. If the direction difference in a certain region fluctuates by more than 60° within 3 or more pixels, such as 10°, 75°, 25°, 90°, then it is determined that this section of the boundary is distorted. After comprehensively judging the results, extract all pixel segments that satisfy the connected component change amplitude greater than 2 and the structural direction jump amplitude greater than 60°, and define them as the edge distortion localization interval.

[0081] S212: Call the edge distortion localization interval, analyze the roughness and contour fitting residuals of the crack boundary pixels, judge the smoothness and consistency, and use the formula:

[0082]

[0083] Calculate the crack contour continuity cooperation value CE r , reconstruct the edge pixel structure to obtain a smooth image of the crack contour, where, represents the boundary roughness of the k-th pixel, representing the roughness of this pixel point at the crack contour boundary, is the mean value of the boundary roughness, used to calculate the average roughness of the crack boundary to standardize the deviation of a single pixel, is the grayscale jump intensity of the k-th pixel, indicating the degree of grayscale change of this pixel point in the image. A high grayscale jump intensity indicates the presence of an edge or a crack, G (avg) is the average value of the grayscale jump intensity, used to calculate the average level of the grayscale jump intensity of all pixel points, so that the jump of a single pixel can be compared with the average state, n ce represents the total number of pixels;

[0084] Call the edge distortion positioning interval, extract the crack edge pixel points in this interval one by one, and perform the analysis operations of boundary roughness and gray level jump intensity. First, with each edge pixel as the center, construct a 3×3 neighborhood window, and count the square of the difference between the gray level value of the center pixel and the average gray level of the neighborhood pixels to obtain the boundary roughness value of this pixel point For example, the gray level value of the first pixel is 148, and the neighborhood average is 145, then the roughness is (148 - 145) 2 = 9, and the normalized value is 0.36; the gray level value of the second pixel is 150, and the neighborhood average is 147, the roughness is (150 - 147) 2 = 9, and after normalization it is 0.34; the gray level value of the third pixel is 143, and the neighborhood average is 141, the roughness is (143 - 141) 2 = 4, and the normalized value is 0.28, then the roughness value set of the 3 pixels in this area is: Calculate the average roughness as:

[0085]

[0086] Next, obtain the gray level jump intensity of each pixel point Defined as the absolute difference between the current pixel gray level value and the average gray level of its neighborhood. For the first pixel, |148 - 145| = 3, for the second pixel it is |150 - 147| = 3, and for the third pixel it is |143 - 141| = 2, then the gray level jump intensity set is: The average gray level jump intensity is:

[0087]

[0088] Substitute the above data into the formula to calculate the crack profile continuity coordination value. Calculate the numerator part:

[0089] (0.36 - 0.3267) 2 = 0.0011;

[0090] (0.34 - 0.3267) 2 = 0.0002;

[0091] (0.28 - 0.3267) 2 = 0.0022;

[0092] The sum of the numerators = 0.0011 + 0.0002 + 0.0022 = 0.0035;

[0093] Calculate the denominator part:

[0094] |3 - 2.67| = 0.33;

[0095] |3 - 2.67| = 0.33;

[0096] |2 - 2.67| = 0.67;

[0097] The sum of the denominators = 0.33 + 0.33 + 0.67 = 1.33;

[0098] Final CE r is:

[0099]

[0100] Thus, the collaborative value of the crack profile continuity is approximately 0.00263. Comparing it with the set reference threshold of 0.005, it can be seen that both the fluctuation range of the current crack edge roughness and the deviation of the jump intensity are at a low level. This value is lower than the reference value, indicating that the structural continuity of the crack edge is strong. Pixel reconstruction of the profile can be performed to reconstruct the crack edge structure in this area and output a smooth image of the crack profile.

[0101] S213: According to the smooth image of the crack profile, remove background noise through opening operation, retain the effective crack area, analyze the connectivity and closure of the crack, identify the key geometric features of the crack, and obtain the optimized information of the crack edge;

[0102] Use a structural element with a radius of 3 to perform erosion and dilation operations on the image, remove isolated small areas in the background, and only retain the main crack structure with a connected area length greater than 40 pixels after detection. Subsequently, establish an equally spaced pixel sequence along the main axis in the crack area, and perform geometric feature analysis on each segment. Extract the average width of every 10 pixels. If there is a fluctuation of more than 5 pixels in width in this segment, it is defined as a geometric change segment. Then calculate the curvature value through the three-point difference method. If the continuous curvature value in the segment exceeds 1.5 per 10 pixels, it means that there is a rapid direction change in this segment. The formula used in the curvature calculation process is where S is the area of the triangle and L is the average side length. For example, if the three-point coordinates are (20, 20), (22, 23), and (25, 28) respectively, calculate the area S as 7.5 and L as 6.3, then κ ≈ 0.3. Perform frequency analysis on the curvature values of all segments, identify the part with a frequency greater than 1 time per 10 pixels as the mutation feature segment, and then combine the structural closure index. Set the judgment criterion as the closure degree greater than 0.95. For example, if the distance between the head and tail of the crack is less than 1 pixel, it is considered to be well closed. Finally, extract the key geometric features of all crack areas that meet the conditions of a crack connectivity greater than 50 pixels, an average width between 5 and 12 pixels, and a curvature frequency between 0.8 and 2.0, and output the optimized information of the crack edge.

[0103] Please refer to Figure 4 , the specific steps for obtaining the crack area segmentation result are as follows:

[0104] S311: Based on the optimized information of the crack edge, compare the pixel values of the image with those of the background area, analyze the gray - scale difference of the data, determine the key difference baseline by sorting the gray - scale differences, and obtain the crack pixel difference interval;

[0105] Extract the pixel points near the crack edge and the pixel points of the adjacent background area from the image. For the image gray - scale matrix, select two pixel points P1 and P2 located in the crack edge area respectively, and their extracted gray - scale values are 135 and 140 respectively. At the same time, select two reference pixel points Q1 and Q2 in the image background area, and their gray - scale values are 120 and 123 respectively. Calculate the gray - scale difference between the crack - edge pixels and the background pixels. The difference between P1 and Q1 is 135 - 120 = 15, and the difference between P2 and Q2 is 140 - 123 = 17. The obtained differences form a difference data set [15, 17]. After sorting this data set, an ascending sequence [15, 17] is obtained. Further calculate its first - order derivative, that is, the difference between the two differences 17 - 15 = 2. Identify the distribution mutation point through this difference change amount. Since the amount of data is small, directly use the mean value 16 of the two values as the difference baseline, extend 2 units upward and downward with this value as the center to construct a difference range [14, 18], and combine with the image gray - scale distribution frequency diagram to eliminate the values with a difference frequency less than 5% (assuming that the differences 13 and 19 have low frequencies and are eliminated). Finally, establish the crack pixel difference interval as [14, 18].

[0106] S312: Based on the crack pixel difference interval, analyze the pixel gradient and the background intensity difference, and use the formula:

[0107]

[0108] Obtain the optimal threshold range TB for crack segmentation opt , screen the pixel areas that meet the conditions, and obtain the optimized parameters of the crack gray - scale boundary, where GB k represents the local gradient change intensity of the k - th pixel, BE k represents the average background pixel intensity within the k - th pixel, which is used to compare the difference with the crack area, DB k represents the gray - scale difference between the k - th pixel block and the adjacent block, which is used to identify the crack edge, n ce represents the total number of pixels;

[0109] Based on the crack pixel difference interval [14, 18], extract the local gradient values of two pixels P1 and P2 in the image. Assume that the maximum gray value in the neighborhood of P1 is 150 and the minimum is 120, then the local gradient change intensity GB = 150 - 120 = 30. For P2, the maximum in the neighborhood is 160 and the minimum is 125, then GB = 35. Use the gray mean value of the 5×5 pixel block around P1 and P2 as the background average pixel intensity. Assume that the BE of the neighborhood of P1 is 125 and the BE of the neighborhood of P2 is 130. At the same time, measure the difference between the gray mean values of the regions of P1 and P2 and their circumscribed 8-neighborhoods respectively, set as DB = 6, DB = 8, and substitute into the formula:

[0110] Item 1:

[0111] Item 2:

[0112] Calculate the mean value:

[0113]

[0114] Obtain the optimal threshold range TB for crack segmentation opt = -37.34. Since it is negative, it indicates that the gray value change of crack pixels is higher than the background intensity and the regional gray value is unbalanced. Subsequently, use its absolute value or offset remapping method to perform gray boundary interval conversion, so as to obtain the optimized parameters of the crack gray boundary.

[0115] S313: Based on the optimized parameters of the crack gray boundary, evaluate the continuity of the edge points in the region, judge whether there are fracture or connection problems, locate the end points and continuous regions of the crack, and obtain the crack region segmentation result;

[0116] Extract the regions where P1 and P2 are located for edge point location analysis, evaluate the connectivity of the eight-neighborhood pixel groups of each pixel, calculate the Euclidean distance between P1 and P2. Assume the coordinates are (100, 105) and (106, 108) respectively, then the distance is √((106 - 100) 2 +(108 - 105) 2 ) = √(36 + 9) = √45 ≈ 6.7 pixels. Set the critical length for fracture judgment to 10 pixels. Since 6.7 < 10, it is judged as a continuous region. Further, count the total number of pixels in the regions where P1 and P2 are located as 42, and the region area is 5×10 = 50 pixels. Then the pixel density = 42 / 50 = 0.84. Set the density judgment reference value to 0.6. Since it is greater than the threshold, it meets the crack connectivity requirements. Finally, complete the crack region structure recognition and obtain the crack region segmentation result.

[0117] Please refer to Figure 5 , the specific steps for obtaining the crack connectivity detection data are as follows:

[0118] S411: Based on the crack region segmentation results, analyze the difference between the Euclidean distance between the starting and ending points of the main direction coordinate axis and the crack pixel boundary, identify the spatial span and geometric extension direction of the crack segment, and obtain the crack extension geometric quantity;

[0119] Perform contour pixel tracking on the crack regions in the image, extract the starting and ending coordinate values of each crack region on the main direction coordinate axis. For example, in a crack, its starting coordinate is (120, 340), and the ending point is (300, 340). The pixel difference between the two points in the x-axis direction is 180, and this value can be used as its Euclidean distance. Since the y-axis difference is 0, the Euclidean distance is 180 pixels. Subsequently, screen the edge points of the crack region, and select the outermost boundary pixel points within the main direction range of this region. For example, the left edge point is (121, 341), and the right edge point is (279, 340), and their pixel spacing is 158 pixels. Then the difference between the boundary spacing and the Euclidean distance in the main direction is 22 pixels, indicating that there is a slight deviation or irregular bending at the edge of this crack region. Then select another crack segment with starting and ending points (100, 200) and (280, 210). Its main direction span is 180 pixels, and the maximum spacing of the edge points is 230 pixels, so the difference is 50 pixels. If the set maximum reasonable difference threshold is 30 pixels, then the edge extension of the second crack segment exceeds the main direction control range and is judged as an irregular extension structure. Further calculate the main direction extension value, edge extension range value, and direction angle difference of each crack segment according to such a spatial structure, and jointly construct the spatial span vector set of each crack segment through the Euclidean distance and boundary difference of each crack segment. Finally, organize and summarize the extension direction, pixel length, and edge offset information of multiple crack segments to obtain the crack extension geometric quantity.

[0120] S412: Invoke the crack extension geometric quantity, combine the boundary point width and curvature change, and use the formula:

[0121]

[0122] Calculate the crack non-linear change degree RU, judge whether the crack is a continuous crack or an intermittent crack, and obtain the crack geometric change trend. Among them, wu z is the width value of the z-th measurement point, is the average width, ku z is the curvature value of the z-th measurement point, which is a quantitative index of the bending degree of the crack at this point, is the average curvature, θu z is the direction change angle of the z-th measurement point, indicating the deflection degree of the crack direction at this point, θu max is the maximum angle change value, N ru is the number of key measurement points included in the crack;

[0123] Suppose 2 key measurement points are extracted from a certain crack segment, and the width, curvature, and direction angle data are collected respectively to quantify the non-linear change behavior. The width value of measurement point 1 is 2.0 mm, the curvature value is 0.111 / m, and the direction change angle is 16 degrees. The width value of measurement point 2 is 2.3 mm, the curvature value is 0.141 / m, and the direction change angle is 20 degrees. Then the average width wu of the crack segment is (2.0 + 2.3) / 2 = 2.15 mm, and the average curvature The maximum direction angle θu max = 20 degrees. Substitute it into the formula, where N ru = 2. When calculating the first measurement point item, |2.0 - 2.15| / 2.15 = 0.0698, The direction angle ratio is 16 / 20 = 0.8, then the first item is:

[0124] 0.0698+(0.1225×0.8)=0.0698 + 0.0980 = 0.1678;

[0125] When calculating the second measurement point item, |2.3 - 2.15| / 2.15 = 0.0698, The direction angle ratio is 20 / 20 = 1, and the second item is:

[0126] 0.0698+(0.1225×1)=0.0698 + 0.1225 = 0.1923;

[0127] Take the average value:

[0128] RU=(0.1678 + 0.1923) / 2 = 0.1801;

[0129] It is obtained that the non-linear change degree RU of this crack is 0.1801. If the set threshold is 0.12, it is judged that this crack is an intermittent crack, and then the corresponding crack geometric change trend is obtained. By constructing the normalized product of the width deviation ratio, the square root value of the curvature offset, and the angle change, a comprehensive non-linear behavior quantification mechanism is established, which can still form an effective identification standard under the condition of less data volume, and realize the numerical quantification and delimitation of complex crack structures such as bending, expansion, and folding.

[0130] S413: According to the crack geometric change trend, analyze the relationship between the distance and angle between the crack endpoints, judge whether the adjacent crack segments are in a continuous connection state, and generate crack connectivity detection data;

[0131] Extract the starting and ending endpoint coordinates of each crack segment, and construct adjacent crack pairs by pairwise combination in turn. Calculate the Euclidean distance and the direction angle difference between each pair of endpoints. Taking the end point of crack 1 as (120, 200) and the starting point of crack 2 as (130, 198) as an example, the Euclidean distance is Pixels with direction angles of 20° and 23.2° respectively, and the difference is 3.2°. If the preset connectivity discrimination condition is that the distance between endpoints is less than or equal to 12 pixels and the angle difference does not exceed 5°, then this crack segment can be regarded as a continuous structure. If there are multiple crack segments that meet the above conditions, the segments will be marked with the same crack cluster number, and the corresponding connectivity mapping relationship will be recorded. Finally, through the coordinate matching matrix and the comparison of direction differences, it is judged whether there is geometric connectivity between all crack segments, and the output is organized into a set of matrix-connected comparison relationships.

[0132] Please refer to Figure 6 , and the steps for obtaining the crack positioning result are specifically as follows:

[0133] S511: Based on the crack connectivity detection data, quantitatively measure the spatial continuity of the fractured cracks, compare it with the regional average crack length and the trend change, judge whether there is a structural fracture in the cracks, and obtain the fractured crack identification set;

[0134] The cracks are divided into several detection units, the spatial continuity index of each crack is obtained and its measured value is recorded. The measurement method is based on the length difference of the connectivity section between adjacent crack endpoints. A spatial connection network of the crack and the surrounding cracks is constructed in space. Taking cracks F001 to F004 as examples, their spatial continuity indexes are measured to be 3.2m, 2.1m, 5.4m, and 1.6m respectively. Subsequently, the length of each crack is measured and calculated through the coordinate distance formula. For example, the starting and ending points of F002 are (2.1, 1.5) and (4.2, 2.3), then its length is √[(4.2 - 2.1) 2 +(2.3 - 1.5) 2 ≈ 2.3m. By comparing the deviation value between the obtained crack length and the regional average crack length of 2.5m, the deviation of F002 is 0.2m. Combining the measurement of the trend change value and using the line segment direction angle calculation method, analyze the angle change degree between the current crack line segment and the adjacent line segments. The difference between the direction angle of F002 and the average direction angle of the surrounding cracks is 6.3°. Taking the spatial continuity offset less than 2.5m, the length deviation not exceeding ±0.4m, and the trend change less than 10° as the structural fracture determination threshold, and comparing and judging all cracks item by item accordingly. Finally, cracks F002 and F004 that do not meet the determination criteria are screened out to obtain the fractured crack identification set.

[0135] S512: Based on the fractured crack identification set, analyze the center point position and the starting and ending point coordinates of the cracks, calculate the deviation of the average distance between them and the adjacent cracks, and evaluate the trend deviation angle to obtain the crack coordinate offset interval;

[0136] Extract the central point positions of F002 and F004 respectively, calculate their average values through the start and end coordinates. For example, the central point of F002 is [(4.1 + 6.5) / 2] = 5.3m. At the same time, collect the corresponding start and end coordinate data, calculate the average distance value from it to the nearest neighbor crack, and judge its deviation degree through difference comparison. For example, the distance between F002 and the adjacent crack is 2.7m. Compared with the set reference distance of 2.0m, the deviation is 0.7m, exceeding the distance difference threshold of 0.5m. Subsequently, evaluate the deviation angle of the trend. Using the direction vector included angle formula, the deviation angle of F002 is 12.5°. Compared with the reference value of the trend change angle of 10°, it is 2.5° larger. Based on the adjacent distance deviation threshold and the trend deviation angle threshold, jointly judge the spatial deviation state of the crack. Mark the direction component that needs to be corrected for the crack on the spatial coordinate as east-north, and the deviation distance is 0.7m. Thus, construct the spatial correction vector range of each crack to obtain the crack coordinate deviation interval.

[0137] S513: According to the crack coordinate deviation interval, adjust the position of the central point coordinates of the crack, re-identify its geometric shape, determine the current position coordinates of the crack, and obtain the crack positioning result;

[0138] Adjust the position of the central point coordinates of F002 and F004 respectively. The original central point of F002 is 5.3m. After correction, the coordinate is offset 0.7m east-north from 5.3m, and the new coordinate is 5.98m. Re-extract the full-section coordinates of the corresponding corrected crack, construct a new crack line segment, calculate its corrected length value and direction angle change, and obtain geometric shape indicators. The length of F002 after correction is 2.4m, and the direction angle change is 8.3°. Comparing with the original data, it is found that the structural deviation decreases. Combine all the corrected crack data to recalculate the standard deviation of the structural stability. If the standard deviation is less than the initial structural reference variance value of 0.35, confirm that its shape is stable and the position is coordinated, and use the crack coordinates that meet the conditions as the final corrected coordinates to obtain the crack positioning result.

[0139] A water conservancy project crack detection system based on intelligent vision recognition. The system includes:

[0140] The image processing and enhancement module obtains the water conservancy project crack image, divides the image into multiple regions according to the illumination information, detects the brightness change of each region, calculates its local contrast, performs brightness adjustment, and refines the contrast between the crack and the background by adjusting the regional contrast weight to obtain a locally enhanced image;

[0141] Based on the locally enhanced image, the crack edge optimization module extracts the crack contour, performs a morphological closing operation to fill in the missing parts of the crack edge, eliminates the edge distortion caused by noise or uneven illumination, and then removes the background noise through a morphological opening operation, and analyzes the crack morphological characteristics according to the geometric characteristics of the crack to obtain the crack edge optimization information;

[0142] Based on the crack edge optimization information, the crack area segmentation module compares the image pixel values with the background area pixel values to segment the crack area. By analyzing the geometric features of the crack area, it determines whether there are fracture or connection problems, locates the endpoints and continuous areas of the crack, and obtains the crack area segmentation result;

[0143] Based on the crack area segmentation result, the crack connectivity analysis module performs connectivity analysis on the crack area according to the geometric features of the crack, calculates the fracture endpoints, length, width and curvature of the crack, determines whether the crack is a continuous crack or an intermittent crack, and analyzes the relative positions of the fracture endpoints of the crack to generate crack connectivity detection data;

[0144] Based on the crack connectivity detection data, the crack location calibration module uses geometric constraints to locate misjudged fractured cracks, calibrates the crack position according to the geometric features of the crack, determines the current position coordinates of the crack, and obtains the crack location result.

[0145] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the disclosed technical content to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A crack detection method for hydraulic engineering based on intelligent vision recognition, characterized in that, The following steps are involved: S1: Obtain a crack image of a hydraulic engineering project, divide the image into multiple regions, detect the brightness change of each region, calculate its local contrast, adjust the brightness, and refine the contrast between the crack and the background by adjusting the regional contrast weight to obtain a locally enhanced image; S2: Based on the locally enhanced image, a morphological closing operation is performed to fill in the missing part of the crack edge and eliminate the edge distortion caused by noise or uneven illumination. Then, a morphological opening operation is performed to remove the background noise, retain the effective area of the crack, calculate its length, width and curvature, and obtain the crack edge optimization information; S3: Based on the crack edge optimization information, the crack area is segmented by comparing the image pixel value with the background area pixel value, and the geometric features of the crack area are analyzed to determine whether there is a break or connection problem, thereby obtaining a crack area segmentation result; S4: Based on the crack region segmentation result, a connectivity analysis is performed on the crack region to determine whether the crack is a continuous crack or a discontinuous crack, and the relative positions of the fracture endpoints of the crack are analyzed to generate crack connectivity detection data.

2. The crack detection method for hydraulic engineering based on intelligent vision recognition according to claim 1, wherein, The locally enhanced image includes a regional image after brightness adjustment, a regional image after contrast optimization, and a crack edge refinement image; the crack edge optimization information includes crack edge filling data and crack smoothness data; the crack region segmentation result includes a segmentation threshold, crack region boundary data, and crack endpoint positioning results; the crack connectivity detection data includes crack continuity status and crack fracture endpoint data.

3. The crack detection method for hydraulic engineering based on intelligent vision recognition according to claim 1, characterized in that The steps of acquiring the locally enhanced image are specifically as follows: S111: Acquire a crack image of a hydraulic engineering project, divide the image into multiple regions according to illumination information, monitor brightness changes of each region, calculate its local contrast, and generate a local contrast value; S112: According to the local contrast value, brightness is adjusted according to the contrast difference of each area, by adjusting the contrast weight of each area, using the formula: Obtain the adjusted local contrast CR adjusted , where CR i represents the local contrast value of region i, WR i represents the contrast weight of region i, N cr represents the total number of regions, ΔLR represents the local luminance difference, and CR max represents the maximum value of the local contrast; S113: Based on the adjusted local contrast, the crack edge area is optimized, and the contrast between the crack and the background is refined to obtain a locally enhanced image.

4. The crack detection method for hydraulic engineering based on intelligent vision recognition according to claim 1, characterized in that, The steps for obtaining the crack edge optimization information are specifically as follows: S211: based on the locally enhanced image, extract the grayscale gradient and structural direction change value of the crack contour area, perform a closing operation to fill the discontinuous edge, and obtain an edge distortion positioning interval; S212: calling the edge distortion positioning interval, analyzing the roughness of the crack boundary pixels and the contour fitting residual, judging the smoothness and consistency, using the formula: Calculate the collaborative value CE of crack profile continuity r , reconstruct the edge pixel structure to obtain a smooth image of the crack profile, where represents the boundary roughness of the k-th pixel, is the mean value of the boundary roughness, is the gray level jump intensity of the k-th pixel, G (avg) is the average value of the gray level jump intensity, n ce represents the total number of pixels; S213: Based on the crack contour smoothed image, background noise is removed by opening operation, the effective crack area is retained, the connectivity and closure of the crack are analyzed, key geometric features of the crack are identified, and crack edge optimization information is obtained.

5. The crack detection method for hydraulic engineering based on intelligent visual recognition according to claim 1, characterized in that The steps for obtaining the crack region segmentation result are specifically as follows: S311: Based on the crack edge optimization information, compare the image pixel value with the background area pixel value, analyze the grayscale difference of the data, sort the grayscale difference, determine the key difference baseline, and obtain the crack pixel difference interval; S312: Based on the crack pixel difference interval, analyze the pixel gradient and the background intensity difference, using the formula: Obtain the optimal threshold range TB for crack segmentation opt , screen the pixel regions that meet the conditions, and obtain the optimized parameters of the crack grayscale boundary, where GB k represents the local gradient change intensity of the k-th pixel, BE k represents the average background pixel intensity within the k-th pixel, DB k represents the grayscale difference between the k-th pixel block and its adjacent blocks, and n ce represents the total number of pixels; S313: Based on the crack gray boundary optimization parameters, evaluate the continuity of the edge points within the region, determine whether there are problems of fracture or connection, locate the endpoints and continuous regions of the crack, and obtain the crack region segmentation result.

6. The crack detection method for hydraulic engineering based on intelligent visual recognition according to claim 1, characterized in that The specific steps for obtaining the crack connectivity detection data are as follows: S411: Based on the crack region segmentation result, analyze the difference between the Euclidean distance between the starting and ending points of the main direction coordinate axis and the crack pixel boundary, identify the spatial span and geometric extension direction of the crack line segment, and obtain the crack extension geometric quantity; S412: Invoke the crack extension geometric quantity, combine the boundary point width and curvature change, using the formula: Calculate the non - linear change degree RU of the crack, determine whether the crack is a continuous crack or an intermittent crack, and obtain the geometric change trend of the crack, where wu z is the width value of the z - th measurement point, is the average width, ku z is the curvature value of the z - th measurement point, is the average curvature, θu z is the direction change angle of the z - th measurement point, θu max is the maximum angle change value, N ru is the number of key measurement points included in the crack; S413: According to the crack geometric change trend, analyze the distance and angle relationship between the crack endpoints, determine whether the adjacent crack segments are in a continuous connection state, and generate the crack connectivity detection data.

7. The crack detection method for hydraulic engineering based on intelligent vision recognition according to claim 1, wherein, The steps further include: S5: Based on the crack connectivity detection data, use geometric constraints to locate the misjudged fractured cracks, calibrate the crack position according to the crack geometric characteristics, determine the current position coordinates of the crack, and verify the geometric shape of the crack to obtain the crack positioning result; The crack positioning result includes the crack position coordinates, the crack geometric shape calibration data, and the crack correction information.

8. The method for detecting cracks in hydraulic engineering based on intelligent vision recognition according to claim 7, wherein The specific steps for obtaining the crack positioning result are as follows: S511: Based on the crack connectivity detection data, quantitatively measure the spatial continuity of the fractured cracks, compare it with the regional average crack length and trend change, and determine whether there are structural fractures in the cracks to obtain the fractured crack identification set; S512: Based on the fractured crack identification set, analyze the center point position and the starting and ending point coordinates of the crack, calculate the deviation from the average spacing between it and the adjacent cracks, and evaluate the trend deviation angle to obtain the crack coordinate offset interval; S513: According to the crack coordinate offset interval, adjust the position of the center point coordinate of the crack, re-identify its geometric shape, determine the current position coordinates of the crack, and obtain the crack positioning result.

9. A crack detection system for hydraulic engineering based on intelligent visual recognition, characterized in that, Execute according to the water conservancy project crack detection method based on intelligent vision recognition described in any one of claims 1-8. The system includes: The image processing enhancement module acquires the water conservancy project crack image, divides the image into multiple regions according to the illumination information, detects the brightness change of each region, calculates its local contrast, performs brightness adjustment, and refines the contrast between the crack and the background by adjusting the regional contrast weight to obtain the locally enhanced image; The crack edge optimization module, based on the locally enhanced image, extracts the crack contour, performs a morphological closing operation to fill in the missing parts of the crack edge, eliminate the edge distortion caused by noise or uneven illumination, and then removes the background noise through a morphological opening operation, and analyzes the crack morphological characteristics according to the geometric characteristics of the crack to obtain the crack edge optimization information; The crack area segmentation module divides the crack area based on the optimized crack edge information by comparing the image pixel values with the background area pixel values. By analyzing the geometric features of the crack area, it determines whether there are problems of fracture or connection, locates the endpoints and continuous areas of the crack, and obtains the crack area segmentation result; The crack connectivity analysis module performs connectivity analysis on the crack area based on the crack area segmentation result according to the geometric features of the crack, calculates the fracture endpoints, length, width, and curvature of the crack, determines whether the crack is a continuous crack or an intermittent crack, and analyzes the relative positions of the fracture endpoints of the crack to generate crack connectivity detection data; The crack location calibration module locates the misjudged fracture cracks using geometric constraints based on the crack connectivity detection data, calibrates the crack position according to the crack geometric features, determines the current position coordinates of the crack, and obtains the crack location result.

Images