CN117593651B - Tunnel crack segmentation recognition method - Google Patents
Tunnel crack segmentation recognition method Download PDFInfo
- Publication number
- CN117593651B CN117593651B CN202410074052.7A CN202410074052A CN117593651B CN 117593651 B CN117593651 B CN 117593651B CN 202410074052 A CN202410074052 A CN 202410074052A CN 117593651 B CN117593651 B CN 117593651B
- Authority
- CN
- China
- Prior art keywords
- channel
- value
- abnormal
- image
- crack
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 34
- 230000011218 segmentation Effects 0.000 title claims abstract description 23
- 230000002159 abnormal effect Effects 0.000 claims abstract description 154
- 230000008859 change Effects 0.000 claims abstract description 83
- 238000012544 monitoring process Methods 0.000 claims abstract description 13
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000005070 sampling Methods 0.000 claims description 31
- 238000013528 artificial neural network Methods 0.000 claims description 15
- 230000002708 enhancing effect Effects 0.000 claims description 7
- 230000009286 beneficial effect Effects 0.000 description 10
- 230000014509 gene expression Effects 0.000 description 9
- 238000013527 convolutional neural network Methods 0.000 description 7
- 230000002093 peripheral effect Effects 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000007796 conventional method Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000001105 regulatory effect Effects 0.000 description 2
- 230000007704 transition Effects 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000008602 contraction Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a tunnel crack segmentation and identification method, which belongs to the technical field of image processing, and adopts a mode of subtracting an original image of a hole surface in a tunnel from a real-time monitoring image to solve the problem of influence caused by similar crack trace in the tunnel, meanwhile, the method is more convenient for finding new change of the tunnel, protruding a tunnel change position, splitting a difference image to obtain three single-channel images, carrying out enhancement processing on the single-channel images, further increasing the distance between the channel value of an abnormal region and the channel value of a normal region, enabling the abnormal region to be more obvious, improving the precision of extracting abnormal pixels, finding out pixels on the crack from the abnormal pixels, taking intersection of pixels on the crack of three single-channel enhanced images, obtaining tunnel crack, and improving the precision of crack segmentation and identification.
Description
Technical Field
The invention relates to the technical field of image processing, in particular to a tunnel crack segmentation and identification method.
Background
Tunnel cracks are common defects of tunnels, and the tunnel inner hole surface is extruded and cracked due to unbalanced temperature, expansion caused by heat and contraction caused by cold. The existing tunnel crack segmentation and identification method can realize real-time monitoring of tunnel crack conditions by collecting images of tunnels, but the existing tunnel crack segmentation and identification method usually adopts a convolutional neural network, such as a YOLO neural network, to identify crack targets so as to obtain crack positions, but the convolutional neural network comprises a large number of convolutional layers and pooling layers, so that the problems of high computational complexity and large calculation amount exist, the tunnels are formed by segmented pouring or segment splicing, more traces similar to cracks exist, the traces are easy to be misidentified, and the problem of low crack identification precision exists.
Disclosure of Invention
Aiming at the defects in the prior art, the tunnel crack segmentation and identification method provided by the invention solves the following technical problems:
1. the convolutional neural network is adopted to perform tunnel crack segmentation recognition, so that the problems of high calculation complexity and large calculation amount exist;
2. the tunnel has more traces similar to cracks, is easy to be identified by mistake, and the existing tunnel crack segmentation identification method has the problem of low crack identification precision.
In order to achieve the aim of the invention, the invention adopts the following technical scheme: a tunnel crack segmentation recognition method comprises the following steps:
s1, collecting a real-time monitoring image of a tunnel inner hole surface;
s2, subtracting pixel values of the same pixel point positions from the real-time monitoring image and the original image of the tunnel inner hole surface to obtain a difference image;
s3, splitting the difference image to obtain a single-channel image, wherein the single-channel image comprises: an R-channel image, a G-channel image, and a B-channel image;
s4, enhancing the single-channel image to obtain a single-channel enhanced image, and finding out abnormal pixel points in the single-channel enhanced image, wherein the single-channel enhanced image comprises: r-channel enhanced image, G-channel enhanced image, and B-channel enhanced image;
s5, finding out pixel points on the cracks according to the positions of abnormal pixel points on the single-channel enhanced image;
s6, taking intersection of pixel points on cracks of the R channel enhancement image, the G channel enhancement image and the B channel enhancement image to obtain tunnel cracks.
In summary, the invention has the following beneficial effects: according to the method, a mode of subtracting the original image of the hole surface in the tunnel from the original image of the hole surface in the tunnel is adopted, the influence caused by the fact that similar cracks exist in the tunnel is solved, meanwhile, the new change of the tunnel is more conveniently found, the tunnel change position is protruded, the difference image is split to obtain three single-channel images, the single-channel images are subjected to enhancement processing, the distance between the channel value of an abnormal region and the channel value of a normal region is further increased, the abnormal region is more obvious, the precision of extracting abnormal pixels is improved, the pixels on the cracks are found out from the abnormal pixels, the intersection of the pixels on the cracks of the three single-channel enhanced images is taken, the tunnel cracks are obtained, and the precision of crack segmentation and identification is improved.
According to the method, the pixel points on the corresponding cracks are found according to the distribution condition of the channel values on the difference image, the calculated amount and complexity are far smaller than those of the convolutional neural network, and the problems of high calculated complexity and large calculated amount existing in the conventional method for carrying out tunnel crack segmentation identification by adopting the convolutional neural network are solved.
In the invention, at the same position in the tunnel, the pixel value of the position is different between the real-time monitoring image and the original image of the tunnel inner hole surface due to the light brightness change possibly caused by the position imaging, but the influence of the light brightness change on the tunnel inner hole surface is integral, so that the influence caused by the trace similar to the crack in the tunnel is solved by adopting a mode of subtracting the real-time monitoring image from the original image of the tunnel inner hole surface, and a new change area in the tunnel can be bulged.
Further, the step S4 includes the following sub-steps:
s41, carrying out enhancement processing on the single-channel image to obtain a single-channel enhanced image;
s42, calculating an abnormal value of each pixel point according to a neighborhood channel value of each pixel point in the single-channel enhanced image;
s43, when the abnormal value of the pixel point is larger than the abnormal threshold value, the pixel point is an abnormal pixel point.
Further, the step S41 includes the following sub-steps:
s411, clustering the single-channel images according to the distance of each channel value to find out the area with the largest area occupation ratio, and classifying the area as a normal area;
and S412, carrying out enhancement processing on the single-channel image according to the channel value in the normal region to obtain the single-channel enhanced image.
The beneficial effects of the above further scheme are: because the invention uses the difference image, and the difference image contains the integral change brought by the light and the new change on the structure in the tunnel, the integral change area brought by the light occupies a larger area, therefore, the invention carries out clustering treatment on the single-channel image according to the distance of each channel value, namely, the similar channel value is classified as a region, and the largest occupied area is a normal region.
Further, the formula for performing enhancement processing on the single-channel image in S412 is as follows:
，
wherein t is z For the enhanced channel value, t is any channel value to be enhanced in the single-channel image, and tanh is hyperbolicTangent function, t i For the i-th channel value in the normal region, N is the number of channel values in the normal region, and i is a positive integer.
The beneficial effects of the above further scheme are: according to the method, the normal region is taken as a target, the abnormal condition of the channel value to be enhanced is reflected according to the distance between any channel value to be enhanced in the single-channel image and the average value of the channels in the normal region, and the greater the distance is, the greater the enhanced channel value is, and the more obvious the distinction from the normal region is.
Further, the step S42 includes the following sub-steps:
s421, taking each pixel point in the single-channel enhanced image as a center point;
s422, calculating a first abnormal coefficient of the center point according to the channel value of the inner layer neighborhood range of the center point, wherein the inner layer neighborhood range is 33 neighborhood regions;
s423, calculating a second abnormal coefficient of the center point according to the channel value of the outer layer neighborhood range of the center point, wherein the outer layer neighborhood range is 5Subtracting 3 in the 5 neighborhood region>3 remaining regions of the neighborhood region;
s424, adding the first abnormal coefficient and the second abnormal coefficient to obtain an abnormal value of each pixel point.
The beneficial effects of the above further scheme are: after enhancement, the method is more beneficial to extracting the abnormal pixel points, each pixel point in the single-channel enhanced image is taken as a central point, and the abnormal value of the central point is determined according to the distribution condition of the channel values in two ranges, so that whether the central point is positioned at the position of abnormal change of the channel values or not is expressed by the abnormal value.
Further, the formula for calculating the first anomaly coefficient of the center point in S422 is:
，
wherein u is 1 A first anomaly coefficient t being the center point j The j channel value is the j channel value of the inner layer neighborhood range, H is a fixed constant, and j is a positive integer;
the formula for calculating the second anomaly coefficient of the center point in S423 is:
，
wherein u is 2 A second anomaly coefficient, t, being the center point k The kth channel value for the outer neighborhood range.
The beneficial effects of the above further scheme are: the present invention selects two ranges: the inner neighborhood range and the outer neighborhood range, so that the selection precision of abnormal pixel points is improved, whether the channel value of the area fluctuates or not can be reflected through the channel value distribution condition of the neighborhood, and the fluctuation represents the possible transition point of the area between the tattoo and the normal area.
Further, the step S5 includes the following sub-steps:
s51, enhancing M of abnormal pixel points on the image according to single channelsThe method comprises the steps of obtaining the position of each abnormal pixel point in an M neighborhood range, and obtaining a transverse change vector, a longitudinal change vector and a direction change vector of the abnormal pixel point, wherein M is a positive integer;
s52, calculating suspected crack coefficients of the abnormal pixel points by adopting a three-layer BP neural network according to the transverse change vector, the longitudinal change vector and the direction change vector of the abnormal pixel points;
s53, obtaining a crack value of each abnormal pixel point according to the suspected crack coefficient of the abnormal pixel point;
and S54, when the crack value is larger than the crack threshold value, the corresponding abnormal pixel point is the pixel point on the crack.
Further, the S51 is specifically: m from abnormal pixel pointEquidistant sampling K+1 abnormal pixel points in the M neighborhood range, and subtracting the abscissa of the n+1 abnormal pixel points from the n abnormal pixel points to obtain an n transverse variation value x in the transverse variation vector b,n K and n are positive integers, and the transverse change vector is: x= { X b,n X is a transverse change vector, and the value range of n is 1~K;
subtracting the ordinate of the (n+1) th sampling abnormal pixel point from the ordinate of the (n) th sampling abnormal pixel point to obtain an nth longitudinal variation value y in the longitudinal variation vector b,n The longitudinal change vector is: y= { Y b,n -wherein Y is a longitudinal variation vector;
the n+1th sampling abnormal pixel point is subtracted from the angle of the n sampling abnormal pixel point to obtain an n-th direction change value r in the direction change vector n The direction change vector is: r= { R n Wherein R is a direction change vector, R n =e n+1 -e n ，e n+1 E is the angle of the n+1th sampling abnormal pixel point n E is the angle of the nth sampling abnormal pixel point n+1 =arctan(y n+1 /x n+1 )，x n+1 The abscissa of the n+1th sampling abnormal pixel point, y n+1 The ordinate, e, of the n+1th sampling abnormal pixel point n =arctan(y n /x n )，x n Is the abscissa, y of the nth sampling abnormal pixel point n For the ordinate of the nth sampled outlier pixel, arctan is an arctangent function.
The beneficial effects of the above further scheme are: the invention is based on M of abnormal pixel pointsThe positions of the abnormal pixel points in the M neighborhood range are obtained, so that the transverse change vector, the longitudinal change vector and the direction change vector of the abnormal pixel points are obtained, the characteristics of the abnormal pixel points are expressed through the change of coordinates in the neighborhood range, and then the three-layer BP neural network is adopted to estimate the abnormal pixel pointsThe suspected crack coefficient of the dot.
The transverse variation vector expression M in the inventionM neighborhood range abscissa variation trend, and longitudinal variation vector expression MTrend of change of ordinate of M neighborhood range, direction change vector expresses M +>And (3) estimating the suspected crack coefficient of the abnormal pixel point by combining the three factors according to the change trend of the pixel point angle in the M neighborhood range.
Further, the three-layer BP neural network in S52 includes: a first input layer, a second input layer, a third input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer;
the first input layer is used for inputting a transverse change vector, and the output end of the first input layer is connected with the input end of the first hiding layer; the second input layer is used for inputting a longitudinal change vector, and the output end of the second input layer is connected with the input end of the second hiding layer; the third input layer is used for inputting a direction change vector, and the output end of the third input layer is connected with the input end of the third hidden layer; the input end of the output layer is respectively connected with the output end of the first hidden layer, the output end of the second hidden layer and the output end of the third hidden layer, and the output end of the output layer is used as the output end of the three-layer BP neural network.
The beneficial effects of the above further scheme are: according to the invention, the three input layers and the three hidden layers respectively process one type of data, so that the complexity of the BP neural network is reduced, and the influence of each type of data on an output result is better regulated and controlled.
Further, the formula for obtaining the crack value of each abnormal pixel in S53 is as follows:
，
wherein L is an abnormal pixel pointCrack value, F o For the suspected crack coefficients of the outlier pixels for which crack values are being calculated,m +.for the outlier pixel that is calculating the crack value>M neighborhood in the first->Suspected crack coefficient of each abnormal pixel, < ->Is a positive integer, C is M +.>The number of abnormal pixel points in M neighborhood range, F th Is a threshold value, gamma is a super threshold coefficient, and gamma is equal to M +.>And the number of suspected crack coefficients in the M neighborhood range is greater than a threshold value.
The beneficial effects of the above further scheme are: according to the method, after the suspected crack coefficient of each abnormal pixel point is obtained, the difference value of the threshold value is determined according to the suspected crack coefficients of other abnormal pixel points in the peripheral range and the suspected crack coefficients of the other abnormal pixel points, the larger the difference value is, the more likely the difference value is, the pixel points on the crack are set, meanwhile, the super threshold coefficient is set, and when most of the suspected crack coefficients in the peripheral range are larger than the threshold value, the abnormal pixel points are classified as the pixel points on the crack.
Drawings
FIG. 1 is a flow chart of a tunnel crack segmentation recognition method;
fig. 2 is a schematic structural diagram of a three-layer BP neural network.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and all the inventions which make use of the inventive concept are protected by the spirit and scope of the present invention as defined and defined in the appended claims to those skilled in the art.
As shown in fig. 1, a tunnel crack segmentation recognition method includes the following steps:
s1, collecting a real-time monitoring image of a tunnel inner hole surface;
s2, subtracting pixel values of the same pixel point positions from the real-time monitoring image and the original image of the tunnel inner hole surface to obtain a difference image;
s3, splitting the difference image to obtain a single-channel image, wherein the single-channel image comprises: an R-channel image, a G-channel image, and a B-channel image;
s4, enhancing the single-channel image to obtain a single-channel enhanced image, and finding out abnormal pixel points in the single-channel enhanced image, wherein the single-channel enhanced image comprises: r-channel enhanced image, G-channel enhanced image, and B-channel enhanced image;
s5, finding out pixel points on the cracks according to the positions of abnormal pixel points on the single-channel enhanced image;
s6, taking intersection of pixel points on cracks of the R channel enhancement image, the G channel enhancement image and the B channel enhancement image to obtain tunnel cracks.
In this embodiment, the original image is a reference image when the tunnel is crack-free, and the same shooting angle and position as the real-time monitoring image are equivalent to a pre-stored crack-free image.
In step S3, the pixel value of each pixel point in the difference image is the pixel difference value of two pixel points at the same position, and therefore, each pixel point in the difference image contains R, G and B-channel components.
The step S4 comprises the following substeps:
s41, carrying out enhancement processing on the single-channel image to obtain a single-channel enhanced image;
s42, calculating an abnormal value of each pixel point according to a neighborhood channel value of each pixel point in the single-channel enhanced image;
s43, when the abnormal value of the pixel point is larger than the abnormal threshold value, the pixel point is an abnormal pixel point.
The step S41 comprises the following substeps:
s411, clustering the single-channel images according to the distance of each channel value to find out the area with the largest area occupation ratio, and classifying the area as a normal area;
in the invention, the distance in the clustering algorithm is changed into the distance of each channel value, and other realization processes are consistent with the prior art and are used for realizing the partitioning of the channel values in the similar distance range;
and S412, carrying out enhancement processing on the single-channel image according to the channel value in the normal region to obtain the single-channel enhanced image.
Because the invention uses the difference image, and the difference image contains the integral change brought by the light and the new change on the structure in the tunnel, the integral change area brought by the light occupies a larger area, therefore, the invention carries out clustering treatment on the single-channel image according to the distance of each channel value, namely, the similar channel value is classified as a region, and the largest occupied area is a normal region.
The formula for enhancing the single-channel image in S412 is as follows:
，
wherein t is z For the enhanced channel value, t is any channel value to be enhanced in the single-channel image, tanh is hyperbolic tangent function, t i For the i-th channel value in the normal region, N is the number of channel values in the normal region, and i is a positive integer.
According to the method, the normal region is taken as a target, the abnormal condition of the channel value to be enhanced is reflected according to the distance between any channel value to be enhanced in the single-channel image and the average value of the channels in the normal region, and the greater the distance is, the greater the enhanced channel value is, and the more obvious the distinction from the normal region is.
The step S42 includes the following sub-steps:
s421, taking each pixel point in the single-channel enhanced image as a center point;
s422, calculating a first abnormal coefficient of the center point according to the channel value of the inner layer neighborhood range of the center point, wherein the inner layer neighborhood range is 33 neighborhood regions;
s423, calculating a second abnormal coefficient of the center point according to the channel value of the outer layer neighborhood range of the center point, wherein the outer layer neighborhood range is 5Subtracting 3 in the 5 neighborhood region>3 remaining regions of the neighborhood region;
s424, adding the first abnormal coefficient and the second abnormal coefficient to obtain an abnormal value of each pixel point.
After enhancement, the method is more beneficial to extracting the abnormal pixel points, each pixel point in the single-channel enhanced image is taken as a central point, and the abnormal value of the central point is determined according to the distribution condition of the channel values in two ranges, so that whether the central point is positioned at the position of abnormal change of the channel values or not is expressed by the abnormal value.
The formula for calculating the first anomaly coefficient of the center point in S422 is:
，
wherein u is 1 A first anomaly coefficient t being the center point j The j channel value is the j channel value of the inner layer neighborhood range, H is a fixed constant, and j is a positive integer;
the formula for calculating the second anomaly coefficient of the center point in S423 is:
，
wherein u is 2 A second anomaly coefficient, t, being the center point k And k is a positive integer, which is the kth channel value of the outer neighborhood range.
The present invention selects two ranges: the inner layer neighborhood range and the outer layer neighborhood range, thereby improving the selection precision of abnormal pixel points, reflecting whether the channel value of the area fluctuates according to the channel value distribution condition of the neighborhood, and representing that the area is likely to be a tattoo and a normal area
Transition point.
The step S5 comprises the following substeps:
s51, enhancing M of abnormal pixel points on the image according to single channelsThe method comprises the steps of obtaining the position of each abnormal pixel point in an M neighborhood range, and obtaining a transverse change vector, a longitudinal change vector and a direction change vector of the abnormal pixel point, wherein M is a positive integer;
in this embodiment, M is a positive integer of 3 or more;
s52, calculating suspected crack coefficients of the abnormal pixel points by adopting a three-layer BP neural network according to the transverse change vector, the longitudinal change vector and the direction change vector of the abnormal pixel points;
s53, obtaining a crack value of each abnormal pixel point according to the suspected crack coefficient of the abnormal pixel point;
and S54, when the crack value is larger than the crack threshold value, the corresponding abnormal pixel point is the pixel point on the crack.
The step S51 specifically includes: m from abnormal pixel pointEquidistant sampling K+1 abnormal pixel points in the M neighborhood range, and subtracting the abscissa of the n+1 abnormal pixel points from the n abnormal pixel points to obtain an n transverse variation value x in the transverse variation vector b,n K and n are positive integers, and the transverse change vector is: x= { X b,n X is a lateral variation vector,the value range of n is 1~K;
in this embodiment, the size of K is set according to the requirement;
subtracting the ordinate of the (n+1) th sampling abnormal pixel point from the ordinate of the (n) th sampling abnormal pixel point to obtain an nth longitudinal variation value y in the longitudinal variation vector b,n The longitudinal change vector is: y= { Y b,n -wherein Y is a longitudinal variation vector;
the n+1th sampling abnormal pixel point is subtracted from the angle of the n sampling abnormal pixel point to obtain an n-th direction change value r in the direction change vector n The direction change vector is: r= { R n Wherein R is a direction change vector, R n =e n+1 -e n ，e n+1 E is the angle of the n+1th sampling abnormal pixel point n E is the angle of the nth sampling abnormal pixel point n+1 =arctan(y n+1 /x n+1 )，x n+1 The abscissa of the n+1th sampling abnormal pixel point, y n+1 The ordinate, e, of the n+1th sampling abnormal pixel point n =arctan(y n /x n )，x n Is the abscissa, y of the nth sampling abnormal pixel point n For the ordinate of the nth sampled outlier pixel, arctan is an arctangent function.
The invention is based on M of abnormal pixel pointsAnd the positions of the abnormal pixel points in the M neighborhood range are obtained, so that the transverse change vector, the longitudinal change vector and the direction change vector of the abnormal pixel points are obtained, the characteristics of the abnormal pixel points are expressed through the change of coordinates in the neighborhood range, and then the three-layer BP neural network is adopted to estimate the suspected crack coefficients of the abnormal pixel points.
The transverse variation vector expression M in the inventionM neighborhood range abscissa variation trend, and longitudinal variation vector expression MM adjacentTrend of change of domain range ordinate, direction change vector expresses M +.>And (3) estimating the suspected crack coefficient of the abnormal pixel point by combining the three factors according to the change trend of the pixel point angle in the M neighborhood range.
As shown in fig. 2, the three-layer BP neural network in S52 includes: a first input layer, a second input layer, a third input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer;
the first input layer is used for inputting a transverse change vector, and the output end of the first input layer is connected with the input end of the first hiding layer; the second input layer is used for inputting a longitudinal change vector, and the output end of the second input layer is connected with the input end of the second hiding layer; the third input layer is used for inputting a direction change vector, and the output end of the third input layer is connected with the input end of the third hidden layer; the input end of the output layer is respectively connected with the output end of the first hidden layer, the output end of the second hidden layer and the output end of the third hidden layer, and the output end of the output layer is used as the output end of the three-layer BP neural network.
According to the invention, the three input layers and the three hidden layers respectively process one type of data, so that the complexity of the BP neural network is reduced, and the influence of each type of data on an output result is better regulated and controlled.
In this embodiment, specific implementation expressions of the hidden layer and the output layer are as follows.
The expression of the first hidden layer is:
，
wherein g 1 For the output of the first hidden layer, sigmoid is an S-type activation function, x b,n Is the nth transverse variation value, w in the transverse variation vector x,n Is x b,n Weights of (2);
the expression of the second hidden layer is:
，
wherein g 2 For the output of the second hidden layer, y b,n Is the nth longitudinal variation value, w in the longitudinal variation vector y,n Is y b,n Weights of (2);
the expression of the third hidden layer is:
，
wherein g 3 R is the output of the third hidden layer n Is the nth direction change value, w in the direction change vector r,n R is n Weights of (2);
the expression of the output layer is:
，
wherein F is the suspected crack coefficient of the abnormal pixel point, tanh is the hyperbolic tangent function, w 1 G is g 1 Weights, w 2 G is g 2 Weights, w 3 G is g 3 Is a weight of (2).
The formula for obtaining the crack value of each abnormal pixel point in S53 is as follows:
，
wherein L is the crack value of the abnormal pixel point, F o For the suspected crack coefficients of the outlier pixels for which crack values are being calculated,m +.for the outlier pixel that is calculating the crack value>M neighborhood in the first->Suspected crack coefficient of each abnormal pixel, < ->Is a positive integer, C is M +.>The number of abnormal pixel points in M neighborhood range, F th Is a threshold value, gamma is a super threshold coefficient, and gamma is equal to M +.>And the number of suspected crack coefficients in the M neighborhood range is greater than a threshold value.
In this embodiment, the threshold value is set according to the requirement.
According to the method, after the suspected crack coefficient of each abnormal pixel point is obtained, the difference value of the threshold value is determined according to the suspected crack coefficients of other abnormal pixel points in the peripheral range and the suspected crack coefficients of the abnormal pixel points, the larger the difference value is, the more likely the difference value is, the pixel points on the crack are set, meanwhile, the super threshold coefficient is set, and when most of the suspected crack coefficients in the peripheral range are larger than the threshold value, the abnormal pixel points are classified as the pixel points on the crack, and the crack contour is fully extracted.
According to the method, a mode of subtracting the original image of the hole surface in the tunnel from the original image of the hole surface in the tunnel is adopted, the influence caused by the fact that similar cracks exist in the tunnel is solved, meanwhile, the new change of the tunnel is more conveniently found, the tunnel change position is protruded, the difference image is split to obtain three single-channel images, the single-channel images are subjected to enhancement processing, the distance between the channel value of an abnormal region and the channel value of a normal region is further increased, the abnormal region is more obvious, the precision of extracting abnormal pixels is improved, the pixels on the cracks are found out from the abnormal pixels, the intersection of the pixels on the cracks of the three single-channel enhanced images is taken, the tunnel cracks are obtained, and the precision of crack segmentation and identification is improved.
According to the method, the pixel points on the corresponding cracks are found according to the distribution condition of the channel values on the difference image, the calculated amount and complexity are far smaller than those of the convolutional neural network, and the problems of high calculated complexity and large calculated amount existing in the conventional method for carrying out tunnel crack segmentation identification by adopting the convolutional neural network are solved.
In the invention, at the same position in the tunnel, the pixel value of the position is different between the real-time monitoring image and the original image of the tunnel inner hole surface due to the light brightness change possibly caused by the position imaging, but the influence of the light brightness change on the tunnel inner hole surface is integral, so that the influence caused by the trace similar to the crack in the tunnel is solved by adopting a mode of subtracting the real-time monitoring image from the original image of the tunnel inner hole surface, and a new change area in the tunnel can be bulged.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (6)
1. The tunnel crack segmentation and identification method is characterized by comprising the following steps of:
s1, collecting a real-time monitoring image of a tunnel inner hole surface;
s2, subtracting pixel values of the same pixel point positions from the real-time monitoring image and the original image of the tunnel inner hole surface to obtain a difference image;
s3, splitting the difference image to obtain a single-channel image, wherein the single-channel image comprises: an R-channel image, a G-channel image, and a B-channel image;
s4, enhancing the single-channel image to obtain a single-channel enhanced image, and finding out abnormal pixel points in the single-channel enhanced image, wherein the single-channel enhanced image comprises: r-channel enhanced image, G-channel enhanced image, and B-channel enhanced image;
s5, finding out pixel points on the cracks according to the positions of abnormal pixel points on the single-channel enhanced image;
s6, taking intersection of pixel points on cracks of the R channel enhanced image, the G channel enhanced image and the B channel enhanced image to obtain tunnel cracks;
the step S5 comprises the following substeps:
s51, enhancing M of abnormal pixel points on the image according to single channelsThe method comprises the steps of obtaining the position of each abnormal pixel point in an M neighborhood range, and obtaining a transverse change vector, a longitudinal change vector and a direction change vector of the abnormal pixel point, wherein M is a positive integer;
s52, calculating suspected crack coefficients of the abnormal pixel points by adopting a three-layer BP neural network according to the transverse change vector, the longitudinal change vector and the direction change vector of the abnormal pixel points;
s53, obtaining a crack value of each abnormal pixel point according to the suspected crack coefficient of the abnormal pixel point;
s54, when the crack value is larger than the crack threshold value, the corresponding abnormal pixel point is the pixel point on the crack;
the step S51 specifically includes: m from abnormal pixel pointEquidistant sampling K+1 abnormal pixel points in the M neighborhood range, and subtracting the abscissa of the n+1 abnormal pixel points from the n abnormal pixel points to obtain an n transverse variation value x in the transverse variation vector b,n K and n are positive integers, and the transverse change vector is: x= { X b,n X is a transverse change vector, and the value range of n is 1~K;
subtracting the ordinate of the (n+1) th sampling abnormal pixel point from the ordinate of the (n) th sampling abnormal pixel point to obtain an nth longitudinal variation value y in the longitudinal variation vector b,n The longitudinal change vector is: y= { Y b,n -wherein Y is a longitudinal variation vector;
the n+1th sampling abnormal pixel point is subtracted from the angle of the n sampling abnormal pixel point to obtain an n-th direction change value r in the direction change vector n The direction change vector is: r= { R n Wherein R is a direction change vector, R n =e n+1 -e n ，e n+1 The angle of the n+1th sampling abnormal pixel point，e n E is the angle of the nth sampling abnormal pixel point n+1 =arctan(y n+1 /x n+1 )，x n+1 The abscissa of the n+1th sampling abnormal pixel point, y n+1 The ordinate, e, of the n+1th sampling abnormal pixel point n =arctan(y n /x n )，x n Is the abscissa, y of the nth sampling abnormal pixel point n An ordinate of the nth sampling abnormal pixel point, wherein arctan is an arctangent function;
the three-layer BP neural network in the S52 comprises: a first input layer, a second input layer, a third input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer;
the first input layer is used for inputting a transverse change vector, and the output end of the first input layer is connected with the input end of the first hiding layer; the second input layer is used for inputting a longitudinal change vector, and the output end of the second input layer is connected with the input end of the second hiding layer; the third input layer is used for inputting a direction change vector, and the output end of the third input layer is connected with the input end of the third hidden layer; the input end of the output layer is respectively connected with the output end of the first hidden layer, the output end of the second hidden layer and the output end of the third hidden layer, and the output end of the output layer is used as the output end of the three-layer BP neural network;
the formula for obtaining the crack value of each abnormal pixel point in S53 is as follows:
，
wherein L is the crack value of the abnormal pixel point, F o For the suspected crack coefficients of the outlier pixels for which crack values are being calculated,m +.for the outlier pixel that is calculating the crack value>M neighborhood in the first->Suspected crack coefficient of each abnormal pixel, < ->Is a positive integer, C is M +.>The number of abnormal pixel points in M neighborhood range, F th Is a threshold value, gamma is a super threshold coefficient, and gamma is equal to M +.>And the number of suspected crack coefficients in the M neighborhood range is greater than a threshold value.
2. The tunnel crack segmentation recognition method according to claim 1, wherein the S4 includes the sub-steps of:
s41, carrying out enhancement processing on the single-channel image to obtain a single-channel enhanced image;
s42, calculating an abnormal value of each pixel point according to a neighborhood channel value of each pixel point in the single-channel enhanced image;
s43, when the abnormal value of the pixel point is larger than the abnormal threshold value, the pixel point is an abnormal pixel point.
3. The tunnel crack segmentation recognition method according to claim 2, wherein the S41 includes the sub-steps of:
s411, clustering the single-channel images according to the distance of each channel value to find out the area with the largest area occupation ratio, and classifying the area as a normal area;
and S412, carrying out enhancement processing on the single-channel image according to the channel value in the normal region to obtain the single-channel enhanced image.
4. The tunnel crack segmentation recognition method according to claim 3, wherein the formula for performing the enhancement processing on the single-channel image in S412 is as follows:
，
wherein t is z For the enhanced channel value, t is any channel value to be enhanced in the single-channel image, tanh is hyperbolic tangent function, t i For the i-th channel value in the normal region, N is the number of channel values in the normal region, and i is a positive integer.
5. The tunnel crack segmentation recognition method according to claim 2, wherein the S42 includes the sub-steps of:
s421, taking each pixel point in the single-channel enhanced image as a center point;
s422, calculating a first abnormal coefficient of the center point according to the channel value of the inner layer neighborhood range of the center point, wherein the inner layer neighborhood range is 33 neighborhood regions;
s423, calculating a second abnormal coefficient of the center point according to the channel value of the outer layer neighborhood range of the center point, wherein the outer layer neighborhood range is 5Subtracting 3 in the 5 neighborhood region>3 remaining regions of the neighborhood region;
s424, adding the first abnormal coefficient and the second abnormal coefficient to obtain an abnormal value of each pixel point.
6. The tunnel crack segmentation recognition method according to claim 5, wherein the formula for calculating the first anomaly coefficient of the center point in S422 is:
，
wherein u is 1 A first anomaly coefficient t being the center point j The j channel value is the j channel value of the inner layer neighborhood range, H is a fixed constant, and j is a positive integer;
the formula for calculating the second anomaly coefficient of the center point in S423 is:
，
wherein u is 2 A second anomaly coefficient, t, being the center point k And k is a positive integer, which is the kth channel value of the outer neighborhood range.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410074052.7A CN117593651B (en) | 2024-01-18 | 2024-01-18 | Tunnel crack segmentation recognition method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410074052.7A CN117593651B (en) | 2024-01-18 | 2024-01-18 | Tunnel crack segmentation recognition method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117593651A CN117593651A (en) | 2024-02-23 |
| CN117593651B true CN117593651B (en) | 2024-04-05 |
Family
ID=89922339
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410074052.7A Active CN117593651B (en) | 2024-01-18 | 2024-01-18 | Tunnel crack segmentation recognition method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117593651B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118212592B (en) * | 2024-05-21 | 2024-07-12 | 成都航空职业技术学院 | Method for identifying abnormal behavior in passenger cabin |
| CN118470551B (en) * | 2024-07-15 | 2024-09-13 | 四川凉山水洛河电力开发有限公司 | Slope loose rock mass identification method |
| CN118706078A (en) * | 2024-08-09 | 2024-09-27 | 四川交通职业技术学院 | Remote monitoring method and system adopting internet of things (IoT) and Base Station (BS) framework aiming at tunnel |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105510964A (en) * | 2015-11-27 | 2016-04-20 | 中国石油大学(华东) | Seismic recognition method of low-order strike-slip faults in complex structural areas |
| CN106651872A (en) * | 2016-11-23 | 2017-05-10 | 北京理工大学 | Prewitt operator-based pavement crack recognition method and system |
| CN107818562A (en) * | 2017-10-23 | 2018-03-20 | 广东电网有限责任公司东莞供电局 | Online detection method for cracks of air duct insulation encapsulating layer of dry-type hollow parallel reactor |
| CN108460419A (en) * | 2018-03-07 | 2018-08-28 | 中国科学院武汉岩土力学研究所 | The fracture parameters extracting method of drilling optical imagery and the fusion of radar imagery information |
| CN110930352A (en) * | 2019-02-25 | 2020-03-27 | 研祥智能科技股份有限公司 | Object color difference defect detection method and device, computer equipment and storage medium |
| CN115661021A (en) * | 2021-07-09 | 2023-01-31 | 长鑫存储技术有限公司 | Defect detection method, device, equipment and storage medium |
| WO2023082418A1 (en) * | 2021-11-09 | 2023-05-19 | 国网江苏省电力有限公司南通供电分公司 | Power utility tunnel settlement crack identification method based on artificial intelligence technology |
| CN117173188A (en) * | 2023-11-03 | 2023-12-05 | 泸州通鑫显示科技有限公司 | Glass scar identification method |
| CN117274263A (en) * | 2023-11-22 | 2023-12-22 | 泸州通源电子科技有限公司 | Display scar defect detection method |
| CN117333825A (en) * | 2023-12-01 | 2024-01-02 | 尚古智造(山东)智能装备有限公司 | Cable bridge monitoring method based on computer vision |
| CN117372338A (en) * | 2023-09-15 | 2024-01-09 | 瑞声科技(新加坡)有限公司 | Mirror defect detection method, device, equipment and readable storage medium |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7336849B2 (en) * | 2004-05-13 | 2008-02-26 | Destiny Technology Corporation | Exposure correction method for digital images |
-
2024
- 2024-01-18 CN CN202410074052.7A patent/CN117593651B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105510964A (en) * | 2015-11-27 | 2016-04-20 | 中国石油大学(华东) | Seismic recognition method of low-order strike-slip faults in complex structural areas |
| CN106651872A (en) * | 2016-11-23 | 2017-05-10 | 北京理工大学 | Prewitt operator-based pavement crack recognition method and system |
| CN107818562A (en) * | 2017-10-23 | 2018-03-20 | 广东电网有限责任公司东莞供电局 | Online detection method for cracks of air duct insulation encapsulating layer of dry-type hollow parallel reactor |
| CN108460419A (en) * | 2018-03-07 | 2018-08-28 | 中国科学院武汉岩土力学研究所 | The fracture parameters extracting method of drilling optical imagery and the fusion of radar imagery information |
| CN110930352A (en) * | 2019-02-25 | 2020-03-27 | 研祥智能科技股份有限公司 | Object color difference defect detection method and device, computer equipment and storage medium |
| CN115661021A (en) * | 2021-07-09 | 2023-01-31 | 长鑫存储技术有限公司 | Defect detection method, device, equipment and storage medium |
| WO2023082418A1 (en) * | 2021-11-09 | 2023-05-19 | 国网江苏省电力有限公司南通供电分公司 | Power utility tunnel settlement crack identification method based on artificial intelligence technology |
| CN117372338A (en) * | 2023-09-15 | 2024-01-09 | 瑞声科技(新加坡)有限公司 | Mirror defect detection method, device, equipment and readable storage medium |
| CN117173188A (en) * | 2023-11-03 | 2023-12-05 | 泸州通鑫显示科技有限公司 | Glass scar identification method |
| CN117274263A (en) * | 2023-11-22 | 2023-12-22 | 泸州通源电子科技有限公司 | Display scar defect detection method |
| CN117333825A (en) * | 2023-12-01 | 2024-01-02 | 尚古智造(山东)智能装备有限公司 | Cable bridge monitoring method based on computer vision |
Non-Patent Citations (4)
| Title |
|---|
| An image processing-based crack detection technique for pressed panel products;Yinan Miao等;《Journal of Manufacturing Systems》;20201031;287-297 * |
| Image-Based Crack Detection Methods: A Review;Hafiz Suliman Munawar等;《Infrastructures 2021》;20210814;第6卷(第8期);1-20 * |
| 基于机器视觉的轮毂瑕疵检测系统开发;唐楚;《中国优秀硕士学位论文全文数据库_工程科技Ⅱ辑》;20220215;C035-571 * |
| 基于漏磁数据成像的管道缺陷识别方法研究;王之桢;《中国优秀硕士学位论文全文数据库_工程科技Ⅰ辑》;20220415;B019-205 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117593651A (en) | 2024-02-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN117593651B (en) | Tunnel crack segmentation recognition method | |
| CN114862862B (en) | Pump body cold shut defect identification method and system based on image processing | |
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN109214380B (en) | License plate inclination correction method | |
| CN109635814B (en) | Forest fire automatic detection method and device based on deep neural network | |
| CN111444948B (en) | Image feature extraction and matching method | |
| CN116188468B (en) | HDMI cable transmission letter sorting intelligent control system | |
| CN109255799B (en) | Target tracking method and system based on spatial adaptive correlation filter | |
| CN109741358B (en) | Superpixel segmentation method based on adaptive hypergraph learning | |
| CN115639248A (en) | System and method for detecting quality of building outer wall | |
| CN116168037A (en) | Method and system for calculating bending degree of wire crimping based on image processing | |
| CN117314940B (en) | Laser cutting part contour rapid segmentation method based on artificial intelligence | |
| CN116823820B (en) | Aluminum-carbon integral stopper rod contour bending detection method | |
| CN116740053B (en) | Management system of intelligent forging processing production line | |
| CN113160097A (en) | Infrared image quantization method based on histogram transformation | |
| CN106485264B (en) | It is divided based on gradient sequence and the curve of mapping policy describes and matching process | |
| CN113222028B (en) | Image feature point real-time matching method based on multi-scale neighborhood gradient model | |
| CN114581475A (en) | Laser stripe segmentation method based on multi-scale saliency features | |
| CN109785331B (en) | Sonar image segmentation method based on self-adaptive pixel value constraint and MRF | |
| CN112949389A (en) | Haze image target detection method based on improved target detection network | |
| CN112017108A (en) | Satellite ortho-image color relative correction method based on independent model method block adjustment | |
| CN110648341B (en) | Target boundary detection method based on scale space and subgraph | |
| CN117974657B (en) | Cable surface defect detection method based on computer vision | |
| CN111008555A (en) | Unmanned aerial vehicle image small and weak target enhancement extraction method | |
| CN116681695B (en) | Quality detection method for anti-deformation template end face |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |