A Deep Learning-Based Real-Time Monitoring Method for Dynamic Crack Propagation in Concrete

Abstract

This invention discloses a real-time monitoring method for the dynamic propagation of concrete cracks based on deep learning. The method includes: improving the YOLOv7 model and training and validating the improved model to obtain a crack target detection model; acquiring images of the concrete to be monitored in real time and inputting the acquired images into the crack target detection model to obtain the crack target detection result corresponding to the current image; determining whether the crack has propagated at the current moment based on a preset crack propagation judgment benchmark and the crack target detection results corresponding to two adjacent frames; and outputting the dynamic propagation change value of the crack parameters when it is determined that the crack has propagated at the current moment. This invention can identify newly generated crack damage in concrete structures in real time and monitor the dynamic propagation of existing cracks in real time, thereby achieving more efficient and accurate monitoring of dynamic crack propagation.

Description

Technical Field

[0001] This invention relates to the field of structural health monitoring technology, and in particular to a real-time monitoring method for the dynamic propagation of concrete cracks based on deep learning. Background Technology

[0002] Concrete crack monitoring is a crucial part of structural health monitoring. Real-time monitoring of the early and later stages of concrete cracking is of great significance. It can not only promptly alert professionals to maintenance and reinforcement in the early stages of crack propagation, significantly reducing repair costs, but also lower the probability of accidents.

[0003] Currently, researchers have proposed traditional contact-based crack monitoring methods and non-contact monitoring methods based on deep learning and image processing techniques. Unfortunately, contact-based methods not only require prior estimation of crack locations, but also involve expensive and easily damaged monitoring equipment. Non-contact monitoring methods, on the other hand, first utilize deep learning or image processing techniques to identify all crack images, then use image processing techniques to convert the identified crack binarized masks into crack parameters or skeleton lines. Finally, the crack parameters or skeleton lines of the current image are compared with those of the previous frame to determine if the crack has expanded. However, when only the crack width has expanded while the crack length (shape) has not, methods based on crack skeleton line shape context matching will fail. It is worth noting that due to the characteristics of image processing algorithms, these methods are time-consuming in the parameter conversion process, making real-time monitoring impossible.

[0004] Furthermore, all images, including those without crack propagation, require the same processing, leading to inefficiency. In other words, current research on crack monitoring based on deep learning algorithms and image processing techniques focuses more on crack identification at a specific moment, with very little research on real-time monitoring of crack dynamic propagation throughout the entire service life. This is a crucial basis for reasonably assessing the current service performance of concrete structures. Therefore, it is essential to find a real-time monitoring method for the dynamic propagation of concrete cracks based on deep learning. Summary of the Invention

[0005] This invention provides a real-time monitoring method for the dynamic propagation of concrete cracks based on deep learning, in order to solve the technical problems of low detection accuracy, long time consumption and low efficiency in the existing technology.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] On one hand, the present invention provides a real-time monitoring method for the dynamic propagation of concrete cracks based on deep learning, the method comprising:

[0008] The YOLOv7 network model was improved and optimized, and the improved and optimized YOLOv7 network model was trained and validated on a pre-set concrete crack target detection dataset to obtain a crack target detection model.

[0009] Images of the concrete to be monitored are acquired in real time, and the acquired images are input into the crack target detection model. The crack target detection model is then used to obtain the crack target detection result corresponding to the current image.

[0010] Based on the preset crack propagation judgment criteria, the crack propagation is determined at the current moment based on the crack target detection results corresponding to the two adjacent frames of images output by the crack target detection model.

[0011] When it is determined that the crack is propagating at the current moment, the dynamic propagation change value of the crack parameters is calculated and output.

[0012] Furthermore, the improvement and optimization of the YOLOv7 network model includes:

[0013] A global attention mechanism is introduced at layers 12, 63, and 76 of the YOLOv7 network model.

[0014] Furthermore, the crack propagation judgment criteria include:

[0015] If the pixel size of the crack target prediction rectangle in the current frame image is different from that in the previous frame image, it is determined that the crack has expanded at the current moment.

[0016] If the difference in pixel size area between the crack target prediction rectangle of the current frame image and the crack target prediction rectangle of the previous frame image is within a preset range, then the similarity between the images within the prediction rectangles of the current frame image and the previous frame image is further calculated to determine whether the crack has expanded at the current moment.

[0017] Further, the step of determining whether the crack has expanded at the current moment by calculating the similarity between the images within the rectangular prediction boxes of the current frame image and the previous frame image includes:

[0018] Calculate the grayscale values ​​of the pixels within the rectangular prediction boxes in the current frame and the previous frame respectively; plot the histograms corresponding to the current frame and the previous frame respectively;

[0019] The grayscale values ​​of the image pixels within the rectangular prediction boxes in both frames are normalized using the following formula:

[0020]

[0021] Where dst(i,j) represents the normalized gray value; src(i,j) represents a single gray value of the original image; src(x,y) represents all gray values ​​in the original image; max represents the upper limit of the normalization range; and min represents the lower limit of the normalization range.

[0022] Based on the normalized image pixel grayscale values, the similarity between the current frame image and the image within the rectangular prediction bounding box of the previous frame image is calculated using the following formula:

[0023]

[0024] Where d(H1,H2) represents the similarity calculation value between the current frame image and the image within the rectangular prediction box of the previous frame image; H1(I) and H2(I) represent the histograms of the two frames, respectively. and These represent the mean values ​​of the histograms of the two images respectively; I represents the index of all pixels or gray levels in the image.

[0025] When the similarity calculation value is less than 1, it is determined that the crack has expanded at the current moment.

[0026] Furthermore, the calculation and output of the dynamic propagation change values ​​of the crack parameters includes:

[0027] Based on the size of the crack, the extracted crack image is segmented using an improved and optimized DeepLabv3+ segmentation model or an image segmentation algorithm based on irregular ROIs to obtain a crack binarization mask.

[0028] The generated crack binarized mask is used to calculate parameters by a preset crack parameter quantization algorithm, and the dynamic expansion and change values ​​of crack parameters are output.

[0029] Furthermore, the step of segmenting the extracted crack image based on the crack size using an improved and optimized DeepLabv3+ segmentation model or an image segmentation algorithm based on irregular ROIs includes:

[0030] Based on the crack width, the crack is divided into a wide crack portion and a narrow crack portion. An image segmentation algorithm based on irregular ROI is used to segment the narrow crack portion in the crack image, and an improved and optimized DeepLabv3+ segmentation model is used to segment the wide crack portion in the crack image.

[0031] Furthermore, the improved and optimized DeepLabv3+ segmentation model introduces an attention mechanism into the decoding part of the DeepLabv3+ model.

[0032] Furthermore, the segmentation process of the image segmentation algorithm based on irregular ROIs includes:

[0033] For the crack target prediction rectangle output by the crack target detection model, the polygonal closed box selection method is used to select the crack region within the rectangle;

[0034] Median filtering and adaptive thresholding are applied to the selected image to obtain a binary mask of the crack image.

[0035] Connectivity analysis is performed on the obtained binary mask of the crack image to obtain the number and area of ​​all connected components. Connectivity analysis refers to finding and labeling independent connected components in the image. If the number of connected components in the binary mask of the crack image exceeds one, the current binary mask contains noise. Connected components with an area smaller than a preset threshold should be deleted, and only the connected component with the largest area should be retained, which is the real crack.

[0036] Furthermore, the crack parameter quantification algorithm can be any one of the following: crack parameter calculation method based on the infinitesimal element method, crack width method based on the centerline, and local average crack width method.

[0037] Furthermore, the crack parameters include crack length, crack width, and crack angle.

[0038] In another aspect, the present invention also provides an electronic device comprising a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the above-described method.

[0039] In another aspect, the present invention also provides a computer-readable storage medium storing at least one instruction that is loaded and executed by a processor to implement the above-described method.

[0040] The beneficial effects of the technical solution provided by this invention include at least the following:

[0041] The present invention provides a real-time monitoring method for the dynamic propagation of concrete cracks, which combines a crack target detection algorithm with a benchmark method for dynamic crack propagation, thereby achieving real-time automatic monitoring of crack dynamic propagation. By employing an image segmentation algorithm based on deep learning and image processing technology, as well as a crack parameter calculation method, it is possible to achieve more efficient and higher-precision quantification of the degree of dynamic crack propagation. This enables real-time identification of newly generated crack damage in concrete structures and real-time monitoring of the dynamic propagation of existing cracks. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram of the execution flow of the real-time monitoring method for dynamic propagation of concrete cracks based on deep learning provided in an embodiment of the present invention.

[0044] Figure 2 This is a schematic diagram of the crack detection result of the previous frame image provided in an embodiment of the present invention;

[0045] Figure 3 This is a schematic diagram of the crack detection result of the current frame image provided in an embodiment of the present invention;

[0046] Figure 4 This is a schematic diagram of the crack prediction rectangle of the previous frame image provided in an embodiment of the present invention;

[0047] Figure 5 This is a schematic diagram of the crack prediction rectangle of the current frame image provided in an embodiment of the present invention;

[0048] Figure 6 This is a flowchart of an image segmentation algorithm based on irregular ROIs provided in an embodiment of the present invention. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0050] First Embodiment

[0051] This embodiment provides a real-time monitoring method for the dynamic propagation of concrete cracks based on deep learning. The method mainly consists of two parts: the first part uses an improved and optimized YOLOv7 crack target real-time detection model to determine whether crack propagation has occurred using a crack propagation benchmark method. The second part performs image segmentation on the image of the propagating crack and outputs the dynamic propagation change values ​​of various crack parameters.

[0052] This method for real-time monitoring of dynamic crack propagation in concrete can be implemented using electronic equipment, which can be a terminal or a server. Specifically, the execution flow of this method is as follows: Figure 1 As shown, it includes the following steps:

[0053] S1. Improve and optimize the YOLOv7 network model, and train and validate the improved and optimized YOLOv7 network model on the concrete crack target detection dataset to obtain the crack target detection model.

[0054] Specifically, in this embodiment, the YOLOv7 network model is improved and optimized by introducing a global attention mechanism at layers 12, 63, and 76 to preserve channel and spatial information, thereby further improving the performance of the deep neural network. Furthermore, the improved and optimized YOLOv7 model is trained and validated on a concrete crack detection dataset to maximize crack detection accuracy while ensuring inference speed meets the requirements of real-time monitoring.

[0055] Furthermore, in training the YOLOv7 network model, this embodiment divides the training into multiple stages. In each stage, the learning rate is gradually reduced as training progresses. The formula for reducing the learning rate is as follows:

[0056]

[0057] Where η is the learning rate; W old W represents the model weights in the previous iteration. new The model weights from the previous iteration are estimated by minimizing a loss function (L); V t λ is the cumulative exponential moving average of past gradients; λ is a preset hyperparameter, whose value ranges from greater than 0 to less than 1, and its commonly used values ​​are 0.5, 0.9 and 0.99. Let L be the partial derivative of the weights in the old model.

[0058] S2. Combine the crack target detection model with video software for concrete crack monitoring to acquire images of the concrete to be monitored in real time, and input the acquired images into the crack target detection model to obtain the crack target detection result corresponding to the current image, thereby realizing real-time identification of concrete cracks.

[0059] S3. During the monitoring of concrete crack development, based on the crack propagation judgment criteria, the current frame crack image output by the crack target detection model is compared with the previous frame crack image to determine whether the crack has propagated at the current moment. The crack propagation judgment criteria used in this embodiment are as follows:

[0060] When the pixel size of the predicted bounding box of the current frame crack image output by the crack target detection model is inconsistent with the pixel size of the predicted bounding box of the previous frame crack image, it can be determined that the crack has expanded at the current moment.

[0061] Since the pixel coordinates of the top-left and bottom-right corners of the predicted bounding box generated by the model are known, the result of whether the crack has expanded can be calculated using formula (1). Figure 2 and Figure 3 Taking the crack image shown as an example, the pixel coordinates of the upper left and lower right corners of the predicted bounding box of the previous crack image are (154, 101) and (181, 394), respectively, while the pixel coordinates of the upper left and lower right corners of the predicted bounding box of the current crack image are (159, 94) and (197, 422), respectively. According to formula (1), the difference in pixel area between the predicted bounding boxes of the two frames is 4553 (Δs>0), which means that the crack in the current frame image has expanded.

[0062] Δs=|(x2-x1)(y2-y1)-(x2 ′ -x1 ′ )(y2 ′ -y1 ′ )| (1)

[0063] In the formula, Δs represents the difference in pixel area between the predicted bounding boxes of the crack images in the previous and next frames, and (x1,y1) and (x2,y2) represent the top-left and bottom-right pixel coordinates of the predicted bounding box of the crack image in the current frame generated by the crack target detection model, respectively. ′ ,y1 ′ (x2) ′ ,y2 ′ ) represent the top-left and bottom-right pixel coordinates of the predicted bounding box of the previous frame crack image generated by the crack target detection model, respectively.

[0064] When the pixel size and area of ​​the predicted bounding box of the current frame crack image output by the crack target detection model are basically the same as those of the predicted bounding box of the previous frame crack image, further judgment is required. Specifically, the gray values ​​of the image pixels within the bounding box in the two frames are counted respectively, the image pixel histogram is plotted and normalized, and finally the correlation method is used to accurately judge the similarity between the two histograms. When the similarity calculation value is less than 1, it can be judged that the crack has expanded at the current time.

[0065] It should be noted that, since using the number of pixel grayscale values ​​as the statistical result has certain limitations, before comparing the similarity of the images corresponding to the two histograms, a linear normalization method is used to normalize the pixel grayscale values ​​of the predicted bounding boxes of the crack images in the two frames before and after, as shown in formula (2). Then, the correlation method is used to calculate the similarity of the images corresponding to the two normalized histograms, as shown in formulas (3) and (4). Figure 4 and Figure 5Taking the crack image shown as an example, the difference between the pixel area of ​​the predicted rectangle of the crack image in the previous frame and the crack image in the current frame is close to 0. According to the calculation of formulas (2) to (4), the image similarity calculation value within the rectangle in the two images is 0.29, indicating that the crack has expanded at the current moment.

[0066]

[0067] In the formula, dst(i,j) represents the normalized gray value, src(i,j) represents a single gray value of the original image, and src(x,y) represents all gray values ​​in the original image. max represents the upper limit of the normalization range, and min represents the lower limit of the normalization range.

[0068]

[0069]

[0070] In the formula, d(H1,H2) represents the calculated similarity value between two images, and H1(I) and H2(I) are the histograms of the two images, respectively. and ...

[0071] S4. If the crack has not expanded at the current moment, continue monitoring the next frame of video image; if it has expanded, extract the crack image of the current frame and perform the next image segmentation step.

[0072] S5. Use the improved and optimized DeepLabv3+ segmentation model or an image segmentation algorithm based on irregular ROI to segment the extracted crack image and obtain a binary mask of the crack.

[0073] It should be noted that, since the collected crack image datasets typically contain limited samples of small cracks, the improved and optimized DeepLabv3+ model, while capable of end-to-end segmentation of crack images with low processing time and high efficiency, performs well on wide crack images but performs poorly on small crack images, sometimes failing to segment them completely. In contrast, image segmentation algorithms based on irregularly shaped areas of interest (ROIs) achieve good segmentation results on all crack images with high accuracy, but are time-consuming. Therefore, in crack monitoring, the advantages of both methods are fully combined: the irregularly shaped ROI-based image segmentation algorithm is used to segment the small crack portions of the crack image, while the improved DeepLabv3+ model is used to segment the other portions, thus achieving high-precision and high-efficiency crack image segmentation. The threshold used to distinguish the width of the crack can be set based on experience. This embodiment does not impose specific limitations on this. Generally, the cracks at the tip of the crack extension are smaller, and the image segmentation algorithm based on irregular ROI can obtain high-precision segmentation results. In other parts, the cracks are relatively larger, and the improved and optimized DeepLabv3+ segmentation model can be used for segmentation to improve segmentation efficiency.

[0074] The improved and optimized DeepLabv3+ segmentation model introduces an attention mechanism into the decoding part of the DeepLabv3+ model. This mechanism further focuses on crack feature information in the optimized feature map, reducing the attention given to non-crack feature information, thereby improving crack segmentation accuracy. The implementation steps are as follows:

[0075] 1) The DeepLabv3+ model mainly consists of two parts: encoding and decoding. In the decoding part, after the high-level feature map and the shallow feature map are cascaded and fused, an attention mechanism is introduced to improve the model, further focusing on the crack feature information in the optimized feature map and reducing the attention to non-crack feature information.

[0076] 2) The improved DeepLabv3+ model was trained and validated on the concrete crack segmentation dataset to obtain a high-precision concrete crack segmentation model.

[0077] 3) Input the expanded crack image into the improved and optimized DeepLabv3+ model, perform end-to-end segmentation, and output a crack binarization mask.

[0078] Image segmentation algorithms based on irregular ROIs are image processing methods consisting of irregular ROI selection, filtering algorithms, thresholding algorithms, and connected component analysis for noise reduction. The segmentation process is as follows: Figure 6 As shown, the steps are as follows:

[0079] 1) The video of the crack image is transmitted to the target detection model, which can generate a predicted bounding box of the crack image and extract the image within the bounding box.

[0080] 2) Use the polygonal closed bounding box selection method to select the crack area within the rectangular box. This area can also be called an irregular ROI.

[0081] 3) Perform median filtering and adaptive thresholding on the selected crack image to obtain a binarized mask of the crack image. It should be noted that other combinations of filtering and thresholding algorithms can be selected as needed.

[0082] 4) Perform connected component analysis on the obtained binary mask of the crack image to obtain the number and area of ​​all connected components. Connected component analysis refers to finding and labeling independent connected components in the image. Since there is only one real crack selected by the irregular ROI in the above steps, if the number of connected components in the binary mask of the crack image exceeds one, the binary mask contains noise. Therefore, connected components with smaller areas (noise) should be deleted, and only the connected component with the largest area should be retained, which is the real crack.

[0083] S6. Calculate the parameters of the generated crack binarized mask using the crack parameter quantization method, and output the dynamic expansion and change values ​​of each crack parameter.

[0084] The crack parameter quantification methods used in this embodiment include, but are not limited to: crack parameter calculation methods based on the infinitesimal element method, crack width methods based on the centerline, and local average crack width methods. The output crack parameters mainly include: crack length, crack width, and crack angle.

[0085] In summary, this embodiment provides a real-time monitoring method for the dynamic propagation of concrete cracks based on deep learning. This method combines crack target detection algorithms with benchmark methods for dynamic crack propagation, achieving real-time automatic monitoring of crack dynamic propagation. By employing image segmentation algorithms and crack parameter calculation methods based on deep learning and image processing technologies, it is possible to achieve more efficient and higher-precision quantification of the degree of dynamic crack propagation. This enables real-time identification of newly generated crack damage in concrete structures and real-time monitoring of the dynamic propagation of existing cracks.

[0086] Second Embodiment

[0087] This embodiment provides an electronic device, which includes a processor and a memory; wherein the memory stores at least one instruction, which is loaded and executed by the processor to implement the method of the first embodiment.

[0088] The electronic device can vary considerably depending on its configuration or performance, and may include one or more processors (central processing units, CPUs) and one or more memories, wherein the memories store at least one instruction that is loaded by the processor and executed in accordance with the above method.

[0089] Third Embodiment

[0090] This embodiment provides a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the method of the first embodiment described above. The computer-readable storage medium may be a ROM, random access memory, CD-ROM, magnetic tape, floppy disk, or optical data storage device, etc. The instruction stored therein can be loaded and executed by a processor in a terminal.

[0091] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0092] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0094] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0095] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. A method for real-time monitoring of dynamic propagation of concrete cracks based on deep learning, characterized in that, The deep learning-based real-time monitoring method for dynamic propagation of concrete cracks includes: The YOLOv7 network model was improved and optimized, and the improved and optimized YOLOv7 network model was trained and validated on a pre-set concrete crack target detection dataset to obtain a crack target detection model. Images of the concrete to be monitored are acquired in real time, and the acquired images are input into the crack target detection model. The crack target detection model is then used to obtain the crack target detection result corresponding to the current image. Based on the preset crack propagation judgment criteria, the crack propagation is determined at the current moment based on the crack target detection results corresponding to the two adjacent frames of images output by the crack target detection model. When it is determined that the crack is propagating at the current moment, calculate and output the dynamic propagation change values ​​of the crack parameters; The improvements and optimizations to the YOLOv7 network model include: A global attention mechanism is introduced at layers 12, 63, and 76 of the YOLOv7 network model. The criteria for judging crack propagation include: If the pixel size of the crack target prediction rectangle in the current frame image is different from that in the previous frame image, it is determined that the crack has expanded at the current moment. If the difference in pixel size area between the crack target prediction rectangle of the current frame image and the crack target prediction rectangle of the previous frame image is within a preset range, then the similarity between the images within the rectangle prediction rectangles of the current frame image and the previous frame image is further calculated to determine whether the crack has expanded at the current moment. The calculation and output of the dynamic propagation change values ​​of the crack parameters includes: Based on the size of the crack, the extracted crack image is segmented using an improved and optimized DeepLabv3+ segmentation model or an image segmentation algorithm based on irregular ROIs to obtain a crack binarization mask. The generated crack binarization mask is used to calculate parameters by a preset crack parameter quantization algorithm, and the dynamic expansion and change values ​​of crack parameters are output. The improved and optimized DeepLabv3+ segmentation model introduces an attention mechanism into the decoding part of the DeepLabv3+ model.

2. The method for real-time monitoring of dynamic propagation of concrete cracks based on deep learning as described in claim 1, characterized in that, The step of determining whether the crack has expanded at the current moment by calculating the similarity between the images within the rectangular prediction boxes of the current frame image and the previous frame image includes: Calculate the grayscale values ​​of the pixels within the rectangular prediction boxes in the current frame and the previous frame respectively; plot the histograms corresponding to the current frame and the previous frame respectively; The grayscale values ​​of the image pixels within the rectangular prediction boxes in both frames are normalized using the following formula: Where dst(i,j) represents the normalized gray value; src(i,j) represents a single gray value of the original image; src(x,y) represents all gray values ​​in the original image; max represents the upper limit of the normalization range; and min represents the lower limit of the normalization range. Based on the normalized image pixel grayscale values, the similarity between the current frame image and the image within the rectangular prediction bounding box of the previous frame image is calculated using the following formula: Where d(H1,H2) represents the similarity calculation value between the current frame image and the image within the rectangular prediction box of the previous frame image; H1(I) and H2(I) represent the histograms of the two frames, respectively. and These represent the mean values ​​of the histograms of the two images respectively; I represents the index of all pixels or gray levels in the image. When the similarity calculation value is less than 1, it is determined that the crack has expanded at the current moment.

3. The method for real-time monitoring of dynamic propagation of concrete cracks based on deep learning as described in claim 1, characterized in that, The step of segmenting the extracted crack image based on the crack size using an improved and optimized DeepLabv3+ segmentation model or an image segmentation algorithm based on irregular ROIs includes: Based on the crack width, the crack is divided into a wide crack portion and a narrow crack portion. An image segmentation algorithm based on irregular ROI is used to segment the narrow crack portion in the crack image, and an improved and optimized DeepLabv3+ segmentation model is used to segment the wide crack portion in the crack image.

4. The method for real-time monitoring of dynamic propagation of concrete cracks based on deep learning as described in claim 1, characterized in that, The segmentation process of the image segmentation algorithm based on irregular ROIs includes: For the crack target prediction rectangle output by the crack target detection model, the polygonal closed box selection method is used to select the crack region within the rectangle; Median filtering and adaptive thresholding are applied to the selected image to obtain a binary mask of the crack image. Connectivity analysis is performed on the obtained binary mask of the crack image to obtain the number and area of ​​all connected components. Connectivity analysis refers to finding and labeling independent connected components in the image. If the number of connected components in the binary mask of the crack image exceeds one, the current binary mask contains noise. Connected components with an area smaller than a preset threshold should be deleted, and only the connected component with the largest area should be retained, which is the real crack.

5. The method for real-time monitoring of dynamic propagation of concrete cracks based on deep learning as described in claim 1, characterized in that, The crack parameter quantification algorithm can be any one of the following: crack parameter calculation method based on the infinitesimal element method, crack width method based on the centerline, and local average crack width method.

6. The method for real-time monitoring of dynamic propagation of concrete cracks based on deep learning as described in claim 1, characterized in that, The crack parameters include crack length, crack width, and crack angle.

Images