A quantitative evaluation method for pipeline defect detection based on CCTV and deep learning

Abstract

The present invention discloses a quantitative evaluation method for pipeline defect detection based on CCTV and deep learning, including constructing an improved pipeline crack recognition model and making a dataset for training the pipeline crack recognition model; obtaining images from pipeline videos, inputting the obtained images into a pre-trained pipeline crack recognition model, and identifying and segmenting the images containing cracks; removing redundant pixels in the images, reducing the crack width to one pixel wide, and obtaining the center of the pipeline in the crack image by means of circle detection method O ( x c , y c ) and multiple specific pixels on the crack A i ( x i , y i ) coordinates; obtaining the length, shape and width of the pipeline crack through the CDQ quantization method; evaluating and making corresponding repair treatments according to the calculated crack length and width data based on the crack grade determination principle; this method combines an improved segmentation model and an innovative crack quantization method, can characterize the basic information of the crack, and preliminarily evaluate its severity in a drainage pipe with poor and complex environment.

Description

Technical Field

[0001] The present invention relates to a pipeline defect detection quantitative evaluation method based on CCTV and deep learning. Background Art

[0002] Urban drainage pipes are one of the most important underground infrastructures, responsible for collecting and transporting domestic sewage, industrial wastewater and urban rainwater. Due to aging, improper operation and maintenance of pipes, cracks, deformation, corrosion and other defects are becoming increasingly serious, which can easily affect flow capacity, sewage overflow, water environment pollution and urban flooding, and also lead to a poor internal environment of the drainage pipes. Failure to detect, evaluate and repair drainage pipe defects in a timely manner may cause economic losses to the city and even the safety of life and property of residents.

[0003] Closed-circuit television (CCTV) has been applied to drainage pipe inspection to detect internal defects and reduce the impact of defects. Although CCTV can show the inside of the drainage pipe, inspectors still need to spend a lot of time to find and evaluate defects from a large number of images, especially in poor environment and complex drainage pipes.

[0004] Inspectors need to evaluate defects according to the latest specifications and develop repair plans, but due to the broad definition of drain pipe specifications, the uniqueness of shooting angles, and the different experience of each inspector, the defect assessment process is somewhat subjective. Therefore, developing a method to automatically detect, quantify and evaluate drain pipe defects is both important and difficult. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a quantitative assessment method for pipeline defect detection based on CCTV and deep learning. The method combines an improved segmentation model with an innovative crack quantification method, so that it can characterize basic information such as the length, width and morphology of the cracks, and preliminarily assess their severity in drainage pipes with poor and complex environments.

[0006] To achieve the above object, the present invention provides a pipeline defect detection quantitative evaluation method based on CCTV and deep learning, comprising the steps of: constructing an improved pipeline crack recognition model and preparing a data set for training the pipeline crack recognition model;

[0007] Step 2: Acquire an image from the pipeline video, input the acquired image into a pre-trained pipeline crack recognition model, and use the pipeline crack recognition model to identify and segment an image containing cracks from the current image;

[0008] Step 3: Remove redundant pixels in the image, reduce the crack width to one pixel, and use the circle detection calculation method to obtain the coordinate position of the center of the pipeline in the crack image, that is, O ( xc , y c ) and multiple specific pixels on the crack A i ( x i , y i );

[0009] Step 4: Obtain the length, shape and width of the pipeline crack by using the quantification method of crack length, shape and width;

[0010] Step 5: Evaluate the crack grade based on the calculated crack length and width data according to the crack grade determination principles and repair the cracks according to the corresponding grade.

[0011] Furthermore, the pipeline crack recognition model includes an optimized backbone network, an optimized ASPP structure and an optimized decoder.

[0012] The optimized backbone network adopts the MobileNet V2 structure and downsamples the original image three times. The feature information is transmitted once before the last two samplings. The feature information transmitted for the first time is named low-latitude feature information, and the feature information transmitted for the second time is mid-latitude feature information.

[0013] The optimized ASPP structure performs a feature information transmission, which is named as high-dimensional feature information. The optimized ASPP structure includes 4 convolution blocks with different expansion rates, 1 global average pooling module and a depth-separable convolution module.

[0014] The optimized decoder receives low-dimensional feature information, mid-latitude feature information and high-dimensional feature information and simultaneously adopts the depthwise separable convolution module in the optimized decoder to fuse the low-dimensional feature information, mid-latitude feature information and high-dimensional feature information in the splicing layer and restore the fused image to form a segmented image of the same size as the input image.

[0015] Furthermore, the specific steps of the crack morphology quantification method in step 4 include the following:

[0016] ①. Expand the pipeline into a rectangle and make the bottom side H-H' of the pipeline the length of the bottom side of the rectangle. Scan along the length of the pipeline to obtain multiple pixels. A i ( x i , y i ),in A 1( x 1, y 1) is the first pixel detected on the crack. A 1(x 1, y 1) With O ( x c , y c ) The length of the pipe bottom to the point is calculated by length scaling conversion A 1, that is, through (Rs is the actual radius of the pipe section) calculated;

[0017] in: , is the angle rad A1 The degree expression of ;

[0018] ,for A 1 -O and HO The angle between

[0019] ②, because when reconstructing the crack shape, the pixels A 1 is regarded as the reference point, in the actual rectangular domain A X coordinate of 1 X A1 = 0, that is, pixels in the actual rectangular domain A 1 coordinate system is (0, Y A1 );

[0020] ③. Calculate the Y coordinates of all pixels on the crack using the formula in step ①. Y Ai ;

[0021] Then calculate the X coordinates of all pixels in the crack in the actual rectangular domain X Ai , which is calculated by the following formula:

[0022] , the horizontal distance between adjacent pixels corresponds to the actual length of a single pixel;

[0023] ④. Obtain all pixels on the crack through step ③ A i The coordinates in the actual rectangular domain are ( X Ai , Y Ai ); the shape of the crack is drawn in the actual rectangular domain according to the obtained coordinates, and then the crack on the pipe wall is transferred from the curved surface to the plane for shape reconstruction.

[0024] Furthermore, the specific steps of the crack width quantification method in step 4 are as follows: randomly draw an inscribed circle inside the crack, and among multiple inscribed circles, obtain the diameter of the largest inscribed circle as the crack width w.

[0025] Furthermore, in step 4, the specific steps of the crack length quantification method are as follows:

[0026] ① Based on the center coordinates of the pipe section O ( x c , y c ) and multiple specific pixels on the crack A i ( x i , y i ), and obtain the actual length of each pixel on the crack. The calculation formula for the actual length of each pixel is:

[0027] in λ i is the actual length of each pixel, C i For point O Centered on OA i is the circumference of the new circle with radius, C s is the actual pipe cross-section perimeter, R s is the actual radius of the pipe section;

[0028] ② After calculating the actual length of each pixel, the total length of the entire crack is calculated by summing up. The summation formula is:

[0029] in L i equal λ i , which is the actual length of each pixel, α is the correction factor. When the resolution is 1080P or 480P, α is 1.2 or 1.8.

[0030] Furthermore, in the verification process of crack length and width quantification, the mean square error (MSE) is introduced, and its calculation formula is: ;

[0031] in, N is the total number of all morphological cracks of the same length, observed i For the actual value of each crack, predicted i The predicted value for each crack.

[0032] Furthermore, the crack grade determination principle in step 5 is specifically as follows: it includes four grades: a first-grade crack, with a length value of 0-200 mm and / or a width value of 0-0.2 mm, and a first-grade crack does not need to be repaired;

[0033] Secondary cracks, with a length of 200-300 mm and / or a width of 0.2-2.0 mm, are repaired by non-excavation methods;

[0034] Grade 3 cracks, with a length value of 300-400 mm and / or a width value of 2.0-5.0 mm, are repaired by non-excavation methods or traditional excavation methods;

[0035] Grade 4 cracks: length value > 400mm and / or width value > 5.0mm. Grade 4 cracks are serious and the pipe needs to be replaced;

[0036] When the length value and the width value are evaluated to different levels, the higher level of the two is regarded as the final level of the evaluation.

[0037] An electronic device comprises: a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute the computer executable instructions, which, when executed by the processor, implement the steps of the quantitative evaluation method for pipeline defect detection based on CCTV and deep learning.

[0038] A computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the steps of a pipeline defect detection quantitative evaluation method based on CCTV and deep learning.

[0039] Beneficial effects: 1. This application improves the pipeline crack recognition model, including the optimized backbone network, ASPP structure and decoder, which can extract more detailed features, including the overall morphology of defects such as pipeline cracks, so that the model can learn more feature information from more scales and improve the accuracy of the segmentation model.

[0040] 2. This application also provides information about the quantification of crack length, shape and width and combines it with a pipeline crack identification model to improve the detection and evaluation of cracks, so that it can characterize basic information such as the length, width and shape of the cracks, and preliminarily evaluate their severity in drainage pipes with poor and complex environments. This not only improves the efficiency and accuracy of crack detection, but also enables large-scale applications and improves the accuracy of pipeline detection and evaluation quantification. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1A schematic diagram of the steps of the quantitative evaluation method for pipeline defect detection based on CCTV and deep learning;

[0042] Figure 2 This is the optimized DeepLab V3 Plus Pro network structure diagram;

[0043] Figure 3 This is the structural diagram of the optimized backbone network;

[0044] Figure 4 It is the structural diagram of the optimized ASPP structure;

[0045] Figure 5 This is the structure diagram of the optimized decoder;

[0046] Figure 6 Obtain a schematic diagram for a single width crack;

[0047] Figure 7 Quantify the principle diagram of crack morphology;

[0048] Figure 8 Schematic diagram for crack width quantification;

[0049] Fig. 9 It is a partial schematic diagram of the experimental data set;

[0050] Fig.10 This is a partial schematic diagram of the actual data set;

[0051] Fig.11 Schematic diagram of crack identification case of experimental data set;

[0052] Fig.12 This is a data graph of CDQ accuracy results of experimental datasets combined with different segmentation models;

[0053] Fig.13 Schematic diagram of crack quantification case of experimental data set;

[0054] Fig.14 Schematic diagram of crack identification case of actual data set;

[0055] Fig.15 This is a graph showing the CDQ accuracy results of different segmentation models combined with actual data sets;

[0056] Fig.16 Schematic diagram of crack quantification case for actual data set. DETAILED DESCRIPTION

[0057] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation mode, structure, characteristics and effects of the present invention are described in detail below in combination with the accompanying drawings and preferred embodiments.

[0058] Refer to Figure 1-Figure 16 As shown, the present application provides a quantitative evaluation method for pipeline defect detection based on CCTV and deep learning, and a data set is prepared for training the pipeline crack recognition model. The data set includes two parts: an experimental data set and an actual data set, as follows:

[0059] Experimental Dataset

[0060] Image created by simulating cracks in the lab with black tape, e.g. Fig. 9 As shown. In the training dataset, photos of cracks of different widths, lengths, and shapes were taken at random shooting angles. In the validation dataset, we created cracks of two different widths (2mm and 8mm) and two different lengths (80mm and 150mm) to simulate level 2 and level 4 cracks. In the crack form, five different forms of cracks were artificially created, including: vertical straight line, vertical undulation, inclined, horizontal straight line, and horizontal undulation cracks. A total of 45 pictures were used for training; 30 pictures were used for validation. In order to improve the accuracy of the segmentation model and avoid data leakage, the training dataset was enhanced to 270 pictures through the geometric transformation method, and processed through operations such as rotation, width translation, height translation, shearing, and scaling. Since the internal environment of the pipeline is clear and the image resolution is high, 300 pictures are sufficient to train the segmentation model.

[0061] Actual Dataset

[0062] Contains 214 real CCTV inspection images with cracks, which are from the Sewer-ML dataset and pipeline inspection projects in a certain region of China, such as Fig.10 As shown in Figure 2. Real datasets are used to verify the performance of the proposed CDQ calculation method in an environment more complex than laboratory conditions. In these datasets, 139 images (109 from China; 30 from Sewer-ML) are used for training, while 75 (55 from China; 20 from Sewer-ML) are used for validation. Similarly, the training dataset is also enhanced to 770 images through the geometric transformation method.

[0063] like Figure 2-Figure 5 As shown, an improved pipeline crack recognition model is constructed, named DeepLab V3Plus Pro model, and the recognition model is trained with the above data set. The pipeline crack recognition model includes an optimized backbone network, an optimized ASPP structure and an optimized decoder;

[0064] The optimized backbone network adopts the MobileNet V2 structure and downsamples the original image three times. The feature information will be transmitted once before the last two samplings. The feature information transmitted for the first time is named low-latitude feature information, and the feature information transmitted for the second time is mid-latitude feature information. The specific MobileNet V2 structure consists of 4 module group 1 modules and 3 module group 2 modules alternating. Module group 1 has a residual module that can extract features from the image stably and efficiently, while a single module group 2 can implement a single downsampling process of features and achieve a more detailed feature extraction process. The low-dimensional feature information and the mid-dimensional feature information respectively contain the overall morphology and more detailed features of defects such as pipeline cracks, so that the model can learn more feature information from more scales and improve the accuracy of the segmentation model.

[0065] The optimized ASPP structure transmits feature information once, named high-dimensional feature information; the optimized ASPP structure includes 4 convolution blocks with different expansion rates, 1 global average pooling module and a depth-separable convolution module; the convolution modules with different expansion rates can provide more feature information at different scales, while the global average pooling module can provide the extraction function of global scale features. Since the backbone network transmits a large number of feature parameters, the depth-separable convolution module in ASPP has the effect of simplifying parameters without reducing the accuracy of the model, realizing an efficient calculation process. High-dimensional feature information, which contains very detailed defect edge features, is used for the model's refined learning process of defects.

[0066] The optimized decoder receives low-dimensional feature information, mid-latitude feature information and high-dimensional feature information and simultaneously adopts the depthwise separable convolution module in the optimized decoder to fuse the low-dimensional feature information, mid-latitude feature information and high-dimensional feature information in the splicing layer and restore the fused image to form a segmented image of the same size as the input image.

[0067] The model also uses the cross entropy loss function, which performs well and is commonly used in CNN-based models. At the same time, the Warm Up method is introduced to alleviate the initial overfitting phenomenon and stabilize the deep structure of the calculation method to better refine the segmentation of the crack edge. The model is trained and validated on a Windows system equipped with an AMD EPYC 9534 processor and two NVIDIA GeForceRTX A6000 GPUs with 48G memory. The batch size and learning rate are set to 32 and 0.0001 respectively. The results show that the mIoU of the model we created is 0.97, and the trained segmentation model is obtained by training the dataset.

[0068] like Figure 6-Figure 8As shown, an image is obtained from a pipeline video, and the obtained image is input into a pre-trained pipeline crack recognition model. The pipeline crack recognition model recognizes and segments an image containing cracks from the current image, and marks the crack image. The marking method adopts a circle detection method, specifically removing pixels that do not need to be counted in the crack image. If there are at least three non-zero pixels on the left, right, bottom, and left and right sides of a pixel, or all pixels on the left and right sides of a pixel are zero, the pixel value is set to zero. Use the edge detection calculation method to mark the fitting points. Select three fitting points to create a circle ( x c , y c , r c ,). The circle will be reflected in the three-dimensional coordinate system to show that the circle we need to find is located at ( x c , y c ,) possibility is used to obtain the coordinate position of the pipeline center in the crack image, that is, O ( x c , y c ) and multiple specific pixels on the crack A i ( x i , y i ), and then the crack length, morphology and width were quantified respectively.

[0069] Specific steps for crack length quantification:

[0070] ① Based on the center coordinates of the pipe section O ( x c , y c ) and multiple specific pixels on the crack A i ( x i , y i ), and obtain the actual length of each pixel on the crack. The calculation formula for the actual length of each pixel is:

[0071] in λ i is the actual length of each pixel, C i For point O Centered on OA i is the circumference of the new circle with radius,C s is the actual pipe cross-section perimeter, R s is the actual radius of the pipe section;

[0072] ② After calculating the actual length of each pixel, the total length of the entire crack is calculated by summing up. The summation formula is:

[0073] in L i equal λ i , which is the actual length of each pixel, and α is a correction factor used to consider the error that may be caused by image resolution. When the resolution is 1080P or 480P, α is 1.2 or 1.8.

[0074] Specific steps for crack morphology quantification:

[0075] ①. Expand the pipeline into a rectangle and make the bottom side H-H' of the pipeline the length of the bottom side of the rectangle. Scan along the length of the pipeline to obtain multiple pixels. A i ( x i , y i ),in A 1( x 1, y 1) is the first pixel detected on the crack. A 1( x 1, y 1) With O ( x c , y c ) The length of the pipe bottom to the point is calculated by length scaling conversion A 1, that is, through Calculated;

[0076] in , the angle rad A1 Converted to degree form, where the angle rad A1 yes A 1 -O and H- O The angle between Calculated;

[0077] Rs is the actual radius of the pipe section;

[0078] ② When reconstructing the crack shape, pixelsA 1 is regarded as the reference point, in the actual rectangular domain A X coordinate of 1 X A1 =0, that is, pixels in the actual rectangular domain A 1 coordinate system is (0, Y A1 );

[0079] ③. Calculate the Y coordinates of all pixels on the crack using the formula in step ①. Y Ai ;

[0080] Then calculate the X coordinates of all pixels in the crack in the actual rectangular domain X Ai , which is calculated by the following formula:

[0081] , the horizontal distance between adjacent pixels corresponds to the actual length of a single pixel;

[0082] ④. Obtain all pixels on the crack through step ③ A i The coordinates in the actual rectangular domain are ( X Ai , Y Ai ); the shape of the crack is drawn in the actual rectangular domain according to the obtained coordinates, and then the crack on the pipe wall is transferred from the curved surface to the plane for shape reconstruction.

[0083] The specific steps of crack width quantification are as follows: Inscribed circles are randomly drawn inside the crack, such as Figure 8 Among these inscribed circles, the largest diameter is regarded as the crack width w.

[0084] The above method for calculating the length, shape and width of cracks is called crack detection and quantification method (CDQ).

[0085] The pipeline cracks were identified and quantified in the experimental data set, and the DeepLab V3Plus Pro crack identification model of the CDQ calculation method of this application was compared with other different crack identification models, including DeepLab V3Plus, FCN-32s and U-Net. The prediction results of other models were not accurate enough at the cracks, while the DeepLab V3 PlusPro model showed better performance, such as Fig.11 And as shown in Table 1.

[0086] In the quantification verification, in order to evaluate the accuracy of the proposed CDQ in crack quantification, we introduced the mean square error (MSE), which is calculated as follows:

[0087]

[0088] Where N is the total number of all morphological cracks of the same length, observedi and predictedi are the actual and predicted values ​​of each crack calculated by the proposed CDQ, respectively. Fig.12 As shown, the results show that the performance of CDQ is satisfactory, with an MSE of 96.8 / 1.1 for the quantification of crack length and width. As for the correction factor, we experimented with several common CCTV equipment resolutions (such as Fig.12 The results show that the CDQ calculation method performs best at resolutions of 1920×1080 and 720×480 when α is 1.2 and 1.8 respectively. Fig.13 shown.

[0089] In the experiment of using actual data sets for model training, the verification results of the four detection models on the actual data sets are as follows: Fig.14 As can be seen from Table 1, the detection accuracy mIoU of the model of the present invention is higher than that of other models, indicating that they perform better in defect segmentation.

[0090] Table 1. Accuracy of DeepLab V3 Plus Pro crack recognition model under experimental data set

[0091]

[0092] To further validate the performance of the proposed CDQ in crack size quantification, 30 real CCTV inspection images were selected from the validation dataset. Based on the quantification results in the experimental dataset, the correction factor α was set to 1.2 to account for the high resolution of these images (1920×1080 pixels). In these images, the diameter of the sewer pipe is 600 mm, based on which the size of the crack is calculated.

[0093] The results of crack size quantification are as follows: Fig.15 This shows that CDQ combined with DeepLab V3 Plus Pro performs better in crack length quantification than FCN-32s and U-Net, with 25 out of 30 cracks correctly classified (83% accuracy). Fig.16 The results show that when the CDQ calculation method is combined with DeepLab V3 Plus Pro, it performs satisfactorily in crack shape characterization.

[0094] This application proposes a new crack assessment method based on length and width quantitative data, taking into account the length and width of the crack. The length and width assessment method is shown in Table 2.

[0095] Table 2 Crack grade assessment method

[0096]

[0097] There are four levels: first-level cracks, with a length of 0-200mm and / or a width of 0-0.2mm, do not need to be repaired; second-level cracks, with a length of 200-300mm and / or a width of 0.2-2.0mm, are repaired by non-excavation methods; third-level cracks, with a length of 300-400mm and / or a width of 2.0-5.0mm, are repaired by non-excavation methods or traditional excavation methods; fourth-level cracks, with a length of >400mm and / or a width of >5.0mm, are severe and require pipe replacement;

[0098] When the length and width values ​​are assessed to different levels, the higher level of the two is considered the final level of the assessment. For a specific crack, if its length and width are assessed to different levels, the higher level is used as the final assessment result. For example, a crack with a length of 240 mm and a width of 6 mm is assessed as level 4. This assessment method takes into account both the crack length and the maximum width, providing a more comprehensive assessment of the severity of the crack.

[0099] The present application also provides an electronic device, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of a quantitative evaluation method for pipeline defect detection based on CCTV and deep learning are implemented.

[0100] A computer-readable storage medium is also provided, which stores computer-executable instructions, which, when executed by a processor, implement the steps of a quantitative evaluation method for pipeline defect detection based on CCTV and deep learning.

[0101] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A pipeline defect detection quantitative assessment method based on CCTV and deep learning, characterized by: include: Step 1: Build an improved pipeline crack recognition model and create a data set for training the pipeline crack recognition model; The pipeline crack recognition model includes an optimized backbone network, an optimized ASPP structure and an optimized decoder; The optimized backbone network adopts the MobileNet V2 structure and downsamples the original image three times. The feature information is transmitted once before the last two samplings. The feature information transmitted for the first time is named low-latitude feature information, and the feature information transmitted for the second time is mid-latitude feature information. The optimized ASPP structure performs a feature information transmission, which is named as high-dimensional feature information. The optimized ASPP structure includes 4 convolution blocks with different expansion rates, 1 global average pooling module and a depth-separable convolution module. The optimized decoder receives low-dimensional feature information, mid-latitude feature information and high-dimensional feature information and simultaneously uses the depthwise separable convolution module in the optimized decoder to fuse the low-dimensional feature information, mid-latitude feature information and high-dimensional feature information in the concatenation layer and restore the fused image to form a segmented image of the same size as the input image; Step 2: Acquire an image from the pipeline video, input the acquired image into a pre-trained pipeline crack recognition model, and use the pipeline crack recognition model to identify and segment an image containing cracks from the current image; Step 3: Remove redundant pixels in the image, reduce the crack width to one pixel, and use the circle detection method to obtain the coordinate position of the center of the pipeline in the crack image, that is, O ( x c , y c ) and multiple specific pixels on the crack A i ( x i , y i ); Step 4: Obtain the length, shape and width of the pipeline crack by using a crack length, shape and width quantification method; the specific steps of the crack shape quantification method include the following: ①. Expand the pipeline into a rectangle and make the bottom side H-H' of the pipeline the length of the bottom side of the rectangle. Scan along the length of the pipeline to obtain multiple pixels. A i ( x i , y i ),in A 1( x 1, y 1) is the first pixel detected on the crack. A 1( x 1, y 1) With O ( x c , y c ) The length of the pipe bottom to the point is calculated by length scaling conversion A 1, that is, through Calculated; in: , is the angle rad A1 The degree expression of ; ,for A 1 -O and HO The angle between Rs is the actual radius of the pipe section; ② When reconstructing the crack shape, pixels A 1 is regarded as the reference point, in the actual rectangular domain A X coordinate of 1 X A1 = 0, that is, pixels in the actual rectangular domain A 1 coordinate system is (0, Y A1 ); ③. Calculate the Y coordinates of all the pixels on the crack using the calculation formula in step ①. Y Ai ; Then calculate the X coordinates of all pixels in the crack in the actual rectangular domain X Ai , which is calculated by the following formula: , the horizontal distance between adjacent pixels corresponds to the actual length of a single pixel; ④. Obtain all pixels on the crack through step ③ A i In the actual rectangular domain, the coordinates ( X Ai , Y Ai ); according to the obtained coordinates, the shape of the crack is drawn in the actual rectangular domain, and then the crack on the pipe wall is transferred from the curved surface to the plane for shape reconstruction; Step 5: Evaluate the crack grade based on the calculated crack length and width data according to the crack grade determination principles and repair the cracks according to the corresponding grade.

2. The pipeline defect detection quantitative assessment method based on CCTV and deep learning according to claim 1 is characterized in that: The specific steps of the crack width quantification method in step 4 are as follows: randomly draw an inscribed circle inside the crack, and among multiple inscribed circles, obtain the diameter of the largest inscribed circle as the crack width w .

3. The pipeline defect detection quantitative assessment method based on CCTV and deep learning according to claim 2 is characterized in that: In step 4, the specific steps of the crack length quantification method are as follows: ① Based on the center coordinates of the pipe section O ( x c , y c ) and multiple specific pixels on the crack A i ( x i , y i ), and obtain the actual length of each pixel on the crack. The calculation formula for the actual length of each pixel is: where λ i is the actual length of each pixel, C i For point O Centered on OA i is the circumference of the new circle with radius, C s is the actual pipe cross-section perimeter, R s is the actual radius of the pipe section; ② After calculating the actual length of each pixel, the total length of the entire crack is calculated by summing up. The summation formula is: in L i equal λ i , which is the actual length of each pixel, α is the correction factor. When the resolution is 1080P or 480P, α is 1.2 or 1.

8.

4. The pipeline defect detection quantitative assessment method based on CCTV and deep learning according to claim 3 is characterized in that: In the verification process of crack length and width quantification, the mean square error (MSE) is introduced, and its calculation formula is: ; in, N is the total number of all morphological cracks of the same length, observed i For the actual value of each crack, predicted i The predicted value for each crack.

5. The pipeline defect detection quantitative assessment method based on CCTV and deep learning according to claim 4 is characterized in that: The crack grade determination principle in step 5 is as follows: there are four grades: grade one crack, with a length value of 0-200 mm and / or a width value of 0-0.2 mm, and grade one crack does not need to be repaired; Secondary cracks, with a length of 200-300 mm and / or a width of 0.2-2.0 mm, are repaired by non-excavation methods; Grade 3 cracks, with a length value of 300-400 mm and / or a width value of 2.0-5.0 mm, are repaired by non-excavation methods or traditional excavation methods; Grade 4 cracks: length value > 400mm and / or width value > 5.0mm. Grade 4 cracks are serious and the pipe needs to be replaced; When the length value and the width value are evaluated to different levels, the higher level of the two is regarded as the final level of the evaluation.

6. An electronic device comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, the steps of the quantitative assessment method for pipeline defect detection based on CCTV and deep learning as described in any one of claims 1 to 5 are implemented.

7. A computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the steps of any one of claims 1 to 5 of the method for quantitative assessment of pipeline defect detection based on CCTV and deep learning.

Images