Water conservancy project lining crack monitoring method and system

Abstract

The invention discloses a hydraulic engineering lining crack monitoring method and system, and belongs to the technical field of optical crack monitoring, and the method comprises the steps: employing a high-definition optical imaging device to collect and carry out the graying conversion of an original image, generating a crack candidate region, and carrying out the combined screening of an average gray scale gradient, a normalized crack length and a background noise threshold value, and obtaining a crack candidate region. The method comprises the following steps: acquiring a crack confidence coefficient score, correcting a leak detection region caused by illumination abrupt change or local low gradient, acquiring a comprehensive detection score, calculating a crack propagation rate in combination with the change of the comprehensive detection score in a time sequence image sequence, and acquiring updated background noise in combination with the crack propagation rate and a current threshold updating period. According to the method, the breakthrough from'static threshold screening 'to'dynamic self-adaptive monitoring' is realized, and the core problems of'erroneous judgment, missing detection and difficulty in dynamic tracking 'in hydraulic engineering lining crack monitoring are comprehensively solved.

Description

Technical Field

[0001] This invention relates to the field of optical crack monitoring technology, and in particular to a method and system for monitoring cracks in the lining of hydraulic engineering projects. Background Technology

[0002] Currently, the monitoring of cracks in the lining of water conservancy projects mainly relies on traditional optical imaging technology combined with fixed threshold algorithms. Its core principle is to screen crack areas by setting static parameters such as grayscale gradient and area.

[0003] However, in practical applications, due to the complex environment of water conservancy projects, existing technologies have the following key problems: First, non-crack areas such as stains, scratches, and water reflections on the lining may be misjudged as cracks due to local high gray-scale gradients. Existing technologies usually judge cracks based on a single gray-scale gradient threshold without considering the spatial morphological characteristics of cracks. Second, crack edges may be judged as non-cracks due to local gradient drops caused by concrete air bubbles. Sudden changes in illumination may also cause a sharp drop in the gray-scale gradient of a crack in a frame, leading to direct missed detection. However, existing technologies are mostly single-frame static detections and do not combine the spatial continuity and temporal change trend of cracks. Finally, when cracks expand rapidly (such as when water pressure causes concrete to crack), a large amount of debris and vibration noise will be generated. The fixed threshold of existing technologies cannot filter these out, leading to misjudgments of "multiple cracks." After the crack stabilizes, its characteristics weaken and it may be judged as a non-crack. This makes it difficult for existing monitoring methods to respond to dynamic changes in crack expansion or environmental noise. Summary of the Invention

[0004] The technical problem this invention aims to solve is that existing technologies suffer from high misjudgment rates of non-crack features, high missed detection rates of cracks, and weak dynamic propagation tracking capabilities. To address this, we propose a method and system for monitoring cracks in hydraulic engineering linings. The specific technical solution adopted is as follows: In a first aspect, one embodiment of the present invention provides a method for monitoring cracks in the lining of hydraulic engineering projects, the method comprising the following steps: A time-series image sequence of the lining surface of a hydraulic engineering project was acquired using a high-definition optical imaging device. After converting the original image to grayscale, a 3×3 Gaussian filter was used to remove sensor noise, and histogram equalization was used to enhance image contrast. Based on the preprocessed image, potential crack edges are extracted using the Canny edge detection algorithm, and the fracture edges are connected by morphological dilation-erosion operations to generate crack candidate regions, while non-crack regions are marked. Calculate the average gray-level gradient and normalized crack length of the crack candidate region, distinguish crack edges from smooth background, calculate background noise threshold, filter non-crack interference with gray-level values ​​lower than background noise threshold, introduce region consistency factor, and verify whether the region is a continuous crack by the gray-level difference of neighboring pixels. By jointly filtering the average gray-level gradient, the normalized crack length, and the background noise threshold, a crack confidence score is obtained, and based on this, a preliminary crack region and a non-crack region are determined from the crack candidate region. SIFT feature point matching is performed on the initially selected crack region at time t-1 and time t in the time-series image sequence, the confidence change is calculated, and the missed areas caused by sudden changes in illumination or local low gradients are corrected based on the crack confidence score, the confidence change and the region consistency factor, and the final crack region is updated to obtain the comprehensive detection score. By combining the comprehensive detection score change of the final crack region at time t-1 and time t in the time-series image sequence, the crack propagation rate is calculated, the background noise threshold of the previous update cycle is obtained, and the updated background noise is obtained by combining the crack propagation rate with the current threshold update cycle. The updated background noise is directly written into the system register and overwrites the background noise threshold of the previous update cycle, and is simultaneously synchronized to the monitoring terminal for record-keeping via the wireless transmission module.

[0005] Preferably, the methods for determining the average gray-level gradient and the background noise threshold are as follows: The optical monitoring image is converted into a grayscale image. A 3×3 Sobel operator is used to traverse all pixels in the crack candidate region. The gradient values ​​in the horizontal and vertical directions are calculated respectively. The square root of the sum of the squares of the gradient values ​​of each pixel is taken to obtain the gradient magnitude. The gradient magnitudes of all pixels in the crack candidate region are counted and the arithmetic mean is calculated as the average grayscale gradient of the crack candidate region. Among them, the gradient magnitude weight of the crack edge region is higher than that of the inner region, and the weight coefficient decreases linearly with the increase of the distance from the pixel to the edge. A pure background region is selected from the monitored image, and the gray value sequence of this region is extracted. After removing high-frequency noise using Gaussian filtering, the mean and standard deviation of the gray values ​​are calculated. The gray value sequence is binarized using the Otsu's method. When the pixel ratio of the background region exceeds 99.5%, the mean plus 2.5 times the standard deviation is used as the initial noise threshold. The gray value change of the background region is monitored in real time through a sliding window. The mean and standard deviation are recalculated every 100 frames of images, and the background noise threshold is dynamically updated to adapt to the light fluctuation.

[0006] Preferably, the specific method for obtaining a crack confidence score based on the joint screening of the average gray-level gradient, the normalized crack length, and the background noise threshold, and thereby determining the initial crack region and non-crack regions from the crack candidate regions, is as follows: The crack length of the crack candidate region is measured by a contour extraction algorithm, and the maximum crack length in the preprocessed image is extracted. The ratio of the crack length to the maximum crack length is denoted as the normalized crack length. The result of multiplying the average gray-level gradient by the normalized crack length is denoted as the spatial comprehensive score of the crack. The result of subtracting the background noise threshold from the spatial comprehensive score is denoted as the crack confidence score. If the crack confidence score is greater than 0, it is determined to be a preliminary crack region; If the crack confidence score is less than or equal to 0, it is determined to be a non-crack area.

[0007] Preferably, the methods for determining the regional consistency factor and the confidence level change are as follows: For each initially selected crack region pixel (x,y), a 5×5 pixel square neighborhood is defined with it as the center, covering 25 pixels. If the initially selected crack region pixel (x,y) is located at the edge of the image and the neighborhood exceeds the image range, the pixels that exceed the boundary are treated as background pixels. The number of positive confidence pixels in the neighborhood is counted, and the ratio of the number of positive confidence pixels in the neighborhood to the total number of pixels in the neighborhood is recorded as the region consistency factor. If the regional consistency factor is greater than or equal to 0.6, it is determined to be the final crack region; If the regional consistency factor is less than 0.6, it is determined to be a non-crack region; The SIFT feature point matching algorithm is used to align the initial crack region in the current frame with the corresponding region in the previous frame. The area overlap rate, edge coincidence, and gray-scale distribution similarity are calculated and weighted to obtain the crack confidence score of the current frame and the crack confidence score of the previous frame. The absolute value of the difference between the crack confidence score of the current frame and the crack confidence score of the previous frame is the confidence change.

[0008] Preferably, the method for correcting missed areas caused by sudden changes in illumination or local low gradients, updating the final crack area, and obtaining the comprehensive detection score based on the crack confidence score, the change in confidence score, and the regional consistency factor is as follows: The product of the crack confidence score of the final crack region and the region consistency factor is recorded as the spatial feature score of the current frame. The sum of the spatial feature score and the confidence change is recorded as the comprehensive detection score.

[0009] Preferably, the method for calculating the crack propagation rate based on the change in the comprehensive detection score of the final crack region between time t-1 and time t in the time-series image sequence is as follows: The overall detection score is recorded as the overall detection score at time t; The difference between the comprehensive detection score at time t and the comprehensive detection score at time t-1 is recorded as the comprehensive detection change score. The time interval between the acquisition of the two frames of images at time t and time t-1, which represent the combined detection scores, is denoted as the time interval. The result of dividing the comprehensive detection change score by the time interval is recorded as the crack propagation rate; When the crack propagation rate is greater than or equal to 0, it reflects the crack propagation in the final crack region; When the crack propagation rate is less than 0, it reflects the weakening of the crack in the final crack region.

[0010] Preferably, the method for updating the background noise threshold based on the two scenarios of crack propagation and crack weakening in the final crack region is as follows: Obtain the background noise threshold from the previous update cycle; The product of the crack propagation rate and the current threshold update cycle is recorded as the threshold adjustment magnitude. The adjusted range is weighted and then added to the background noise threshold of the previous update cycle to obtain the updated background noise. If the final crack region is a crack extension, then the background noise increases after the update; If the final crack region represents a reduction in crack size, then the updated background noise is reduced.

[0011] Secondly, embodiments of the present invention also provide a system for monitoring cracks in the lining of hydraulic engineering projects, comprising: Optical imaging module: Utilizes a high-definition industrial camera and a ring light to acquire grayscale images of the lining surface at a resolution of 1920×1080; Feature extraction module: an embedded FPGA chip that calculates the average gray-level gradient in real time using the Sobel operator and outputs the background noise threshold using the Otsu algorithm; Region Analysis Module: Utilizes an industrial control computer to perform 8-neighborhood traversal and contour regularity analysis, and outputs the region consistency factor; Dynamic tracking module: The processor is equipped with a GPU acceleration unit for the SIFT matching algorithm, which calculates the confidence change and dynamically adjusts the current threshold update cycle, outputting the crack propagation rate and the updated background noise; The output module uses an LCD touchscreen and a 4G transmission unit to display the location, length, and expansion trend of cracks in real time. It writes the updated background noise directly into the system register, overwrites the background noise threshold of the previous update cycle, and uploads it to the cloud monitoring platform.

[0012] The technical effects and advantages of this invention are as follows: In this invention, a "spatial comprehensive score" is constructed by multiplying the average gray-level gradient, which reflects the clarity of the crack edge, and the normalized crack length, which reflects the spatial continuity of the crack. Then, the background noise threshold is subtracted to achieve the filtering of non-crack interference after multi-dimensional feature screening, thereby reducing the misjudgment rate of lining cracks.

[0013] In this invention, the comprehensive detection score incorporates regional consistency and temporal variation, which can correct single-frame detection errors and thus avoid direct filtering. Furthermore, when a sudden change in illumination causes the confidence of the current frame to rise, it can be corrected to a "suspected crack" to avoid missed detection.

[0014] In this invention, the noise threshold is dynamically adjusted by calculating the updated background noise, so that the background noise threshold is linked with the crack propagation speed in real time. Specifically, when the crack propagates rapidly, the background noise threshold automatically increases to filter out debris noise during the propagation process; when the crack stabilizes, the background noise threshold automatically decreases to retain weak feature cracks.

[0015] In addition, the crack confidence score serves as the input for calculating the comprehensive detection score, realizing "spatial filtering → spatiotemporal optimization". The output comprehensive detection score calculates the crack propagation speed, which can drive the update of the background noise threshold of the previous update cycle. The updated background noise threshold, i.e. the updated background noise, reacts to the crack confidence score, forming a dynamic closed loop of "crack features → threshold adjustment → feature re-filtering". Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the method for monitoring cracks in the lining of this water conservancy project. Figure 2 This is a flowchart illustrating the process of acquiring information for monitoring crack conditions according to the present invention. Detailed Implementation

[0017] The present invention will now be described in further detail with reference to the accompanying drawings and preferred embodiments.

[0018] Reference Figure 1 and Figure 2 As shown, the present invention provides a technical solution: a method for monitoring cracks in the lining of hydraulic engineering projects, the method comprising the following steps: Step 1: Use high-definition optical imaging equipment to acquire time-series image sequences of the lining surface of the hydraulic engineering project. After converting the original images to grayscale, use 3×3 Gaussian filtering to remove sensor noise and enhance image contrast through histogram equalization. Based on the preprocessed image, potential crack edges are extracted using the Canny edge detection algorithm, and the fracture edges are connected by morphological dilation-erosion operations to generate crack candidate regions, while non-crack regions are marked.

[0019] Among them, grayscale conversion simplifies color images into single-channel grayscale data, reducing computational complexity; 3×3 Gaussian filtering effectively smooths sensor noise (such as thermal noise of imaging chips) and reduces texture interference; histogram equalization stretches the grayscale dynamic range and enhances the contrast between crack edges and background (such as increasing the grayscale difference of cracks under low light from 10 levels to 30 levels), making the features of fine cracks more prominent. The Canny edge detection algorithm accurately captures potential crack contours through dual-threshold filtering (high threshold for detecting strong edges, low threshold for connecting weak edges); morphological dilation-erosion operation (first dilation connects fracture edges, then erosion removes burrs) aggregates scattered crack pixels into continuous regions, reducing the risk of missed edge fractures caused by bubbles and textures on the concrete surface; and simultaneously labels non-crack areas (such as smooth concrete surfaces and construction joints) to narrow the scope of subsequent analysis, allowing computational resources to focus on high-probability crack areas and improving processing efficiency.

[0020] Thus, high-definition optical imaging and multi-stage preprocessing technology can provide a high-quality image data foundation for crack monitoring.

[0021] Step 2: Calculate the average gray-level gradient and normalized crack length of the crack candidate region, distinguish crack edges from smooth background, calculate background noise threshold, and filter out non-crack interference with gray-level values ​​lower than the background noise threshold; The methods for determining the average gray-level gradient and the background noise threshold are as follows: The optical monitoring image is converted into a grayscale image. A 3×3 Sobel operator is used to traverse all pixels in the crack candidate region. The gradient values ​​in the horizontal and vertical directions are calculated respectively. The square root of the sum of the squares of the gradient values ​​of each pixel is taken to obtain the gradient magnitude. The gradient magnitudes of all pixels in the crack candidate region are counted and the arithmetic mean is calculated as the average grayscale gradient of the crack candidate region. Among them, the gradient magnitude weight of the crack edge region is higher than that of the inner region, and the weight coefficient decreases linearly with the increase of the distance from the pixel to the edge. A pure background area (a uniform concrete surface without cracks, stains, or scratches) is selected from the monitoring image. The gray value sequence of this area is extracted. After removing high-frequency noise using Gaussian filtering, the mean and standard deviation of the gray values ​​are calculated. The gray value sequence is binarized using the maximum inter-class variance method. When the pixel ratio of the background area exceeds 99.5%, the mean plus 2.5 times the standard deviation is used as the initial noise threshold. The gray value change of the background area is monitored in real time through a sliding window. The mean and standard deviation are recalculated every 100 frames of images to dynamically update the background noise threshold to adapt to the light fluctuation. The formula for calculating the average gray-level gradient is as follows: ; ; ; In the formula: G is the average gray-level gradient, G x G represents the gradient value in the horizontal direction. y Let I(x,y) be the gradient value in the vertical direction, I(x,y) be the gray value of pixel (x,y), G(x,y) be the gradient magnitude of each pixel, and N be the total number of pixels in the current crack candidate region, derived from the average gray-level gradient formula: x is the horizontal coordinate of the pixel, which, together with the vertical coordinate y, represents the position of the pixel in the image (e.g., (x,y) is the pixel coordinate).

[0022] Used for scenarios such as gradient calculation and confidence change calculation; The formula for calculating the background noise threshold is as follows: Manually select 3-5 crack-free background areas (such as a smooth concrete surface) in the optical image, each area ≥ 100 × 100 pixels, denoted as set B = {B1, B2, ... B}. k}; ; In the formula: BZ0 is the background noise threshold, M is the total number of pixels in the background region, and B i For the i-th background region, μ b G is the average gradient of the background region. b(x,y) This represents the gradient magnitude of the pixel (x, y) within the background region.

[0023] Furthermore, the specific method for obtaining a crack confidence score by jointly filtering the average gray-level gradient, normalized crack length, and background noise threshold, and then determining the initial crack region and non-crack regions from the crack candidate region based on this score, is as follows: The crack length of the crack candidate region is measured by a contour extraction algorithm, and the maximum crack length in the preprocessed image is extracted. The ratio of crack length to maximum crack length is denoted as the normalized crack length. The result of multiplying the average gray-level gradient by the normalized crack length is denoted as the spatial comprehensive score of the crack. The result of subtracting the background noise threshold from the spatial comprehensive score is denoted as the crack confidence score. If the crack confidence score is greater than 0, it is determined to be a preliminary crack region; If the crack confidence score is less than or equal to 0, it is determined to be a non-crack area.

[0024] The formula for calculating the crack confidence score is as follows: ; In the formula: ZX is the crack confidence score, Lnorm This represents the normalized crack length.

[0025] Previously, crack detection often focused on "brightness changes"—if the pixel brightness at the crack edge changed rapidly (e.g., from dark to bright), it was considered a crack. However, surface scratches and stains may also have localized rapid brightness changes (e.g., a deep scratch with obvious brightness jumps at the edge). In step 2 of this embodiment, two key indicators are used to determine the crack: the rate of brightness change (average grayscale gradient) and the relative length of the crack (normalized crack length). Specifically, a true crack is not just one pixel wide and a few pixels long, but has a certain continuity (for example, occupying at least 30% of the image width). Scratches and stains, on the other hand, are often "short and fragmented." For example, a scratch may have rapid brightness changes, but its length is only 5% of the image. This kind of "small" interference will be filtered out. In addition, a "background noise threshold" is introduced—first, the brightness fluctuation of the "clean area" (the area without cracks or stains) on the lining surface is counted. Only those exceeding this fluctuation range are likely cracks. For example, the background noise threshold is set to "brightness change not exceeding 40". If the brightness of an area changes rapidly (for example, 80), but its length is too short (the overall score is 16), not exceeding 40, it is judged as a "false crack". Thus, through the triple screening of "brightness change + length + background noise", 60 out of 100 areas previously considered cracks are actually stains or scratches. Now, these "false cracks" can be basically screened out, making the monitoring results more reliable.

[0026] At this point, the crack confidence score is obtained, and based on this score, the initial crack region and non-crack region are determined from the crack candidate region.

[0027] Step 3: Introduce a regional consistency factor to verify whether the region is a continuous crack by the difference in grayscale of neighboring pixels; SIFT feature point matching is performed on the initially selected crack regions at time t-1 and time t in the time-series image sequence. The confidence change is calculated. Based on the crack confidence score, the confidence change, and the region consistency factor, the missed regions caused by sudden changes in illumination or local low gradients are corrected and updated to the final crack regions. The comprehensive detection score is obtained. The method for obtaining the comprehensive detection score is as follows: The product of the crack confidence score of the final crack region and the region consistency factor is recorded as the spatial feature score of the current frame. The sum of the spatial feature score and the confidence change is recorded as the comprehensive detection score.

[0028] The formula for calculating the comprehensive test score is as follows: ; In the formula: D is the comprehensive test score, △ZX prevR is the confidence level change, and R is the regional consistency factor, which represents the ratio of the number of positive confidence pixels in the neighborhood to the total number of pixels in the neighborhood (the neighborhood is 5×5 pixels, and the total number of pixels is 25).

[0029] The methods for determining the regional consistency factor, confidence level change, and threshold update period are as follows: For each initially selected crack region pixel (x, y), a 5×5 pixel square neighborhood is defined centered on it (horizontal x-axis: i=-2,-1,0,1,2; vertical y-axis: j=-2,-1,0,1,2), covering 25 pixels, i.e.: Neighboring pixel set = {(x-2,y-2), (x-2,y-1), ..., (x,y), ..., (x+2,y+1), (x+2,y+2)} If the pixel (x, y) in the initially selected crack region is located at the image edge (e.g., x < 2 or y < 2, or x > image width - 3, or y > image height - 3), and the neighborhood exceeds the image range, then the pixels exceeding the boundary are treated as background pixels, i.e., their C(x+i, y+j) = 0 (regardless of whether the pixel actually exists). This avoids statistical errors caused by boundary truncation. The number of positive confidence pixels in the neighborhood is counted. The specific method for counting the number of positive confidence pixels is as follows: Iterate through the 25 pixels in the 5×5 neighborhood and determine whether the confidence level C(x+i,y+j) of each pixel is greater than 0: If C(x+i,y+j)>0, it is recorded as a positive confidence pixel, and the number of positive confidence pixels is accumulated. If C(x+i,y+j)≤0 or the pixel exceeds the image boundary, it is not included in the count; The ratio of the number of positive confidence pixels in the neighborhood to the total number of pixels in the neighborhood is denoted as the region consistency factor. If the regional consistency factor is greater than or equal to 0.6, it is determined to be the final crack region; If the regional consistency factor is less than 0.6, it is determined to be a non-crack region; Using the SIFT feature point matching algorithm, the initial crack region in the current frame is aligned with the corresponding region in the previous frame. The area overlap rate, edge coincidence, and gray-scale distribution similarity are calculated. The crack confidence score of the current frame and the crack confidence score of the previous frame are obtained by weighted summation. The absolute value of the difference between the crack confidence score of the current frame and the crack confidence score of the previous frame is the confidence change. The initial threshold update cycle is set to 50 frames. When the confidence change is greater than 0.2 for three consecutive times (indicating rapid crack expansion), shorten the current threshold update cycle to 30 frames. When the confidence change is less than 0.05 (crack stable), extend the threshold update cycle to 100 frames.

[0030] The formula for calculating the regional consistency factor is as follows: ZX(x+i,y+i)>0 is denoted as K. R = K / (5 × 5); In the formula: R is the region consistency factor, K is the number of positive confidence pixels, representing the number of background regions, according to the figure, it refers to the number of crack-free background regions manually selected (usually 3-5), and the result of 5×5 is the total number of neighborhood pixels; The formula for calculating the change in confidence level is as follows: △ZX prev(x,y) =C t(x,y) -C t-1(x,y) ; In the formula: △ZX prev(x,y) The confidence change for each pixel, C t(x,y) C is the crack confidence score for the current frame. t-1(x,y) The crack confidence score for the previous frame.

[0031] If a crack is partially covered by a water film or shadow, the brightness change becomes weaker (for example, the original brightness change was 80, but after being covered by a water film, it is only 30). It will be judged that "there is no crack", and the real small crack will be missed. In step 3 of this embodiment, we not only look at "this one point", but also look at "the surroundings and the past", whether the surrounding pixels are continuous - the regional consistency factor and the trend of the previous image - the change in confidence. Specifically, cracks are not isolated; the surrounding pixels should also have similar brightness changes. For example, if the brightness change of a certain pixel is weak (score -5), but 80% of the surrounding 5×5 pixels are changing (high regional consistency), it will be considered that "this area may be a crack, let's keep it for a while," instead of directly excluding it. In addition, if the brightness change of this area has been increasing in the previous image (for example, the score was 0.3 in the previous frame, and although it was reduced to 0.2 in this frame due to the shadow, the trend is upward), it will be "comprehensively judged" and will not miss it because of a temporary shadow. In this way, step 3 of this embodiment can find these small cracks through the dual verification of "surrounding continuity + historical trend" and make the monitoring more comprehensive.

[0032] At this point, the missed areas caused by sudden changes in illumination or local low gradients are corrected, and the final crack area and comprehensive detection score are updated.

[0033] Step 4: Calculate the crack propagation rate by combining the comprehensive detection score changes of the final crack region at time t-1 and time t in the time-series image sequence. The overall detection score is recorded as the overall detection score at time t; The difference between the comprehensive detection score at time t and the comprehensive detection score at time t-1 is recorded as the comprehensive detection change score. The time interval between the acquisition of the two frames of images at time t and time t-1, which represent the combined detection scores, is denoted as the time interval. The result of dividing the comprehensive detection change score by the time interval is recorded as the crack propagation rate; When the crack propagation rate is greater than or equal to 0, it reflects the crack propagation in the final crack region. When the crack propagation rate is less than 0, it reflects the weakening of the crack in the final crack region.

[0034] The formula for calculating the crack propagation rate is as follows: ; In the formula: V is the crack propagation rate, D t Let D be the comprehensive detection score at time t. t-1 Δt represents the comprehensive detection score at time t-1, where Δt is the time interval between the acquisition of two image frames.

[0035] Existing methods use fixed "judgment criteria" (e.g., a brightness change exceeding 40 is considered a crack). However, when a crack first begins to expand, debris and vibrations occur around it, and the brightness change of these debris may also exceed 40, leading to "mistaking debris for crack expansion". After stabilization, the brightness change of the crack weakens, and it may be "not visible that it is slowly growing". Step 4 of this embodiment allows the "judgment criteria" (that is, the "background noise threshold" mentioned earlier) to adjust itself. That is, the standard is stricter when the crack expands quickly and looser when the crack is stable. Specifically, when a crack expands rapidly (e.g., at 0.1 mm per second under water pressure), there will be a lot of debris and vibration noise around it. The brightness changes of these noises may also be similar to those of the crack. In this case, raising the judgment standard (e.g., from 40 to 114) can filter out these "false signals" and focus only on the real crack body. However, after the crack stabilizes (e.g., it only grows 0.01 mm in several hours), the brightness change is very weak, but it may be slowly expanding. In this case, lowering the judgment standard (e.g., from 40 to 20) can capture these "weak signals" and prevent them from being missed due to overly strict standards.

[0036] At this point, the updated background noise of the final crack region is obtained.

[0037] Step 5: Obtain the background noise threshold from the previous update cycle, and combine it with the crack propagation rate and the current threshold update cycle to obtain the updated background noise. The method for updating the background noise threshold based on the two scenarios of crack propagation and crack weakening in the final crack region is as follows: Obtain the background noise threshold from the previous update cycle; The product of the crack propagation rate and the current threshold update cycle is recorded as the threshold adjustment magnitude. The adjusted magnitude is weighted and then added to the background noise threshold of the previous update cycle to obtain the updated background noise. If the final crack area is a crack extension, the background noise will increase after the update; If the final crack area is a crack weakening, the background noise will be reduced after the update; After the update, the background noise is directly written into the system register and overwrites the background noise threshold of the previous update cycle. At the same time, it is synchronized to the monitoring terminal for record-keeping via the wireless transmission module.

[0038] The updated background noise is calculated using the following formula: ; In the formula: BZ0 new To update the background noise, BZO prev The background noise threshold for the previous update cycle is q, where q is the weighting coefficient and Δt is the background noise threshold for the previous update cycle. update This is the current threshold update cycle.

[0039] It is worth noting that, based on the real-time state of crack propagation / weakening (determined by crack propagation rate), and using the background noise threshold of the previous cycle as a benchmark, the adjustment range is calculated by combining the crack propagation rate and the current update cycle (e.g., the faster the propagation rate and the shorter the update cycle, the larger the adjustment range), and a weighted coefficient is used to balance sensitivity and stability. When a crack expands rapidly (e.g., the crack expansion rate is >0.1mm / h under water pressure), the threshold is automatically increased (e.g., from the initial 40 grayscale values / pixels to 80) to filter out debris and vibration noise generated during the expansion process (the grayscale gradient of such noise is usually <60). When the crack enters the stable period (crack propagation rate < 0.01 mm / h), the threshold is automatically reduced (e.g., reduced to 25 gray values / pixel) to ensure that weak feature cracks (e.g., low gradient cracks under surface water film coverage, gray gradient = 30) are not misjudged as background.

[0040] The updated threshold is written to the system register in real time, overwriting the old value and synchronizing it to the monitoring terminal. This ensures that the data of the local edge computing and the remote cloud platform are consistent, avoiding deviations in monitoring results caused by threshold asynchrony. At the same time, the threshold adjustment record is automatically archived, providing data traceability for subsequent crack development trend analysis (such as inferring the active and stable periods of cracks through the threshold change curve), reducing the workload of manual threshold verification and improving system operation and maintenance efficiency.

[0041] Thus, adaptive optimization of crack monitoring is achieved through a dynamic threshold adjustment mechanism.

[0042] Based on the same inventive concept as the above methods, this embodiment of the invention also provides a hydraulic engineering lining crack monitoring system, including an optical imaging module, a feature extraction module, a region analysis module, a dynamic tracking module, and a result output module. When the processor executes the program of the industrial control computer, it implements the steps of any one of the above-mentioned hydraulic engineering lining crack monitoring methods.

[0043] It should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this invention should also be within the scope of protection of this invention.

Claims

1. A method of monitoring cracks in a lining of a water engineering structure, characterised in that, The method comprises the following steps: The method comprises the following steps: The method comprises the following steps: Based on the pre-processed image, the potential crack edge is extracted by the Canny edge detection algorithm, the broken edge is connected by the morphological dilation-erosion operation, the crack candidate region is generated, and the non-crack region is marked; The average gray gradient and the normalized crack length of the crack candidate region are calculated to distinguish the crack edge from the smooth background, the background noise threshold is calculated, the non-crack interference with a low gray value is filtered, and the region consistency factor is introduced to verify whether the region is a continuous crack by the gray difference of adjacent pixels; The crack confidence score is obtained by joint screening of the average gray gradient, the normalized crack length and the background noise threshold, and the crack preliminary selection region and the non-crack region are judged from the crack candidate region according to the crack confidence score; The SIFT feature point matching is performed on the crack preliminary selection region at t-1 and t in the time sequence image sequence, the confidence change quantity is calculated, the missed region caused by light mutation or local low gradient is corrected according to the crack confidence score, the confidence change quantity and the region consistency factor, the final crack region is updated, and the comprehensive detection score is obtained; The crack propagation rate is calculated by combining the comprehensive detection score change of the final crack region at t-1 and t in the time sequence image sequence, the background noise threshold of the last update cycle is obtained, and the updated background noise is obtained by combining the crack propagation speed and the current threshold update cycle. The updated background noise is directly written into the system register and covers the background noise threshold of the last update cycle, and is synchronized to the monitoring terminal for archiving through the wireless transmission module.

2. The water conservancy project lining crack monitoring method according to claim 1, characterized in that: The method for determining the average gray gradient and the background noise threshold is respectively: The optical monitoring image is converted into a gray image, a 3*3 Sobel operator is used to traverse all pixel points in the crack candidate region, the gradient values in the horizontal direction and the vertical direction are calculated respectively, the square of the gradient value of each pixel point is taken and then the square root is taken to obtain the gradient amplitude, the gradient amplitudes of all pixel points in the crack candidate region are counted, and the arithmetic mean value is calculated as the average gray gradient of the crack candidate region; The gradient amplitude weight of the crack edge region is higher than that of the internal region, and the weight coefficient decreases linearly with the increase of the distance of the pixel point to the edge; A pure background region in the monitoring image is selected, the gray value sequence of the region is extracted, the mean value and the standard deviation of the gray value are calculated after removing high-frequency noise by Gaussian filtering, the gray sequence is binarized by using the maximum inter-class variance method, when the pixel ratio of the background region is more than 99.5%, the mean value plus 2.5 times the standard deviation is taken as the initial noise threshold, the gray change of the background region is monitored in real time by using a sliding window, the mean value and the standard deviation are recalculated every 100 frames of images, and the background noise threshold is dynamically updated to adapt to light fluctuations.

3. The hydraulic engineering lining crack monitoring method according to claim 1, characterized in that: Based on the joint screening of the average gray gradient, the normalized crack length and the background noise threshold, the crack confidence score is obtained, and the specific method for judging the crack preliminary selection area and the non-crack area from the crack candidate area is: The crack length of the crack candidate area is measured by a contour extraction algorithm, and the maximum length of the crack in the pre-processed image is extracted; The ratio of the crack length to the maximum crack length is recorded as the normalized crack length; The product of the average gray gradient and the normalized crack length is recorded as the spatial comprehensive score of the crack; The result of subtracting the background noise threshold from the spatial comprehensive score is recorded as the crack confidence score; If the crack confidence score is greater than 0, it is determined to be a crack preliminary selection area; If the crack confidence score is less than and equal to 0, it is determined to be a non-crack area.

4. The hydraulic engineering lining crack monitoring method according to claim 1, characterized in that: The specific methods for determining the area consistency factor and the confidence change amount are respectively: For each crack preliminary selection area pixel (x, y), a 5x5 pixel square neighborhood is defined with it as the center, covering 25 pixel points. If the crack preliminary selection area pixel (x, y) is located at the edge of the image, the neighborhood exceeds the image range, at this time, the pixels exceeding the boundary are processed as background pixels, the number of positive confidence pixels in the neighborhood is counted, and the ratio of the number of positive confidence pixels in the neighborhood to the total number of pixels in the neighborhood is recorded as the area consistency factor; If the area consistency factor is greater than and equal to 0.6, it is determined to be a final crack area; If the area consistency factor is less than 0.6, it is determined to be a non-crack area; The SIFT feature point matching algorithm is used to align the crack preliminary selection area of the current frame with the corresponding area of the previous frame, the area overlap rate, the edge coincidence degree and the gray distribution similarity are calculated, and the weighted sum is obtained to get the crack confidence score of the current frame and the crack confidence score of the previous frame. The absolute value of the difference between the crack confidence score of the current frame and the crack confidence score of the previous frame is the confidence change amount.

5. The hydraulic structure lining crack monitoring method according to claim 1, characterized by: Based on the crack confidence score, the confidence change amount and the area consistency factor, the missed detection area caused by sudden change of light or local low gradient is corrected, updated to the final crack area, and the comprehensive detection score is obtained. The method is: The product of the crack confidence score of the final crack area and the area consistency factor is recorded as the spatial feature score of the current frame; The addition result of the spatial feature score and the confidence change amount is recorded as the comprehensive detection score.

6. The hydraulic structure lining crack monitoring method according to claim 1, characterized by: Based on the change of the comprehensive detection score of the final crack area at t-1 time and t time in the time sequence image sequence, the crack propagation rate is calculated. The method is: The comprehensive detection score is recorded as the comprehensive detection score at t time; The subtraction result of the comprehensive detection score at t time and the comprehensive detection score at t-1 time is recorded as the comprehensive detection change score; The time interval between the time of collecting the two frames of images of the comprehensive detection score at t time and t-1 time is recorded as the time interval. The result of dividing the comprehensive detection change score by the time interval is denoted as a crack propagation rate; When the crack propagation rate is greater than and equal to 0, it reflects crack propagation of a final crack region; When the crack propagation rate is less than 0, it reflects crack attenuation of a final crack region.

7. The method of claim 6, wherein: Based on the two cases of crack propagation and crack attenuation of the final crack region, a method for updating a background noise threshold value is: Obtaining a background noise threshold value of a previous update period; A product of the crack propagation rate and the current threshold update period is denoted as an adjustment amplitude of threshold adjustment; After weighting the adjustment amplitude and adding it to the background noise threshold value of the previous update period, an updated background noise is obtained; If the final crack region is crack propagation, the updated background noise is increased; If the final crack region is crack attenuation, the updated background noise is decreased.

8. A hydraulic lining crack monitoring system for performing the method of any one of claims 1-7, characterized in that, Comprise: An optical imaging module: a high-definition industrial camera and a ring-shaped light supplement lamp are used to collect a 1920x1080 resolution gray-scale image of a lining surface; A feature extraction module: an embedded FPGA chip is used to calculate an average gray-scale gradient in real time through a Sobel operator and output a background noise threshold value in combination with an Otsu algorithm; A region analysis module: an industrial control computer is used to perform 8-neighbor traversal and contour rule degree analysis to output a region consistency factor; A dynamic tracking module: a GPU acceleration unit of a processor carrying a SIFT matching algorithm is used to calculate a confidence change amount and dynamically adjust a current threshold update period to output a crack propagation rate and an updated background noise; A result output module: an LCD touch screen and a 4G transmission unit are used to display a crack position, length, and expansion trend in real time, write the updated background noise directly into a system register to overwrite a background noise threshold value of a previous update period, and upload the updated background noise to a cloud monitoring platform.

Images