CN116563202A - Building surface crack identification method and system based on convolutional neural network - Google Patents
Building surface crack identification method and system based on convolutional neural network Download PDFInfo
- Publication number
- CN116563202A CN116563202A CN202211478331.7A CN202211478331A CN116563202A CN 116563202 A CN116563202 A CN 116563202A CN 202211478331 A CN202211478331 A CN 202211478331A CN 116563202 A CN116563202 A CN 116563202A
- Authority
- CN
- China
- Prior art keywords
- crack
- image
- neural network
- building
- convolutional neural
- 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.)
- Pending
Links
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 83
- 238000000034 method Methods 0.000 title claims abstract description 34
- 238000001514 detection method Methods 0.000 claims abstract description 26
- 238000012549 training Methods 0.000 claims description 44
- 238000013528 artificial neural network Methods 0.000 claims description 32
- 238000012545 processing Methods 0.000 claims description 25
- 238000005336 cracking Methods 0.000 claims description 20
- 238000003709 image segmentation Methods 0.000 claims description 16
- 230000003993 interaction Effects 0.000 claims description 7
- 230000011218 segmentation Effects 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000007547 defect Effects 0.000 abstract description 21
- 239000002699 waste material Substances 0.000 abstract description 2
- 238000010276 construction Methods 0.000 description 14
- 230000000694 effects Effects 0.000 description 14
- 230000008569 process Effects 0.000 description 9
- 238000002372 labelling Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000003908 quality control method Methods 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000009429 electrical wiring Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- 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
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30132—Masonry; Concrete
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a building surface crack identification method and system based on a convolutional neural network, firstly, constructing a convolutional neural network model capable of identifying building surface cracks; then, collecting an image of the surface of the building to be detected; and finally, carrying out crack identification and display on the acquired building surface image to be tested by using the constructed convolutional neural network model. According to the scheme, the image recognition is introduced into the crack recognition of the building surface, so that unmanned crack recognition can be realized, and the defects of time and labor waste and high subjectivity of the existing manual detection means are overcome.
Description
Technical Field
The invention relates to the field of civil engineering measuring equipment, in particular to a construction surface crack identification technology.
Background
As a very important working content in the building engineering, the construction quality of the plastering engineering directly influences the use experience of users and the overall construction effect of the building. However, due to the fact that the number of the joint surfaces of plastering engineering is large, quality defects are easy to occur, wherein wall plastering cracking is a common problem in engineering and is also a content that construction quality control is difficult.
In order to ensure the construction quality of the wall, development of detection of the construction quality to find cracks and repair in time are one of the common quality control means. However, the manual detection means adopted at present are time-consuming and labor-consuming, have high subjectivity, and meanwhile, the cracks on the surface of the wall body are sometimes difficult to find, so that the manual detection is easy to ignore, and especially the corner parts of the wall body.
Therefore, how to solve the problems of low efficiency and low precision existing in the existing manual detection of cracks on the surface of a building is a problem to be solved in the field.
Disclosure of Invention
Aiming at the problems of high manual participation, low measurement efficiency, high subjectivity and the like in the existing building surface crack recognition scheme, the invention provides a building surface crack recognition method based on a convolutional neural network, which can realize unmanned inspection on the building surface and quickly recognize cracks so as to overcome the defects of the existing manual recognition; on the basis, the invention also provides a system for identifying the cracks on the building surface, which can realize the method.
In order to achieve the above purpose, the construction surface crack identification method based on convolutional neural network provided by the invention comprises the following steps:
step 1: constructing a convolutional neural network model capable of identifying cracks on the surface of a building;
step 2: collecting an image of the surface of a building to be tested;
step 3: and (3) performing crack identification and display on the acquired building surface image to be tested by using the convolutional neural network model constructed in the step (1).
In some examples of the present invention, the step (1) when constructing the convolutional neural network model includes the following sub-steps:
step 1-1: collecting n building surface crack images in an actual building;
step 1-2: preprocessing a crack image on the surface of a building to form a plurality of sub-images;
step 1-3: manually judging the crack degree in each sub-image, respectively judging that no crack, weak crack and strong crack exist, and respectively giving corresponding label values;
step 1-4: carrying out data enhancement on each split sub-image;
step 1-5: on the basis of an atomic image, adding images with label values corresponding to weak cracks and strong cracks in the images formed by data enhancement into a training set of a neural network, and finally forming an image training set consisting of m crack gray images;
step 1-6: training a convolutional neural network model using the formed image training set, the convolutional neural network model outputting three tag values a, b, c representing the degree of cracking.
In some examples of the present invention, when the sub-images are data-enhanced in the steps 1 to 4, a random number λ is first generated so as to satisfy λ to Be (0.5 ), and then the pixel values x of any two images are expressed by the formulas (1) and (2) 1 ，x 2 And its corresponding tag value y 1 ，y 2 Pixel value and label value x fused into new image new ，y new ：
x new ＝λx 1 +(1-λ)x 2 Formula (1);
y new ＝λy 1 +(1-λ)y 2 formula (2).
In some examples of the present invention, the step (3) includes the following sub-steps when performing crack identification and display:
step 3-1: performing large-scale crack detection on the surface image of the building to be detected based on the constructed convolutional neural network model;
step 3-2: performing small-scale crack grading on the surface image of the building to be tested based on the constructed convolutional neural network model;
step 3-3: combining the identification patterns obtained through the processing of the step 3-1 and the step 3-2 to form the building surface image with the crack grade mark.
In some examples of the present invention, the step 3-1 for large-scale crack detection includes the following sub-steps:
step 3-1-1: dividing an image to be identified into a plurality of 128×128 small images at 64 pixel intervals;
step 3-1-2: scaling the small image obtained in the step 3-1-1 to 64×64 size, and then importing the small image into the convolutional neural network model trained in the step 1 according to rows; processing the input small image by a convolutional neural network model to obtain corresponding label values a, b and c;
step 3-1-3: performing crack identification and determination according to the tag values a, b and c;
step 3-1-4: and (5) corresponding the identification result of the crack to the original image, and marking the original image according to the identification result.
In some examples of the present invention, the step 3-2 for small scale crack grading comprises the following sub-steps:
step 3-2-1: dividing an image to be identified into a plurality of 64 multiplied by 64 small images at intervals of 64 pixels;
step 3-2-2: leading the small images obtained in the step 3-2-1 into the convolutional neural network model trained in the step 1 according to rows, and processing the input small images by the convolutional neural network model to obtain corresponding tag values a, b and c;
step 3-2-3: performing crack grade result identification and determination according to the label values a, b and c;
step 3-2-4: and the identified crack grade result is corresponding to the original image, and the original image is marked according to the identified crack grade result.
In order to achieve the above object, the present invention provides a convolutional neural network-based construction surface crack recognition system, comprising:
a neural network training module that trains a neural network based on the architectural surface crack image, constructs a convolutional neural network model capable of identifying architectural surface cracks,
the building surface image acquisition module is used for acquiring images of the building surface to be detected;
and the crack identification module is used for carrying out crack identification processing on the to-be-detected building surface image acquired by the building surface image acquisition module by utilizing the convolutional neural network model constructed by the neural network training module.
In some examples of the present invention, the convolutional neural network model trained by the neural network training module is capable of generating a plurality of label values representing the degree of cracking for the image processing to be processed.
In some examples of the invention, the crack recognition module includes an image segmentation module, an input module, a decision module, and a merge module,
the image segmentation module is used for segmenting the surface image of the building to be detected into a plurality of sub-images for large-scale crack detection or into a plurality of sub-images for small-scale crack grading;
the input module is in data interaction with the image segmentation module, and a plurality of sub-images obtained by segmentation of the image segmentation module are led into a trained convolutional neural network model according to rows;
the judging module carries out crack detection recognition or crack grading recognition on the image of the surface of the building to be detected according to the output result of the convolutional neural network model, and marks the original image of the surface of the building to be detected according to the recognition result;
and the merging module performs data interaction with the judging module, merges the to-be-detected building surface images with the crack marks and the crack grade marks, and forms a building surface image with the crack grade marks.
Compared with the prior art, the scheme provided by the invention has the beneficial effects that:
1) According to the scheme, the image recognition is introduced into the crack recognition on the surface of the building, so that unmanned crack recognition can be realized, and the defects of time and labor waste and high subjectivity of the existing manual detection means are overcome;
2) The technical scheme of the invention innovatively adopts a double-scale crack identification technology (namely, large-scale crack detection and small-scale crack grading) so as to avoid the special condition that crack information is difficult to extract due to the fact that the crack position is just on an image segmentation line, and ensure that crack identification has higher accuracy.
3) The construction surface crack recognition scheme provided by the invention can be fused with unmanned aerial vehicle, robot and other equipment, so that unmanned and automatic engineering quality detection can be realized.
Drawings
The invention is further described below with reference to the drawings and the detailed description.
FIG. 1 is an overall flow chart of a convolutional neural network-based method for identifying cracks on a building surface in an example of the present invention;
FIG. 2 is a flow chart of the construction of a convolutional neural network model in an example of the invention;
FIG. 3 is a flow chart illustrating crack identification and display in accordance with an embodiment of the present invention;
FIG. 4 is a schematic diagram of a crack identification and display process in accordance with an embodiment of the present invention;
FIG. 5 is a flow chart of a large scale crack detection in an example of the invention;
FIG. 6 is a flow chart of small scale crack grading in an example of the invention;
FIG. 7 is a diagram showing an example of the construction of a system for identifying cracks in a building surface according to an example of the present invention;
FIG. 8 is a graph showing the effect of identifying wall cracks at internal corners in an example of the invention.
Detailed Description
The invention is further described with reference to the following detailed drawings in order to make the technical means, the creation characteristics, the achievement of the purpose and the effect of the implementation of the invention easy to understand.
Aiming at the problems of high manual participation, low measurement efficiency, high subjectivity and the like of the existing building surface crack recognition, the inventor innovates and applies digital image recognition to the defect detection of a building structure through a large amount of researches, provides a building surface crack recognition technology based on a convolutional neural network, can greatly improve the detection precision and reduce the labor cost.
Accordingly, the invention provides a building surface crack recognition method based on a convolutional neural network, which is combined with the method shown in fig. 1, and is mainly formed by matching the following steps:
step 1: constructing a convolutional neural network model capable of identifying cracks on the surface of a building;
step 2: collecting an image of the surface of a building to be tested;
step 3: and (3) performing crack identification and display on the acquired building surface image to be tested by using the convolutional neural network model constructed in the step (1).
In some embodiments of the present invention, referring to fig. 2, in constructing the convolutional neural network model for step (1), the following sub-steps are specifically included:
step 1-1: collecting n building surface crack images in an actual building;
the n building surface crack images collected here are used as samples for subsequent convolutional neural network training. The image of the crack on the surface of the building is not particularly required here, and can be determined according to actual requirements.
Step 1-2: and (3) preprocessing n building surface crack images acquired in the step (1-1) to form a plurality of sub-images.
Specifically, in this step, exposure and contrast are adjusted for the acquired crack image, and the original image is randomly cut into sub-images with a size of 64×64.
In the step, the original image is preferably cut into 64×64 sub-images, so that the subsequent neural network is convenient for sampling training, and the processing efficiency is improved.
Step 1-3: and (3) manually judging the crack degree in each sub-image obtained by processing in the step (1-2), and respectively giving corresponding label values.
By way of example, in this step, the degree of cracking in each sub-image is determined by means of human judgment, and is determined as no cracking, weak cracking and strong cracking, respectively, and the corresponding label values are given as 0.0-no cracking, 0.5-weak cracking and 1.0-strong cracking.
Three crack extent criteria are preferred in this step: 0.0-no crack, 0.5-weak crack and 1.0-strong crack, so that the method can be well matched with the subsequent convolutional neural network model processing, and the recognition accuracy is improved.
Based on three crack extent criteria, preferably: and identifying each sub-image serving as a sample according to the crack-free 0.0-crack, the crack-weak 0.5-crack and the crack-strong 1.0-crack, so that the training effect of the subsequent neural network can be improved, and the recognition processing precision of the convolutional neural network model is improved.
Here, the determination of the degree of cracking in the image is not limited to three levels of no cracking, weak cracking, and strong cracking, and other determination levels may be employed as needed.
The label values given to different degrees of cracking are not limited to three label values of 0.0 (no cracking), 0.5 (weak cracking), and 1.0 (strong cracking), and other label values may be used as needed.
Step 1-4: and (3) carrying out data enhancement on the sub-images which are segmented and processed in the step (1-3) by adopting a mixup algorithm.
In the step, when the data enhancement is carried out on the sub-images, new sub-image data is generated by fine-tuning the original sub-image data, so that enhanced sub-images are formed and are used as new sub-image samples to be supplemented into a training set. Therefore, the subsequent training set can be further perfected and supplemented, and the subsequent neural network training effect is improved.
By way of example, this step preferably achieves data enhancement for each sub-image by:
firstly generating a random number lambda to meet lambda-Be (0.5 ), and then using the pixel value x of any two images according to the formula (1) and the formula (2) 1 ，x 2 And its corresponding tag value y 1 ，y 2 Pixel value and label value x fused into new image new ，y new ：
x new ＝λx 1 +(1-λ)x 2 Formula (1);
y new ＝λy 1 +(1-λ)y 2 formula (2).
Step 1-5: based on the atomic image, combining the images formed by data enhancement in the step 1-4, adding the images with the label value of 0.45-0.55 (weak defect) and 0.9-1.0 (strong defect) to a training set of the neural network, and finally forming the training set consisting of m crack gray level images of 64×64.
Specifically, in this step, the atomic image generated in step 1-3 and the image formed by enhancing the data in step 1-4 are combined, the atomic image with a tag value of 0.45-0.55 (weak defect) and the corresponding enhanced sub-image are extracted simultaneously, and the atomic image with a tag value of 0.9-1.0 (strong defect) and the corresponding enhanced sub-image are extracted; on the basis, all the extracted images are added to a training set of the neural network, and finally the training set consisting of m crack gray level images of 64×64 is formed.
Step 1-6: and training a convolutional neural network model by adopting the formed image training set, wherein the input end of the network is a 64 multiplied by 64 gray scale image, flattening after convolution for a plurality of times, and finally outputting three label values a, b and c representing the crack degree through a fully-connected neural network. The convolution neural network finally obtained can realize the functions of inputting an image and outputting three label values a, b and c representing the crack degree.
By way of example, the correspondence between a, b, c and the degree of cracking in this step is shown in table 1.
TABLE 1 correspondence between tag values a, b, c and crack levels
| Label (Label) | a | b | c |
| Strong defects | 0 | 0 | 1 |
| Weak defect | 0 | 1 | 0 |
| Defect free | 1 | 0 | 0 |
According to the scheme, a convolutional neural network model is trained through an image training set formed by sub-images serving as training samples and achieving crack degree calibration (giving label values corresponding to the crack degrees) in the step 1-3, each training sample sub-image achieving crack degree calibration is firstly input into a neural network based on the corresponding relation in the table 1, the primary neural network outputs three label values a ', b', c 'representing the crack degrees through processing, and parameters of the neural network are adjusted through judging differences between the label values a', b ', c' and the label values a, b and c of the input training sample sub-images. After the sub-images of different training samples are subjected to the operation, the target of the neural network model training can be realized, and the neural network training of the input image can be ensured to output accurate label values.
Therefore, a trained convolutional neural network model is obtained, the convolutional neural network model of the target picture can be realized in the follow-up process, the convolutional neural network model outputs a corresponding label value, and whether the target is a crack target is judged according to the label value.
In some embodiments of the present invention, referring to fig. 3, for step (3), the crack identification and display is performed, specifically including the following sub-steps:
step 3-1: performing large-scale crack detection on the building surface image to be detected acquired in the step (2) based on the constructed convolutional neural network model to form a building surface image with crack marks;
step 3-2: performing small-scale crack grading on the building surface image to be detected acquired in the step (2) based on the constructed convolutional neural network model to form a building surface image with crack grade marks;
step 3-3: combining the identification images (namely the building surface image with the crack mark and the building surface image with the crack grade mark) obtained through the steps 3-1 and 3-2 to form the building surface image with the crack grade mark.
In some embodiments of the present invention, for large-scale crack detection in step 3-1, the method specifically includes the following sub-steps, see fig. 5:
step 3-1-1: the image to be identified is partitioned into 128 x 128 small pictures at 64 pixel intervals.
In the step, the preferable image is divided into a plurality of 128×128 small pictures at intervals of 64 pixels, so that the subsequent neural network is convenient for sampling training, and the processing efficiency is improved.
Meanwhile, the image is divided into a plurality of 128×128 small pictures at intervals of 64 pixels for matching with the small size in the step 3-1-2, and crack identification is realized from a large dimension in the subsequent identification step, so that the accuracy of an identification result is improved.
Step 3-1-2: scaling the small pictures obtained in the step 3-1-1 to 64 multiplied by 64, and sequentially importing the small pictures into the convolutional neural network model trained in the step 1 according to rows; the convolutional neural network model processes the input small pictures in sequence and outputs corresponding tag values a, b and c.
Step 3-1-3: based on label values a, b and c output by the convolutional neural network model, judging whether each small picture has cracks or not: when the label values a, b and c meet the following conditions, judging that the small picture has surface cracks in the corresponding area on the surface image of the building to be detected: c >0.95 or a <0.05 or b+c > alpha; otherwise, judging that no surface crack exists.
In the step, the judgment condition of 'c >0.95 or a <0.05 or b+c > alpha' is adopted, so that certain fault tolerance variation can be realized, and the accuracy of a final judgment result can be ensured.
Here, it should be noted that, in this step, the threshold α may be adjusted to adapt to the requirements of different detection tasks for the recognition accuracy.
Step 3-1-4: and (3) the identification result of the step 3-1-3 on the cracks is corresponding to the original image, and the mark is marked according to the presence of the cracks (light red) and the absence of the cracks (colorless).
Specifically, in the step, the region corresponding to the small picture which is judged to be finished is determined on the building surface image to be detected, and then the region corresponding to the building surface image to be detected is marked according to the crack identification result of the small picture in the step 3-1-3.
For example, color is adopted for marking, and if the crack recognition result of the current small picture is a crack, the corresponding area on the surface image of the building to be detected is marked as light red; if the crack identification result of the current small picture is crack-free, the corresponding area on the image of the surface of the building to be detected is marked as colorless.
It should be noted that other colors may be used as needed for the color of the label, and are not limited to the exemplary embodiment.
And (3) completing processing of all the small pictures segmented in the step (3-1-1) in sequence, and then completing corresponding labeling of all areas on the to-be-tested building surface image, so as to form the building surface image with crack marks.
In some embodiments of the present invention, the small-scale crack grading is performed in step 3-2, specifically comprising the following sub-steps, see fig. 6:
step 3-2-1: the image to be identified is partitioned into 64 x 64 small pictures at 64 pixel intervals.
In the step, the preferable image is divided into a plurality of 64 multiplied by 64 small pictures at intervals of 64 pixels, so that the follow-up neural network is convenient to sample and train, and the processing efficiency is improved.
Meanwhile, the image is divided into a plurality of 64 multiplied by 64 small pictures at intervals of 64 pixels for matching with the large size in the step 3-1-1, and the recognition of the small dimension crack is carried out after the large dimension crack is recognized in the subsequent recognition step, so that the accuracy of the recognition result is improved.
Step 3-2-2: and (3) directly importing the small pictures obtained by segmentation in the step (3-2-1) into a trained convolutional neural network model according to rows, sequentially processing the input small pictures by the convolutional neural network model, and outputting corresponding tag values a, b and c.
Step 3-2-3: and identifying the crack level of each small picture based on the label values a, b and c output by the convolutional neural network model:
when c >0.95 or a <0.05, the grade of the crack on the small picture is considered as a strong defect;
if the requirements are not met, when b+c > alpha, the grade of the crack on the small picture is determined as a middle defect;
if the requirements are still not satisfied, when a > beta, the grade of the crack on the small picture is determined as a weak defect;
otherwise, the small picture is identified to have no crack.
The judging conditions are adopted in the step, so that certain fault tolerance variation can be realized, and the accuracy of a final judging result can be ensured.
Here, it should be noted that, in this step, the threshold values α, β may be adjusted to adapt to the requirements of different detection tasks for the recognition accuracy.
Step 3-2-4: and (3) corresponding the crack grade result identified in the step (3-2-3) to the original image, and carrying out corresponding marking on the corresponding area of the original image according to the crack grade result.
Specifically, in the step, the region corresponding to the identified small picture is determined on the surface image of the building to be detected, and then the region corresponding to the surface image of the building to be detected is marked according to the crack grade identification result of the small picture in the step 3-2-3.
For example, the color is adopted for marking, if the crack grade of the current small picture is determined to be a strong defect, the corresponding area on the surface image of the building to be detected is marked as dark red;
if the crack grade of the current small picture is determined to be a middle defect, marking the corresponding area on the surface image of the building to be detected as red;
if the crack grade of the current small picture is determined to be a weak defect, marking the corresponding area on the surface image of the building to be detected as light red;
and if the crack grade of the current small picture is determined to be crack-free, marking the corresponding area on the surface image of the building to be detected as colorless.
It should be noted that other colors may be used as needed for the color of the label, and are not limited to the exemplary embodiment.
And (3) completing processing of all the small pictures segmented in the step (3-2-1) in sequence, and then completing corresponding labeling of all areas on the to-be-tested building surface image, so as to form the building surface image with the crack grade marks.
In some embodiments of the present invention, for the merging in step 3-3, the formed building surface image has crack mark areas distributed in a large dimension while also having crack grade mark areas distributed in a small dimension based on the merging between the building surface image with crack marks that completes the large dimension identification and the building surface image with crack grade marks that completes the small dimension identification.
Thus, the area with cracks can be determined for each building surface image through the crack mark areas distributed on the building surface image in a large dimension; while crack areas on the building surface image that are not covered by crack-level marked areas in a large dimension are further verified by crack-level marked areas in a small dimension.
If the crack marking area with large dimension distribution is processed on the building surface image, if the crack is just on the dividing line of the large dimension dividing image, the crack marking area with large dimension distribution finally formed cannot cover the crack area; when the crack grade marking area with small dimension distribution is processed on the same building surface image, the same crack will not be positioned on the dividing line of the small dimension dividing image due to different dividing dimensions, and the finally formed crack grade marking area with small dimension distribution will be effectively covered on the crack area;
vice versa, i.e. if a crack is on a segmentation line of a small-dimensional segmented image, the crack will be identified by a crack marker area distributed in a large dimension.
The double-scale crack identification mode is realized, the special condition that crack information is difficult to extract due to the fact that the crack position is just on a picture dividing line can be effectively avoided, and the crack identification is guaranteed to have higher accuracy.
The following is a further description of the solution provided by the present invention by way of specific examples.
In the embodiment, aiming at the building surface crack identification method based on the convolutional neural network, a corresponding software program is formed, and a corresponding building surface crack identification system based on the convolutional neural network is formed. When the software program is run, the building surface crack identification method based on the convolutional neural network is executed, and meanwhile, the building surface crack identification method based on the convolutional neural network is stored in a corresponding storage medium for being called and executed by a processor.
Referring to fig. 7, the construction surface crack recognition system 100 based on the convolutional neural network mainly comprises a neural network training module 110, a construction surface image acquisition module 120 and a crack recognition module 130 when being implemented.
The neural network training module 110 trains the neural network based on the building surface crack image, and constructs a convolutional neural network model capable of identifying the building surface crack.
The building surface image acquisition module 120 is used for acquiring an image of a building surface to be tested.
The crack recognition module 130 performs data interaction with the neural network training module 110 and the building surface image acquisition module 120 respectively, and performs crack recognition processing on the building surface image to be detected acquired by the building surface image acquisition module by using the convolutional neural network model constructed by the neural network training module.
Specifically, when implementing the neural network training module 110 in the present system, it is preferably implemented by adopting the scheme of step 1 and the sub-steps thereof.
The convolutional neural network model trained by the neural network training module 110 can generate a plurality of label values representing the crack degree aiming at the image processing to be processed.
The specific configuration of the building surface image capturing module 120 in the present system may be determined according to practical requirements, and is not limited herein.
The crack recognition module 130 in the present system includes an image segmentation module 131, an input module 132, a determination module 133, and a combination module 134 in terms of construction, and the respective crack recognition modules 130 are constructed by the mutual cooperation of the four functional modules, so as to implement the scheme of step 3 described above.
The image segmentation module 131 is used for segmenting the building surface image to be detected acquired by the building surface image acquisition module 120 into a plurality of sub-images for large-scale crack detection or into a plurality of sub-images for small-scale crack grading.
The input module 132 is in data interaction with the image segmentation module 131, and is configured to guide a plurality of sub-images obtained by segmentation of the image segmentation module 131 into a convolutional neural network model trained by the neural network training module 110 according to rows;
the judging module 133 performs crack detection recognition or crack grading recognition on the to-be-detected building surface image according to the output result of the convolutional neural network model, marks the original image of the to-be-detected building surface according to the recognition result, and forms a building surface image with crack marks and a building surface image with crack grade marks.
The merging module 134 performs data interaction with the determining module 133, and merges the building surface image with the crack mark formed by the determining module 133 with the building surface image with the crack grade mark to form the building surface image with the crack grade mark.
The following illustrates the process of performing crack identification on a building surface of the building surface crack identification system based on the convolutional neural network formed in this example.
Before the building surface crack recognition system performs building surface crack recognition, firstly, n building surface crack images in an actual building are acquired by the building surface image acquisition module 120 and used for training by the neural network training module 110 so as to construct a convolutional neural network model capable of recognizing the building surface crack.
And the automatic recognition of the building surface cracks can be performed by the building surface crack recognition system after the convolutional neural network model training capable of recognizing the building surface cracks is completed.
As shown in connection with fig. 4, an image of the building surface to be measured is acquired by the building surface image acquisition module 120 and transmitted to the image segmentation module 131.
The image segmentation module 131 simultaneously performs the following segmentation processing on the received image of the building surface to be detected:
(1) Dividing an image of the surface of a building to be detected into a plurality of 128×128 first small pictures at intervals of 64 pixels;
(2) The picture to be identified is partitioned into 64 x 64 second small pictures at 64 pixel intervals.
The image segmentation module 131 transmits the segmented 128×128 first small pictures and the segmented 64×64 second small pictures to the input module 132.
The input module 132 scales the received first 128×128 small pictures to 64×64, and then imports the scaled first small pictures into a convolutional neural network model trained in the system in rows;
the input module 132 directs the received 64 x 64 second thumbnail images in a row into a convolutional neural network model trained in the system.
The convolutional neural network model in the system respectively and sequentially processes the input first small picture and the second small picture, and outputs corresponding tag values a, b and c.
The determination module 133 performs large-scale crack detection on the to-be-detected building surface image based on the label values a, b, c corresponding to each first small picture, and the whole process is as follows in combination with fig. 5:
the determining module 133 performs, for each first thumbnail, the following determination labeling process on the tag values a, b, c after receiving the corresponding tag values a, b, c output by the convolutional neural network model:
c >0.95 or a <0.05 or b+c > alpha, if so, judging that the small picture has surface cracks in the corresponding area on the surface image of the building to be detected; marking the corresponding area on the surface image of the building to be detected as light red;
if the two images are not established, judging that the small picture has no surface crack in the corresponding area on the surface image of the building to be detected, and marking the corresponding area on the surface image of the building to be detected as colorless, namely the original picture.
The determining module 133 sequentially completes processing of all the first small pictures, and then completes corresponding labeling of all areas on the surface image of the building to be detected, thereby forming the surface image of the building with crack marks.
Meanwhile, the determining module 133 performs small-scale crack grading identification on the to-be-detected building surface image based on the label values a, b and c corresponding to each second small picture, and the whole process is as follows in combination with fig. 6:
the determining module 133 performs, for each second small picture, the following determination labeling process on the tag values a, b, c after receiving the corresponding tag values a, b, c output by the convolutional neural network model:
firstly, judging that c is more than 0.95 or a is less than 0.05, and if so, judging that the grade of the crack on the second small picture is a strong defect; marking the corresponding area on the surface image of the building to be detected as dark red;
if not, further judging that b+c > alpha is true, and if so, judging that the grade of the crack on the second small picture is a middle defect; marking the corresponding area on the surface image of the building to be detected as red;
if not, further judging a > beta, and if so, judging the crack grade on the second small picture as a weak defect; marking the corresponding area on the surface image of the building to be detected as light red;
if not, determining that the second small picture has no crack; the corresponding area on the surface image of the building to be detected is not marked, and the building is the original image.
In this way, the determining module 133 sequentially completes processing all the second small pictures, and then completes corresponding labeling of all the areas on the surface image of the building to be tested, thereby forming the surface image of the building with the crack grade marks.
The determination module 133 transmits the resulting architectural surface image with the crack signature to the merge module 134 simultaneously with the architectural surface image with the crack grade signature.
The merge module 134 merges the two images to give a final crack identification and display in terms of final shade of color.
The method has the advantages that the automatic and accurate identification of the cracks on the surface of the building is realized by adopting the double-scale crack identification mode, the special condition that the crack information is difficult to extract due to the fact that the crack position is just on the picture dividing line can be effectively avoided, and the crack identification can be guaranteed to have higher accuracy.
To further verify the crack recognition effect of the solution provided by the present invention, the present example further carried out a practical application test of the solution of the present invention.
The practical project application results show that the crack identification technology provided by the invention can effectively identify cracks on the surface of a building, has better identification effect on cracks which are difficult to be found by naked eyes,
referring to fig. 8, an effect diagram of the method for identifying wall cracks at internal corners is shown.
Wherein fig. 8a is a graph of the actual effect of the reentrant corner crack 1; fig. 8b is a graph showing the effect of the recognition result of the reentrant corner crack 1 according to the present invention;
FIG. 8c is a graph showing the actual effect of the reentrant corner crack 2; and fig. 8d is a graph showing the effect of the recognition result of the reentrant corner crack 2 according to the present invention.
As can be seen from the practical application effect shown in FIG. 8, the scheme of the invention can realize accurate recognition effect even for wall cracks at the internal corners as shown in FIG. 8.
The above method of the present invention, or specific system units, or parts thereof, are pure software structures, and can be distributed on physical media, such as hard disks, optical discs, or any electronic devices (such as smart phones, computer readable storage media), when the machine loads the program codes and executes (such as smart phones loads and executes), the machine becomes a device for implementing the present invention. The methods and apparatus of the present invention may also be embodied in the form of program code that is transmitted over some transmission medium, such as over electrical wiring, optical fiber, or any other transmission medium, when the program code is received and loaded into and executed by a machine, such as a smart phone, the machine thereby providing an apparatus for practicing the methods.
The foregoing has shown and described the basic principles, principal features and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and that the above embodiments and descriptions are merely illustrative of the principles of the present invention, and various changes and modifications may be made without departing from the spirit and scope of the invention, which is defined in the appended claims. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (9)
1. The building surface crack identification method based on the convolutional neural network is characterized by comprising the following steps of:
step 1: constructing a convolutional neural network model capable of identifying cracks on the surface of a building;
step 2: collecting an image of the surface of a building to be tested;
step 3: and (3) performing crack identification and display on the acquired building surface image to be tested by using the convolutional neural network model constructed in the step (1).
2. The method for identifying cracks on a building surface according to claim 1, wherein the step (1) comprises the following sub-steps when constructing a convolutional neural network model:
step 1-1: collecting n building surface crack images in an actual building;
step 1-2: preprocessing a crack image on the surface of a building to form a plurality of sub-images;
step 1-3: manually judging the crack degree in each sub-image, respectively judging that no crack, weak crack and strong crack exist, and respectively giving corresponding label values;
step 1-4: carrying out data enhancement on each split sub-image;
step 1-5: on the basis of an atomic image, adding images with label values corresponding to weak cracks and strong cracks in the images formed by data enhancement into a training set of a neural network, and finally forming an image training set consisting of m crack gray images;
step 1-6: training a convolutional neural network model using the formed image training set, the convolutional neural network model outputting three tag values a, b, c representing the degree of cracking.
3. The method for recognizing cracks on a building surface according to claim 2, wherein when the sub-images are data-enhanced in the steps 1 to 4, a random number λ is first generated so as to satisfy λ to Be (0.5 ), and then the pixel values x of any two images are expressed by the formulas (1) and (2) 1 ，x 2 And its corresponding tag value y 1 ，y 2 Pixel value and label value x fused into new image new ，y new ：
x new ＝λx 1 +(1-λ)x 2 Formula (1);
y new ＝λy 1 +(1-λ)y 2 formula (2).
4. The method for identifying cracks on a building surface according to claim 1, wherein the step (3) comprises the following sub-steps when identifying and displaying cracks:
step 3-1: performing large-scale crack detection on the surface image of the building to be detected based on the constructed convolutional neural network model;
step 3-2: performing small-scale crack grading on the surface image of the building to be tested based on the constructed convolutional neural network model;
step 3-3: combining the identification patterns obtained through the processing of the step 3-1 and the step 3-2 to form the building surface image with the crack grade mark.
5. The method for identifying cracks on a building surface according to claim 4, wherein the step 3-1 of performing large-scale crack detection comprises the following sub-steps:
step 3-1-1: dividing an image to be identified into a plurality of 128×128 small images at 64 pixel intervals;
step 3-1-2: scaling the small image obtained in the step 3-1-1 to 64×64 size, and then importing the small image into the convolutional neural network model trained in the step 1 according to rows; processing the input small image by a convolutional neural network model to obtain corresponding label values a, b and c;
step 3-1-3: performing crack identification and determination according to the tag values a, b and c;
step 3-1-4: and (5) corresponding the identification result of the crack to the original image, and marking the original image according to the identification result.
6. The method for identifying cracks on a building surface according to claim 4, wherein the step 3-2 of grading small-scale cracks comprises the following sub-steps:
step 3-2-1: dividing an image to be identified into a plurality of 64 multiplied by 64 small images at intervals of 64 pixels;
step 3-2-2: leading the small images obtained in the step 3-2-1 into the convolutional neural network model trained in the step 1 according to rows, and processing the input small images by the convolutional neural network model to obtain corresponding tag values a, b and c;
step 3-2-3: determining a crack grade result according to the label values a, b and c;
step 3-2-4: and the identified crack grade result is corresponding to the original image, and the original image is marked according to the identified crack grade result.
7. Building surface crack identification system based on convolutional neural network, characterized by comprising:
a neural network training module that trains a neural network based on the architectural surface crack image, constructs a convolutional neural network model capable of identifying architectural surface cracks,
the building surface image acquisition module is used for acquiring images of the building surface to be detected;
and the crack identification module is used for carrying out crack identification processing on the to-be-detected building surface image acquired by the building surface image acquisition module by utilizing the convolutional neural network model constructed by the neural network training module.
8. The architectural surface crack recognition system of claim 7, wherein the convolutional neural network model trained by the neural network training module is capable of generating a plurality of tag values representing crack levels for the image processing to be processed.
9. The architectural surface crack recognition system of claim 7, wherein the crack recognition module comprises an image segmentation module, an input module, a decision module, and a merge module,
the image segmentation module is used for segmenting the surface image of the building to be detected into a plurality of sub-images for large-scale crack detection or into a plurality of sub-images for small-scale crack grading;
the input module is in data interaction with the image segmentation module, and a plurality of sub-images obtained by segmentation of the image segmentation module are led into a trained convolutional neural network model according to rows;
the judging module carries out crack detection recognition or crack grading recognition on the image of the surface of the building to be detected according to the output result of the convolutional neural network model, and marks the original image of the surface of the building to be detected according to the recognition result;
and the merging module performs data interaction with the judging module, merges the to-be-detected building surface images with the crack marks and the crack grade marks, and forms a building surface image with the crack grade marks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211478331.7A CN116563202A (en) | 2022-11-23 | 2022-11-23 | Building surface crack identification method and system based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211478331.7A CN116563202A (en) | 2022-11-23 | 2022-11-23 | Building surface crack identification method and system based on convolutional neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116563202A true CN116563202A (en) | 2023-08-08 |
Family
ID=87486718
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211478331.7A Pending CN116563202A (en) | 2022-11-23 | 2022-11-23 | Building surface crack identification method and system based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116563202A (en) |
-
2022
- 2022-11-23 CN CN202211478331.7A patent/CN116563202A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111612763B (en) | Mobile phone screen defect detection method, device and system, computer equipment and medium | |
| US11551341B2 (en) | Method and device for automatically drawing structural cracks and precisely measuring widths thereof | |
| CN111080622B (en) | Neural network training method, workpiece surface defect classification and detection method and device | |
| CN108346144B (en) | Automatic bridge crack monitoring and identifying method based on computer vision | |
| CN113436169B (en) | Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation | |
| CN111257341B (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN114994061B (en) | Machine vision-based steel rail intelligent detection method and system | |
| CN105574161B (en) | A kind of brand logo key element recognition methods, device and system | |
| CN106355579A (en) | Defect detecting method of cigarette carton surface wrinkles | |
| CN113284154B (en) | Steel coil end face image segmentation method and device and electronic equipment | |
| CN105223208B (en) | A kind of circuit board detecting template and preparation method thereof, circuit board detecting method | |
| CN111598942A (en) | Method and system for automatically positioning electric power facility instrument | |
| CN112381175A (en) | Circuit board identification and analysis method based on image processing | |
| CN111103307A (en) | Pcb defect detection method based on deep learning | |
| CN110503623A (en) | A method of Bird's Nest defect on the identification transmission line of electricity based on convolutional neural networks | |
| CN113780200A (en) | Computer vision-based pavement multi-disease area detection and positioning method | |
| CN115184380A (en) | Printed circuit board welding spot abnormity detection method based on machine vision | |
| CN110728269B (en) | High-speed rail contact net support pole number plate identification method based on C2 detection data | |
| CN114266881A (en) | Pointer type instrument automatic reading method based on improved semantic segmentation network | |
| CN113962951B (en) | Training method and device for detecting segmentation model, and target detection method and device | |
| CN113705564B (en) | Pointer type instrument identification reading method | |
| CN113962929A (en) | Photovoltaic cell assembly defect detection method and system and photovoltaic cell assembly production line | |
| CN114155226A (en) | Micro defect edge calculation method | |
| CN113657339A (en) | Instrument pointer counting and reading method and medium based on machine vision | |
| CN115937555A (en) | Industrial defect detection algorithm based on standardized flow model |
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 |