CN117975374B - Intelligent visual monitoring method for double-skin wall automatic production line - Google Patents
Intelligent visual monitoring method for double-skin wall automatic production line Download PDFInfo
- Publication number
- CN117975374B CN117975374B CN202410369999.0A CN202410369999A CN117975374B CN 117975374 B CN117975374 B CN 117975374B CN 202410369999 A CN202410369999 A CN 202410369999A CN 117975374 B CN117975374 B CN 117975374B
- Authority
- CN
- China
- Prior art keywords
- sub
- block
- pixel point
- gradient
- target
- 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 77
- 238000004519 manufacturing process Methods 0.000 title claims abstract description 32
- 230000000007 visual effect Effects 0.000 title claims abstract description 29
- 238000012544 monitoring process Methods 0.000 title claims abstract description 24
- 239000013598 vector Substances 0.000 claims abstract description 105
- 230000005764 inhibitory process Effects 0.000 claims abstract description 23
- 238000012937 correction Methods 0.000 claims abstract description 16
- 238000001514 detection method Methods 0.000 claims abstract description 6
- 238000012216 screening Methods 0.000 claims abstract description 6
- 230000007547 defect Effects 0.000 claims description 12
- 230000000739 chaotic effect Effects 0.000 claims description 11
- 238000003708 edge detection Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 8
- 238000010606 normalization Methods 0.000 claims description 7
- 238000010276 construction Methods 0.000 claims description 2
- 238000012545 processing Methods 0.000 abstract description 3
- 238000005286 illumination Methods 0.000 description 4
- 239000006185 dispersion Substances 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000004566 building material Substances 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000005192 partition Methods 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of image data processing, in particular to an intelligent visual monitoring method for a double-skin wall automatic production line, which comprises the following steps: acquiring a double-skin wall image; constructing a square window and sub-blocks according to the double-skin wall image; obtaining a feature vector according to the sub-block; obtaining the degree of confusion according to the gradient of the pixel points in the sub-blocks; screening out the characteristic vector of the sub-block according to the characteristic vector; according to the feature vector of the sub-block, obtaining symmetrical similarity; obtaining gradient regularity according to the symmetry similarity and the confusion degree; obtaining the inhibition degree according to the gradient regularity; obtaining a gradient correction amplitude according to the inhibition degree; obtaining a crack pixel point according to the gradient correction amplitude; and (5) according to the number of the crack pixel points, obtaining a visual detection result of the double-skin wall automatic production line. According to the invention, the texture of the double-skin wall image is inhibited, so that the accuracy of visual monitoring of the double-skin wall automatic production line is improved.
Description
Technical Field
The invention relates to the technical field of image data processing, in particular to an intelligent visual monitoring method for a double-skin wall automatic production line.
Background
In the production process of building materials, intelligent visual monitoring systems are widely used, remarkable progress is made in identifying and classifying wall defects, and the production quality of the wall can be improved; in the production process of the double-skin wall, the phenomenon that the wall structure is unstable can be caused by expansion and shrinkage of concrete in the wall in the curing process, so that the crack on the surface of the wall is detected by utilizing an edge detection mode, and timely treatment measures can be taken when the crack problem occurs by the system, so that the further expansion of the defect range is avoided.
When the canny edge detection is utilized to detect defects on the surface of the wall, the traditional canny edge detection only considers the overall gradient, different texture effects can be formed on the surface of the wall due to the fact that different concrete pouring modes are possibly adopted by the wall, gradient values of texture pixel points are relatively large, when crack areas are detected, poor distinguishing between textures and cracks can be caused, proper threshold values cannot be selected to obtain accurate texture areas, and therefore monitoring results of the double-skin wall automatic production line are inaccurate.
Disclosure of Invention
The invention provides an intelligent visual monitoring method for a double-skin wall automatic production line, which aims to solve the existing problems.
The intelligent visual monitoring method for the double-skin wall automatic production line adopts the following technical scheme:
The embodiment of the invention provides an intelligent visual monitoring method for a double-skin wall automatic production line, which comprises the following steps of:
acquiring a double-skin wall image; constructing a square window corresponding to each pixel point in the double-skin wall image and sub-blocks in the square window; obtaining a feature vector of each pixel point according to the gradient of the pixel point in the sub-block;
In the double-skin wall image, marking any pixel as a target pixel; marking a square window corresponding to the target pixel point as a target window; obtaining the chaotic degree of each sub-block in the target window according to the gradient of all the pixel points in each sub-block in the target window;
Screening out the characteristic vector of each sub-block in the target window according to the characteristic vector of the target pixel point; according to the feature vector of each sub-block in the target window, obtaining the symmetrical similarity of each sub-block in the target window; obtaining the gradient regularity of the target pixel point according to the symmetry similarity and the chaotic degree of each sub-block in the target window;
Obtaining the inhibition degree of each pixel point according to the gradient regularity of the pixel points in the square window corresponding to each pixel point in the double-skin wall image;
Obtaining a gradient correction amplitude of each pixel point according to the inhibition degree of each pixel point in the double-skin wall image; based on the gradient correction amplitude value of each pixel point in the double-skin wall image, obtaining a crack pixel point by utilizing an edge detection algorithm; and (5) according to the number of the crack pixel points, obtaining a visual detection result of the double-skin wall automatic production line.
Further, the construction of the square window corresponding to each pixel point in the double-skin wall image and the sub-blocks in the square window comprises the following specific steps:
In the double-skin wall image, an N multiplied by N square window is constructed by taking any pixel point as a center, wherein N represents a preset side length; dividing the square window into a plurality of sub-blocks with the size of M multiplied by M; m represents the preset side length of the sub-block.
Further, the step of obtaining the feature vector of each pixel according to the gradient of the pixel in the sub-block includes the following specific steps:
Calculating the gradient vector of each pixel point by utilizing a Sobel operator according to the gray value of the pixel point in each sub-block; according to the gradient vector of each pixel point in each sub-block, constructing a gradient histogram of HD (high definition) bin directions of each sub-block, wherein the HD represents the number of preset bin directions; the abscissa of the gradient histogram is the bin direction; the ordinate is the accumulated value of the gradient values of all pixel points in the sub-block in the bin direction corresponding to the respective angle, and the accumulated value is recorded as a response value;
In the double-skin-wall image, the feature vector of any pixel point is as follows:
In the method, in the process of the invention, Is the characteristic vector of the pixel point; /(I)、/>/>Respectively representing response values in the 1 st, 2 nd and HD bin directions in the 1 st sub-block in the square window corresponding to the pixel point; /(I)And/>Respectively representing a response value in the 1 st bin direction and a response value in the 2 nd bin direction in the 2 nd sub-block in the square window corresponding to the pixel point; /(I)And representing the response value in the HD bin direction in the ZK sub-block in the square window corresponding to the pixel point.
Further, the obtaining the chaotic degree of each sub-block in the target window according to the gradient of all the pixel points in each sub-block in the target window comprises the following specific calculation method:
In the method, in the process of the invention, The degree of confusion for the p-th sub-block in the target window; /(I)The number of pixel points contained in the p-th sub-block in the target window is the number; /(I)Is a linear normalization function; /(I)Gradient vector of the z pixel point in the p-th sub-block in the target window; /(I)The module length of the gradient vector of the z pixel point in the p-th sub-block in the target window is given; /(I)Gradient vector of the z pixel point in the p-th sub-block in the target window; /(I)The average gradient vector of the gradient vector of all pixel points in the p-th sub-block in the target window; /(I)For/>And/>A sine value of the included angle; /(I)As a function of absolute value.
Further, the step of screening out the feature vector of each sub-block in the target window according to the feature vector of the target pixel point includes the following specific steps:
and in the characteristic vector of the target pixel point, the sequence formed by the response values in all bin directions corresponding to each sub-block in the target window is recorded as the characteristic vector of each sub-block.
Further, the step of obtaining the symmetrical similarity of each sub-block in the target window according to the feature vector of each sub-block in the target window comprises the following specific steps:
Marking any sub-block in the target window as a target sub-block; marking the sub-block of the target sub-block symmetrical to the central sub-block of the target window as a reference sub-block;
s1: subtracting the characteristic vector of the central sub-block of the target window from the characteristic vector of the target sub-block to obtain a difference vector of the target sub-block; subtracting the characteristic vector of the reference sub-block from the characteristic vector of the central sub-block of the target window to obtain a difference vector of the reference sub-block;
S2: calculating cosine similarity of the difference vector of the target sub-block and the difference vector of the reference sub-block;
s3: the positions of all response values in the feature vector of the target sub-block are moved backwards by one bit, and the last response value of the feature vector of the target sub-block is moved to the first response value position of the feature vector of the target sub-block, so that a new feature vector of the target sub-block is obtained;
S4: repeating S1 to S3 until the repetition number is equal to HD-1 times, and stopping repeating;
And (5) recording the maximum cosine similarity obtained in S1 to S4 as the symmetrical similarity of the target subblocks.
Further, the gradient regularity of the target pixel is obtained according to the symmetry similarity and the chaotic degree of each sub-block in the target window, which comprises the following specific calculation modes:
In the method, in the process of the invention, Representing the gradient regularity of the target pixel points; /(I)Representing the number of sub-blocks in the target window after the central sub-block is removed; /(I)Indicating the confusion degree of the q-th sub-block after the central sub-block is removed in the target window; /(I)Representing the degree of confusion of the center sub-block of the target window; /(I)Representing the symmetrical similarity of the qth sub-block after the center sub-block is removed from the target window.
Further, the step of obtaining the inhibition degree of each pixel point according to the gradient regularity of the pixel point in the square window corresponding to each pixel point in the double-skin wall image comprises the following specific steps:
In the method, in the process of the invention, The inhibition degree of the ith pixel point; /(I)Is a linear normalization function; /(I)In a square window corresponding to the ith pixel point, the gradient regularity is greater than the number of the pixel points of the gradient regularity of the ith pixel point; The number of the pixel points in the square window corresponding to the ith pixel point; /(I) The gradient regularity of the ith pixel point; The gradient regularity of the j-th pixel except the i-th pixel is in the square window corresponding to the i-th pixel.
Further, the step of obtaining the gradient correction amplitude of each pixel point according to the inhibition degree of each pixel point in the double-skin wall image comprises the following specific steps:
And calculating a difference value of 1 minus the inhibition degree of the ith pixel point, and recording the product of the difference value and the gradient amplitude of the ith pixel point as the gradient correction amplitude of the ith pixel point.
Further, the visual detection result of the double-skin wall automatic production line is obtained according to the number of the crack pixel points, and the method comprises the following specific steps:
Acquiring the number of the crack pixel points, and judging that the crack defect does not exist on the surface of the double-skin wall when the number of the crack pixel points is smaller than a preset crack threshold LW; and when the number of the crack pixel points is larger than or equal to a preset crack threshold LW, judging that the crack defect exists on the surface of the double-skin wall.
The technical scheme of the invention has the beneficial effects that: according to the method, the chaotic degree of each sub-block in the target window is obtained through the gradient of all pixel points in each sub-block in the target window, and the possibility of cracks in the sub-blocks is primarily judged; according to the symmetrical similarity of each sub-block in the target window, gradient regularity of the target pixel is obtained, and the possibility that each pixel is a texture pixel is judged more accurately; according to the gradient regularity of each pixel point in the double-skin wall image and the gradient regularity of the pixel points in the square window corresponding to each pixel point, the inhibition degree of each pixel point is obtained, and the possibility that each pixel point is a texture pixel point is quantized more accurately. According to the invention, the crack pixel points are obtained through the inhibition degree of each pixel point and the edge detection algorithm, so that the intelligent visual monitoring result of the double-skin wall automatic production line is more accurate.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of steps of an intelligent visual monitoring method for a double-skin wall automatic production line;
Fig. 2 is a schematic diagram of a split of a double skin wall according to the present embodiment.
Detailed Description
In order to further describe the technical means and effects adopted by the invention to achieve the preset aim, the following is a detailed description of specific implementation, structure, characteristics and effects of the intelligent visual monitoring method for the double-skin wall automatic production line according to the invention with reference to the accompanying drawings and the preferred embodiment. In the following description, different "one embodiment" or "another embodiment" means that the embodiments are not necessarily the same. Furthermore, the particular features, structures, or characteristics of one or more embodiments may be combined in any suitable manner.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
The invention provides a specific scheme of an intelligent visual monitoring method for a double-skin wall automatic production line, which is specifically described below with reference to the accompanying drawings.
Referring to fig. 1, a flowchart of steps of an intelligent visual monitoring method for a double-skin wall automatic production line according to an embodiment of the invention is shown, the method includes the following steps:
step S001: acquiring a double-skin wall image; constructing a square window corresponding to each pixel point in the double-skin wall image and sub-blocks in the square window; and obtaining the characteristic vector of each pixel point according to the gradient of the pixel point in the sub-block.
Specifically, the produced double-skin wall is horizontally placed on a conveyor belt, an industrial camera is arranged above the conveyor belt, a lens of the industrial camera is aligned to the surface of the double-skin wall on the conveyor belt, and RGB images of the surface of the double-skin wall are acquired in the working process of the conveyor belt.
And carrying out graying operation on the RGB image of the surface of the double-skin wall to obtain the double-skin wall image, as shown in figure 2.
It should be noted that, the conventional HOG features are obtained by partitioning an image and analyzing the overall gradient features of each partition in the image, but the description of a single pixel point is rough, so that the HOG features of each point in a neighborhood centered on each point are obtained by using the manner of obtaining each gradient histogram in the conventional HOG feature calculation process, so as to further analyze the properties of each center point. The basic idea of the conventional HOG feature is to divide the image into small units, calculate the direction and magnitude of the gradient in each unit, and then make statistics and normalization on the gradient information to form a global feature vector.
HOG feature calculation is a well known technique, and specific methods are not described here.
Specifically, in the double-skin wall image, an n×n square window is constructed with any pixel point as the center, N represents a preset side length, the preset side length n=9 is taken as an example for description in this embodiment, other values may be set in other embodiments, and this embodiment is not limited specifically. And in the square window, sequentially constructing M multiplied by M sub-blocks according to the sequence of M step sizes from left to right and from top to bottom, recording the number of the sub-blocks as ZK, wherein M represents the side length of a preset sub-block, namely dividing the square window into a plurality of sub-blocks with the size of M multiplied by M.
In this embodiment, the preset sub-block side length m=3 is taken as an example for description, and other values may be set in other embodiments, which is not specifically limited.
Calculating the gradient vector of each pixel point by utilizing a Sobel operator according to the gray value of the pixel point in each sub-block; according to the gradient vector of each pixel point in each sub-block, constructing a gradient histogram of HD (high definition) bin directions of each sub-block, wherein the HD represents the number of preset bin directions, and the abscissa of the gradient histogram is the bin direction; the ordinate is the accumulated value of the gradient values of all the pixel points in the sub-block added to the bin direction corresponding to the respective angles, and the accumulated value is recorded as a response value. In this embodiment, the preset bin direction number hd=12 is described as an example, and other values may be set in other embodiments, which is not specifically limited. The method for calculating the gradient vector of the pixel point and the method for calculating the gradient histogram are known techniques, and specific methods are not described herein. The bin direction is one of the well-known HOG features.
In the double-skin-wall image, the feature vector of any pixel point is as follows:
In the method, in the process of the invention, Is the characteristic vector of the pixel point; /(I)、/>/>Respectively representing response values in the 1 st, 2 nd and HD bin directions in the 1 st sub-block in the square window corresponding to the pixel point; /(I)And/>Respectively representing a response value in the 1 st bin direction and a response value in the 2 nd bin direction in the 2 nd sub-block in the square window corresponding to the pixel point; /(I)And representing the response value in the HD bin direction in the ZK sub-block in the square window corresponding to the pixel point.
According to the method, the characteristic vector of each pixel point in the double-skin wall image is obtained.
Step S002: in the double-skin wall image, marking any pixel as a target pixel; marking a square window corresponding to the target pixel point as a target window; and obtaining the confusion degree of each sub-block in the target window according to the gradients of all the pixel points in each sub-block in the target window.
It should be noted that, the grain distribution on the double-skin wall is more regular and uniform, and the distribution of the cracks is more random, so that the preliminary judgment is made on the defect area of the double-skin wall through the chaotic degree of the sub-blocks. However, since the gradient amplitude of each pixel point in the sub-block may be affected by uneven illumination, the gradient amplitude of the pixel point in the sub-block with weaker illumination is smaller, so that the response value of part of the sub-block in the bin direction is smaller, the gradient direction and the magnitude in each sub-block need to be integrated, and the regularity quantization is performed by the chaotic degree in each sub-block.
In the double-skin wall image, marking any pixel as a target pixel;
marking a square window corresponding to the target pixel point as a target window;
the calculation mode of the chaotic degree of the p-th sub-block in the target window is as follows:
In the method, in the process of the invention, The degree of confusion for the p-th sub-block in the target window; /(I)The number of pixel points contained in the p-th sub-block in the target window is the number; /(I)Is a linear normalization function; /(I)Gradient vector of the z pixel point in the p-th sub-block in the target window; /(I)The module length of the gradient vector of the z pixel point in the p-th sub-block in the target window is given; /(I)Gradient vector of the z pixel point in the p-th sub-block in the target window; /(I)The average gradient vector of the gradient vector of all pixel points in the p-th sub-block in the target window; /(I)For/>And/>A sine value of the included angle; /(I)As a function of absolute value.
In the method, in the process of the invention,The larger the value of the degree of dispersion of the gradient direction of the z pixel point in the p-th sub-block in the target window is, the more the gradient direction is dispersed, and the more the p-th sub-block in the target window is disordered; The smaller the value of the degree of correction of the degree of dispersion of the gradient direction of the z-th pixel point in the p-th sub-block in the target window, which indicates that the z-th pixel point in the p-th sub-block in the target window is more likely to be affected by weaker illumination conditions, the more the degree of dispersion of the gradient direction of the pixel point needs to be corrected.
According to the method, the confusion degree of each sub-block in the target window is obtained.
Step S003: screening out the characteristic vector of each sub-block in the target window according to the characteristic vector of the target pixel point; according to the feature vector of each sub-block in the target window, obtaining the symmetrical similarity of each sub-block in the target window; and obtaining the gradient regularity of the target pixel point according to the symmetry similarity and the confusion degree of each sub-block in the target window.
It should be noted that, after obtaining the degree of confusion of each sub-block in each square window, the regular property in each square window is not only dependent on the degree of confusion of the sub-blocks in the square window, but also because the grain trend of the concrete surface tends to appear in the continuous condition in the corresponding neighborhood of each point, the correlation between the sub-blocks at symmetrical positions in the square window needs to be analyzed, if the variation among the sub-blocks at symmetrical positions is relatively regular, the gradient of the central pixel point of the square window is greatly contributed by the grain, but not the actual concrete surface crack defect.
Specifically, in the feature vector of the target pixel, the sequence of response values in all bin directions corresponding to each sub-block in the target window is recorded as the feature vector of each sub-block. Specifically, the feature vector of the p-th sub-block is:
In the method, in the process of the invention, Feature vector for the p-th sub-block; /(I)The response value of the p-th sub-block in the target window in the 1 st bin direction is obtained; /(I)The response value of the p-th sub-block in the target window in the 2 nd bin direction is given; /(I)The response value of the p-th sub-block in the target window in the HD bin direction is obtained; HD is a preset number of bin directions.
According to the method, the characteristic vector of each sub-block in the target window is obtained.
It should be noted that, the texture may not be in a straight line state, that is, in the adjacent three sub-blocks, the texture direction may be gradually changed and performed along a certain direction, which is reflected inside the feature vector, that is, the peak point of the corresponding feature vector gradually moves, so after the difference vector of the feature vector is obtained, the response value in the difference vector is moved, the cosine similarity of two difference vectors obtained at different positions is calculated, and the cosine similarity is selected to avoid the influence of uneven illumination, thereby being used as a regularity measurement in a single direction.
Further, any sub-block in the target window is marked as a target sub-block; the sub-block of the target sub-block that is symmetrical about the center sub-block of the target window is denoted as a reference sub-block.
The symmetrical similarity of the target sub-block is obtained by the following steps:
s1: subtracting the characteristic vector of the central sub-block of the target window from the characteristic vector of the target sub-block to obtain a difference vector of the target sub-block; subtracting the characteristic vector of the reference sub-block from the characteristic vector of the central sub-block of the target window to obtain a difference vector of the reference sub-block;
S2: calculating cosine similarity of the difference vector of the target sub-block and the difference vector of the reference sub-block;
s3: the positions of all response values in the feature vector of the target sub-block are moved backwards by one bit, and the last response value of the feature vector of the target sub-block is moved to the first response value position of the feature vector of the target sub-block, so that a new feature vector of the target sub-block is obtained;
s4: repeating S1 to S3 until the repetition number is equal to HD-1 times, and stopping repeating.
The maximum cosine similarity obtained in S1 to S4 is recorded as the symmetrical similarity of the target sub-block;
According to the method, the symmetrical similarity of each sub-block in the target window is obtained. The method for calculating the cosine similarity is a known technique, and the specific method is not described here.
The gradient regularity of the target pixel is calculated by the following steps:
In the method, in the process of the invention, Representing the gradient regularity of the target pixel points; /(I)Representing the number of sub-blocks in the target window after the central sub-block is removed; /(I)Indicating the confusion degree of the q-th sub-block after the central sub-block is removed in the target window; /(I)Representing the degree of confusion of the center sub-block of the target window; /(I)Representing the symmetrical similarity of the qth sub-block after the center sub-block is removed from the target window.
In the method, in the process of the invention,The larger the value of the weight coefficient of the symmetry similarity is, the more likely the q-th sub-block in the target window is removed from the center sub-block and the center sub-block in the target window is a texture area, and the higher the reliability of the symmetry similarity is.
According to the method, the gradient regularity of each pixel point in the double-skin wall image is obtained.
Step S004: and obtaining the inhibition degree of each pixel point according to the gradient regularity of the pixel points in the square window corresponding to each pixel point in the double-skin wall image.
It should be noted that, the gradient regularity is measured based on the target window, when the target pixel is not a crack pixel and a crack defect exists in the target window, the gradient regularity of the target window is also lower, which affects the subsequent edge detection result, and causes the situation that the finally obtained edge detection result generates a wide edge or the trend of the crack edge is blurred, so after the gradient regularity of each pixel is obtained, the gradient regularity of each pixel and the relationship between the regularity of each pixel in the square window corresponding to the pixel need to be considered, and the suppression degree of each pixel is further obtained.
Specifically, in the double-skin wall image, the calculation mode of the inhibition degree of the ith pixel point is as follows:
In the method, in the process of the invention, The inhibition degree of the ith pixel point; /(I)Is a linear normalization function; /(I)In a square window corresponding to the ith pixel point, the gradient regularity is greater than the number of the pixel points of the gradient regularity of the ith pixel point; The number of the pixel points in the square window corresponding to the ith pixel point; /(I) The gradient regularity of the ith pixel point; The gradient regularity of the j-th pixel except the i-th pixel is in the square window corresponding to the i-th pixel.
In the method, in the process of the invention,The larger the value of the weight coefficient of the gradient regularity of the ith pixel point is, the more likely the whole in the square window corresponding to the ith pixel point is a texture area, and the reliability of the gradient regularity is higher; /(I)The method comprises the steps that in a square window corresponding to an ith pixel point, the larger the value of a weight coefficient of gradient regularity of other pixel points except the ith pixel point is, the larger the reference value of the gradient regularity of the other pixel points for calculating the inhibition degree of the ith pixel point is; The larger the value of the overall gradient regularity of the pixel points except the ith pixel point in the square window corresponding to the ith pixel point, the higher the probability that the ith pixel point is a texture pixel point, and the greater the inhibition degree of the ith pixel point.
Step S005: obtaining a gradient correction amplitude of each pixel point according to the inhibition degree of each pixel point in the double-skin wall image; based on the gradient correction amplitude value of each pixel point in the double-skin wall image, obtaining a crack pixel point by utilizing an edge detection algorithm; and (5) according to the number of the crack pixel points, obtaining a visual detection result of the double-skin wall automatic production line.
It should be noted that, by adjusting the gradient amplitude of the non-crack pixel point in the double-skin wall image, the accuracy of the crack pixel point in the edge detection result is improved.
Specifically, in the double-skin wall image, the gradient amplitude of each pixel point is obtained by utilizing a Sobel operator.
The gradient correction amplitude of the ith pixel point is:
Wherein, Correcting the amplitude value for the gradient of the ith pixel point; /(I)The inhibition degree of the ith pixel point; /(I)The gradient magnitude for the i-th pixel.
According to the method, the gradient correction amplitude value of each pixel point in the double-skin wall image is obtained.
Specifically, based on the gradient correction amplitude of each pixel point in the double-skin wall image, a Canny edge detection algorithm is utilized to obtain a crack pixel point, wherein the crack pixel point is an edge pixel point detected by an edge; acquiring the number of the crack pixel points, and judging that the crack defect does not exist on the surface of the double-skin wall when the number of the crack pixel points is smaller than a preset crack threshold LW; when the number of the crack pixel points is larger than or equal to a preset crack threshold LW, judging that the crack defect exists on the surface of the double-skin wall, and alarming and timely processing are needed; the preset crack threshold lw=10 is described as an example, and other values may be set in other embodiments, which is not limited in this example.
The present invention has been completed.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the invention, but any modifications, equivalent substitutions, improvements, etc. within the principles of the present invention should be included in the scope of the present invention.
Claims (9)
1. The intelligent visual monitoring method for the double-skin wall automatic production line is characterized by comprising the following steps of:
acquiring a double-skin wall image; constructing a square window corresponding to each pixel point in the double-skin wall image and sub-blocks in the square window; obtaining a feature vector of each pixel point according to the gradient of the pixel point in the sub-block;
In the double-skin wall image, marking any pixel as a target pixel; marking a square window corresponding to the target pixel point as a target window; obtaining the chaotic degree of each sub-block in the target window according to the gradient of all the pixel points in each sub-block in the target window;
Screening out the characteristic vector of each sub-block in the target window according to the characteristic vector of the target pixel point; according to the feature vector of each sub-block in the target window, obtaining the symmetrical similarity of each sub-block in the target window; obtaining the gradient regularity of the target pixel point according to the symmetry similarity and the chaotic degree of each sub-block in the target window;
Obtaining the inhibition degree of each pixel point according to the gradient regularity of the pixel points in the square window corresponding to each pixel point in the double-skin wall image;
Obtaining a gradient correction amplitude of each pixel point according to the inhibition degree of each pixel point in the double-skin wall image; based on the gradient correction amplitude value of each pixel point in the double-skin wall image, obtaining a crack pixel point by utilizing an edge detection algorithm; according to the number of the crack pixel points, a visual detection result of the double-skin wall automatic production line is obtained;
the method for obtaining the feature vector of each pixel point according to the gradient of the pixel point in the sub-block comprises the following specific steps:
Calculating the gradient vector of each pixel point by utilizing a Sobel operator according to the gray value of the pixel point in each sub-block; according to the gradient vector of each pixel point in each sub-block, constructing a gradient histogram of HD (high definition) bin directions of each sub-block, wherein the HD represents the number of preset bin directions; the abscissa of the gradient histogram is the bin direction; the ordinate is the accumulated value of the gradient values of all pixel points in the sub-block in the bin direction corresponding to the respective angle, and the accumulated value is recorded as a response value;
In the double-skin-wall image, the feature vector of any pixel point is as follows:
In the method, in the process of the invention, Is the characteristic vector of the pixel point; /(I)、/>/>Respectively representing response values in the 1 st, 2 nd and HD bin directions in the 1 st sub-block in the square window corresponding to the pixel point; /(I)And/>Respectively representing a response value in the 1 st bin direction and a response value in the 2 nd bin direction in the 2 nd sub-block in the square window corresponding to the pixel point; /(I)And representing the response value in the HD bin direction in the ZK sub-block in the square window corresponding to the pixel point.
2. The intelligent visual monitoring method for the double-skin wall automatic production line according to claim 1, wherein the construction of the square window and the sub-blocks in the square window corresponding to each pixel point in the double-skin wall image comprises the following specific steps:
In the double-skin wall image, an N multiplied by N square window is constructed by taking any pixel point as a center, wherein N represents a preset side length; dividing the square window into a plurality of sub-blocks with the size of M multiplied by M; m represents the preset side length of the sub-block.
3. The intelligent visual monitoring method for the double-skin wall automatic production line according to claim 1, wherein the obtaining the chaotic degree of each sub-block in the target window according to the gradient of all pixel points in each sub-block in the target window comprises the following specific calculation method:
In the method, in the process of the invention, The degree of confusion for the p-th sub-block in the target window; /(I)The number of pixel points contained in the p-th sub-block in the target window is the number; /(I)Is a linear normalization function; /(I)Gradient vector of the z pixel point in the p-th sub-block in the target window; /(I)The module length of the gradient vector of the z pixel point in the p-th sub-block in the target window is given; Gradient vector of the z pixel point in the p-th sub-block in the target window; /(I) The average gradient vector of the gradient vector of all pixel points in the p-th sub-block in the target window; /(I)For/>And/>A sine value of the included angle; /(I)As a function of absolute value.
4. The intelligent visual monitoring method for the double-skin-wall automatic production line according to claim 1, wherein the step of screening out the feature vector of each sub-block in the target window according to the feature vector of the target pixel point comprises the following specific steps:
and in the characteristic vector of the target pixel point, the sequence formed by the response values in all bin directions corresponding to each sub-block in the target window is recorded as the characteristic vector of each sub-block.
5. The intelligent visual monitoring method for the double-skin wall automatic production line according to claim 1, wherein the method for obtaining the symmetrical similarity of each sub-block in the target window according to the feature vector of each sub-block in the target window comprises the following specific steps:
Marking any sub-block in the target window as a target sub-block; marking the sub-block of the target sub-block symmetrical to the central sub-block of the target window as a reference sub-block;
s1: subtracting the characteristic vector of the central sub-block of the target window from the characteristic vector of the target sub-block to obtain a difference vector of the target sub-block; subtracting the characteristic vector of the reference sub-block from the characteristic vector of the central sub-block of the target window to obtain a difference vector of the reference sub-block;
S2: calculating cosine similarity of the difference vector of the target sub-block and the difference vector of the reference sub-block;
s3: the positions of all response values in the feature vector of the target sub-block are moved backwards by one bit, and the last response value of the feature vector of the target sub-block is moved to the first response value position of the feature vector of the target sub-block, so that a new feature vector of the target sub-block is obtained;
S4: repeating S1 to S3 until the repetition number is equal to HD-1 times, and stopping repeating;
And (5) recording the maximum cosine similarity obtained in S1 to S4 as the symmetrical similarity of the target subblocks.
6. The intelligent visual monitoring method for the double-skin-wall automatic production line according to claim 1, wherein the gradient regularity of the target pixel is obtained according to the symmetry similarity and the confusion degree of each sub-block in the target window, and the specific calculation method comprises the following steps:
In the method, in the process of the invention, Representing the gradient regularity of the target pixel points; /(I)Representing the number of sub-blocks in the target window after the central sub-block is removed; /(I)Indicating the confusion degree of the q-th sub-block after the central sub-block is removed in the target window; /(I)Representing the degree of confusion of the center sub-block of the target window; /(I)Representing the symmetrical similarity of the qth sub-block after the center sub-block is removed from the target window.
7. The intelligent visual monitoring method for the double-skin wall automatic production line according to claim 1, wherein the step of obtaining the inhibition degree of each pixel point according to the gradient regularity of the pixel point in the square window corresponding to each pixel point in the double-skin wall image comprises the following specific steps:
In the method, in the process of the invention, The inhibition degree of the ith pixel point; /(I)Is a linear normalization function; /(I)In a square window corresponding to the ith pixel point, the gradient regularity is greater than the number of the pixel points of the gradient regularity of the ith pixel point; /(I)The number of the pixel points in the square window corresponding to the ith pixel point; /(I)The gradient regularity of the ith pixel point; /(I)The gradient regularity of the j-th pixel except the i-th pixel is in the square window corresponding to the i-th pixel.
8. The intelligent visual monitoring method for the double-skin wall automatic production line according to claim 1, wherein the step of obtaining the gradient correction amplitude of each pixel point according to the inhibition degree of each pixel point in the double-skin wall image comprises the following specific steps:
And calculating a difference value of 1 minus the inhibition degree of the ith pixel point, and recording the product of the difference value and the gradient amplitude of the ith pixel point as the gradient correction amplitude of the ith pixel point.
9. The intelligent visual monitoring method for the double-skin wall automatic production line according to claim 1, wherein the visual detection result of the double-skin wall automatic production line is obtained according to the number of the crack pixel points, and the method comprises the following specific steps:
Acquiring the number of the crack pixel points, and judging that the crack defect does not exist on the surface of the double-skin wall when the number of the crack pixel points is smaller than a preset crack threshold LW; and when the number of the crack pixel points is larger than or equal to a preset crack threshold LW, judging that the crack defect exists on the surface of the double-skin wall.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410369999.0A CN117975374B (en) | 2024-03-29 | 2024-03-29 | Intelligent visual monitoring method for double-skin wall automatic production line |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410369999.0A CN117975374B (en) | 2024-03-29 | 2024-03-29 | Intelligent visual monitoring method for double-skin wall automatic production line |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117975374A CN117975374A (en) | 2024-05-03 |
| CN117975374B true CN117975374B (en) | 2024-06-18 |
Family
ID=90855860
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410369999.0A Active CN117975374B (en) | 2024-03-29 | 2024-03-29 | Intelligent visual monitoring method for double-skin wall automatic production line |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117975374B (en) |
Family Cites Families (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107292900A (en) * | 2017-07-05 | 2017-10-24 | 黄河科技学院 | A kind of method for detecting image edge and device based on Canny algorithms |
| CN107993193B (en) * | 2017-09-21 | 2021-06-11 | 沈阳工业大学 | Tunnel lining image splicing method based on illumination equalization and surf algorithm improvement |
| KR20230056807A (en) * | 2021-10-19 | 2023-04-28 | 세종대학교산학협력단 | Hybrid damage evaluation system for masonry construction structure using image data, and method for the same |
| CN113989450B (en) * | 2021-10-27 | 2023-09-26 | 北京百度网讯科技有限公司 | Image processing method, device, electronic equipment and medium |
| CN113989257A (en) * | 2021-11-09 | 2022-01-28 | 国网江苏省电力有限公司南通供电分公司 | Electric power comprehensive pipe gallery settlement crack identification method based on artificial intelligence technology |
| CN114663374A (en) * | 2022-03-14 | 2022-06-24 | 深圳蓄能发电有限公司 | Tunnel wall crack detection method based on bilateral filtering and gamma conversion |
| CN114705689A (en) * | 2022-04-14 | 2022-07-05 | 西安建筑科技大学 | Unmanned aerial vehicle-based method and system for detecting cracks of outer vertical face of building |
| CN115375688B (en) * | 2022-10-25 | 2023-04-28 | 苏州德斯米尔智能科技有限公司 | Automatic detection method for belt conveyor |
| CN116310784A (en) * | 2022-12-20 | 2023-06-23 | 西南石油大学 | Core picture crack identification method based on image leveling processing |
| CN116485719A (en) * | 2023-03-13 | 2023-07-25 | 中山大学 | Self-adaptive canny method for crack detection |
| CN116343157A (en) * | 2023-04-17 | 2023-06-27 | 山东科技大学 | Deep learning extraction method for road surface cracks |
| CN116777898B (en) * | 2023-07-21 | 2024-01-23 | 重庆大学溧阳智慧城市研究院 | Method for realizing crack measurement in 3D printing retaining wall construction process based on AFFormer |
| CN116630813B (en) * | 2023-07-24 | 2023-09-26 | 青岛奥维特智能科技有限公司 | Highway road surface construction quality intelligent detection system |
| CN117237368B (en) * | 2023-11-16 | 2024-01-26 | 临沂市公路事业发展中心兰陵县中心 | Bridge crack detection method and system |
-
2024
- 2024-03-29 CN CN202410369999.0A patent/CN117975374B/en active Active
Non-Patent Citations (2)
| Title |
|---|
| 三星堆城墙干缩裂缝开裂与扩展机理;曲瑾;马建林;杨柏;;工程地质学报;20200615(第03期);全文 * |
| 基于张量投票和目标追踪的墙体裂缝检测算法研究;徐亚明;束进芳;;大地测量与地球动力学;20160415(第04期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117975374A (en) | 2024-05-03 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109559324B (en) | Target contour detection method in linear array image | |
| CN115294140B (en) | Hardware part defect detection method and system | |
| CN115239735A (en) | Communication cabinet surface defect detection method based on computer vision | |
| CN115239704A (en) | Accurate detection and repair method for wood surface defects | |
| CN116385450B (en) | PS sheet wear resistance detection method based on image processing | |
| JP2023139099A (en) | Learning data collecting apparatus, learning data collecting method, and program | |
| US6993187B2 (en) | Method and system for object recognition using fractal maps | |
| CN115797361B (en) | Aluminum template surface defect detection method | |
| CN115861290A (en) | Method for detecting surface defects of skin-touch wooden door | |
| CN117291919B (en) | Visual detection method for lubricating grease on surface of steel rail bolt | |
| CN116883408B (en) | Integrating instrument shell defect detection method based on artificial intelligence | |
| CN115311267A (en) | Method for detecting abnormity of check fabric | |
| CN117808796B (en) | Gear surface damage detection method based on computer vision | |
| CN117952979B (en) | Construction engineering quality detection supervision method and system | |
| CN115222761A (en) | Polyphenyl particle product quality detection method based on computer vision technology | |
| CN117314901B (en) | Scale-adaptive chip detection neural network system | |
| CN117152180B (en) | Mineral casting defect detection method based on artificial intelligence | |
| CN117975374B (en) | Intelligent visual monitoring method for double-skin wall automatic production line | |
| CN117078676B (en) | Breaking hammer shell visual detection method based on image processing | |
| CN117173160A (en) | Image processing-based method for detecting quality of caravan component | |
| CN117011280A (en) | 3D printed concrete wall quality monitoring method and system based on point cloud segmentation | |
| CN116563286B (en) | Mobile hard disk box production quality rapid detection system | |
| CN116993740B (en) | Concrete structure surface defect detection method based on image data | |
| CN118097310B (en) | Method for digitally detecting concrete surface defects | |
| CN117853494B (en) | Visual inspection method for drug particle production process |
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 |