CN111353396A - Concrete crack segmentation method based on SCSEOCUnet - Google Patents
Concrete crack segmentation method based on SCSEOCUnet Download PDFInfo
- Publication number
- CN111353396A CN111353396A CN202010106077.2A CN202010106077A CN111353396A CN 111353396 A CN111353396 A CN 111353396A CN 202010106077 A CN202010106077 A CN 202010106077A CN 111353396 A CN111353396 A CN 111353396A
- Authority
- CN
- China
- Prior art keywords
- crack
- branch
- channel
- module
- input
- 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
- 238000000034 method Methods 0.000 title claims abstract description 26
- 230000011218 segmentation Effects 0.000 title claims abstract description 10
- 230000007246 mechanism Effects 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 14
- 238000012795 verification Methods 0.000 claims description 8
- 230000004913 activation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 3
- 238000013527 convolutional neural network Methods 0.000 abstract description 3
- 238000013473 artificial intelligence Methods 0.000 abstract description 2
- 230000003993 interaction Effects 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 6
- 238000001514 detection method Methods 0.000 description 4
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000003708 edge detection Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000010183 spectrum analysis Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007635 classification algorithm Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/35—Categorising the entire scene, e.g. birthday party or wedding scene
- G06V20/38—Outdoor scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a concrete crack segmentation method based on SCSE OC Unet, and belongs to the technical field of civil engineering and artificial intelligence computer vision interaction. The invention provides a method for extracting concrete apparent cracks by using an SCSEOCUnet convolutional neural network model based on a classical Unet network. The encoder of the SCSEOCUnet network adopts a pre-trained ResNet34 model capable of retaining more detailed information, and a space-channel attention mechanism SCSE module and a context inference OC module are fused to optimize the model, so that the crack information in the image is extracted with higher precision.
Description
Technical Field
The invention relates to the technical field of civil engineering and artificial intelligence computer vision interaction, in particular to a concrete crack segmentation method based on SCSEOCUnet.
Background
Common traditional digital image processing methods comprise edge detection, threshold value method, spectrum analysis method and the like, the methods are only effective on a data set under specific conditions, in an actual environment, crack detection is easily interfered by environmental factors such as fuzziness, shadow, leaves, scratches and the like, and the traditional methods have large detection errors and low model generalization capability due to the noises. Some scholars detect the bridge cracks based on the unmanned aerial vehicle, provide a new bridge crack classification algorithm and a target detection algorithm based on CNN, and obtain specific numerical values of the bridge crack characteristics by using a traditional image processing algorithm after the detection results are quickly positioned. However, none of these methods can extract crack information at the pixel level.
Disclosure of Invention
The invention aims to provide a concrete crack segmentation method capable of extracting crack information at a pixel level.
The technical scheme of the invention is as follows: a concrete crack segmentation method based on SCSEOCUnnet comprises the following steps:
s1: acquiring a picture of the surface of a bridge or the surface of a road with a crack by using an unmanned aerial vehicle;
s2: marking the obtained crack, wherein the pixel value of the crack area is marked as 1, and the pixel value of the non-crack area is marked as 0;
s3: dividing the marked pictures and the original pictures into a training set and a verification set according to the proportion of 0.85: 0.15;
s4: constructing a space-channel attention mechanism SCSE module, constructing a context inference OC module to form an SCSEOCUnet network, training the training set obtained in the step S3, verifying the verification set, and storing the trained model weight parameters;
s5: identifying and processing the newly shot picture by using the trained model, and segmenting a crack region if a crack exists; and if no crack exists, judging the picture to be a normal picture.
In a further technical scheme, in S4, an SCSE module of a space-channel attention mechanism is constructed, a context inference OC module is constructed to form an scseoconet network, the training set obtained in step S3 is trained, the verification set is verified, and trained model weight parameters are stored, wherein the method includes the following steps:
s4.1, a space-channel attention mechanism SCSE module is constructed, and after an input U enters a network, the input U enters two branches, wherein the upper branch is an SC branch, the branch is subjected to 1 × 1 convolution operation to obtain a weight matrix with the same length and width as the input U, and the matrix is multiplied by the U to obtain a characteristic U which is recalibrated in spaceSC(ii) a The lower branch is the SE branch which first passes through a maximum pooling operation to obtain a channelThe number of channels is the same as that of input U, the matrix passes through two full-connection layers, the number of neurons of the first full-connection layer is half of the number of channels, the number of the second full-connection layer is equal to that of the channels, the matrix reduced to the number of the channels by the second full-connection layer is multiplied by U through an activation function, and the characteristic U recalibrated in the channel direction is obtainedSEFinally, the features along the channel and spatial recalibration are combined to output USCSE；
S4.2, for the last layer output X of the network, reducing Channel from 2048 to 512 by convolution of 3 × 3 to be Y, calculating pixel-by-pixel Attention Map and ObjectContext in four branches by a Self-Attention module respectively, wherein the first branch takes all feature maps Y as input, the second branch divides the feature maps into sub-regions of 2 × 2, each sub-region applies shared transform, the third and fourth branches divide the input Y into sub-regions of 3 × 3 and 6 × 6, the transform of each sub-region is not shared, finally splicing the results of each branch along the Channel direction, and increasing the dimension of X by convolution of 1 × 1 to be equal to the dimension of Object Context;
s4.3, the Loss function is a Dice Loss function, and the calculation process is as follows:
wherein L isDiceIs a loss of Dice, yiIs the label value, P, of the ith pixel pointiThe term is set to 1 for the predicted probability value of the ith pixel.
The invention has the beneficial effects that:
1. the method can accurately identify the crack and segment the crack region, has self-adaptability compared with the traditional common traditional digital image processing methods such as edge detection, threshold value method and spectrum analysis method, and does not need to set different parameters according to the environment;
2. compared with the existing deep learning network, the method provided by the invention can effectively reduce the influence of unbalanced samples, improves the segmentation precision, is more stable and strong in robustness for complex environments, and can be used for other disease defects of buildings: such as in the division of the pitted surface of the honeycomb.
3. The invention provides a method for extracting concrete apparent cracks by using an SCSEOCUnet convolutional neural network model based on a classical Unet network. The encoder of the SCSEOCUnet network adopts a pre-trained ResNet34 model capable of retaining more detailed information, and a space-channel attention mechanism SCSE module and a context inference OC module are fused to optimize the model, so that the crack information in the image is extracted with higher precision.
Drawings
FIG. 1 is a schematic flow chart of the present invention,
fig. 2 is a diagram of the scseoconet network structure in the present invention.
Detailed Description
The invention will be further illustrated and understood by the following non-limiting examples.
As shown in fig. 1-2, the present invention provides a concrete crack segmentation method based on scseoconet, comprising the following steps:
s1: acquiring a picture of the surface of a bridge or the surface of a road with a crack by using an unmanned aerial vehicle;
s2: marking the obtained crack, wherein the pixel value of the crack area is marked as 1, and the pixel value of the non-crack area is marked as 0;
s3: dividing the marked pictures and the original pictures into a training set and a verification set according to the proportion of 0.85: 0.15;
s4: constructing a space-channel attention mechanism SCSE module, constructing a context inference OC module to form an SCSEOCUnnet network, training a training set obtained in the step S3, verifying the verification set, and storing trained model weight parameters, wherein the method specifically comprises the following steps:
s4.1, a space-channel attention mechanism SCSE module is constructed, and after an input U enters a network, the input U enters two branches, wherein the upper branch is an SC branch, the branch is subjected to 1 × 1 convolution operation to obtain a weight matrix with the same length and width as the input U, and the matrix is multiplied by the U to obtain a characteristic U which is recalibrated in spaceSC(ii) a The following branch is an SE branch, the branch firstly obtains a weight matrix with the same channel number as the input U through a maximum pooling operation, the matrix passes through two full-connection layers, the number of neurons of the first full-connection layer is half of the channel number, the number of the second full-connection layer is equal to the channel number, the matrix which is reduced to the channel number through the second full-connection layer is multiplied by the U through an activation function, and the characteristic U which is recalibrated in the channel direction is obtainedSEFinally, the features along the channel and spatial recalibration are combined to output USCSE；
S4.2, for the last layer output X of the network, reducing Channel from 2048 to 512 by convolution of 3 × 3 to be Y, calculating pixel-by-pixel Attention Map and ObjectContext in four branches by a Self-Attention module respectively, wherein the first branch takes all feature maps Y as input, the second branch divides the feature maps into sub-regions of 2 × 2, each sub-region applies shared transform, the third and fourth branches divide the input Y into sub-regions of 3 × 3 and 6 × 6, the transform of each sub-region is not shared, finally splicing the results of each branch along the Channel direction, and increasing the dimension of X by convolution of 1 × 1 to be equal to the dimension of Object Context;
s4.3, the Loss function is a Dice Loss function, and the calculation process is as follows:
wherein L isDiceIs a loss of Dice, yiIs the label value, P, of the ith pixel pointiSetting the item as 1 for the prediction probability value of the ith pixel point;
s5: identifying and processing the newly shot picture by using the trained model, and segmenting a crack region if a crack exists; and if no crack exists, judging the picture to be a normal picture.
Claims (2)
1. A concrete crack segmentation method based on SCSEOCUnnet is characterized by comprising the following steps:
s1: acquiring a picture of the surface of a bridge or the surface of a road with a crack by using an unmanned aerial vehicle;
s2: marking the obtained crack, wherein the pixel value of the crack area is marked as 1, and the pixel value of the non-crack area is marked as 0;
s3: dividing the marked pictures and the original pictures into a training set and a verification set according to the proportion of 0.85: 0.15;
s4: constructing a space-channel attention mechanism SCSE module, constructing a context inference OC module to form an SCSEOCUnet network, training the training set obtained in the step S3, verifying the verification set, and storing the trained model weight parameters;
s5: identifying and processing the newly shot picture by using the trained model, and segmenting a crack region if a crack exists; and if no crack exists, judging the picture to be a normal picture.
2. The SCSEOCUnnet-based concrete crack segmentation method as claimed in claim 1, wherein in S4, a space-channel attention mechanism SCSE module is constructed, a context inference OC module is constructed to form an SCSEOCUnnet network, a training set obtained in step S3 is trained, the verification set is verified, and trained model weight parameters are stored, wherein the method comprises the following steps:
s4.1, a space-channel attention mechanism SCSE module is constructed, and after an input U enters a network, the input U enters two branches, wherein the upper branch is an SC branch, the branch is subjected to 1 × 1 convolution operation to obtain a weight matrix with the same length and width as the input U, and the matrix is multiplied by the U to obtain a characteristic U which is recalibrated in spaceSC(ii) a The following branch is an SE branch, the branch firstly obtains a weight matrix with the same channel number as the input U through a maximum pooling operation, the matrix passes through two full-connection layers, the number of neurons of the first full-connection layer is half of the channel number, the number of the second full-connection layer is equal to the channel number, the matrix which is reduced to the channel number through the second full-connection layer is multiplied by the U through an activation function, and the characteristic U which is recalibrated in the channel direction is obtainedSEFinally, the features along the channel and spatial recalibration are combined to output USCSE；
S4.2, for the last layer output X of the network, reducing Channel from 2048 to 512 by convolution of 3 × 3 to be Y, calculating pixel-by-pixel Attention Map and ObjectContext in four branches by a Self-Attention module respectively, wherein the first branch takes all feature maps Y as input, the second branch divides the feature maps into sub-regions of 2 × 2, each sub-region applies shared transform, the third and fourth branches divide the input Y into sub-regions of 3 × 3 and 6 × 6, the transform of each sub-region is not shared, finally splicing the results of each branch along the Channel direction, and increasing the dimension of X by convolution of 1 × 1 to be equal to the dimension of Object Context;
s4.3, the Loss function is a Dice Loss function, and the calculation process is as follows:
wherein L isDiceIs a loss of Dice, yiIs the label value, P, of the ith pixel pointiThe term is set to 1 for the predicted probability value of the ith pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010106077.2A CN111353396A (en) | 2020-02-20 | 2020-02-20 | Concrete crack segmentation method based on SCSEOCUnet |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010106077.2A CN111353396A (en) | 2020-02-20 | 2020-02-20 | Concrete crack segmentation method based on SCSEOCUnet |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111353396A true CN111353396A (en) | 2020-06-30 |
Family
ID=71192339
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010106077.2A Pending CN111353396A (en) | 2020-02-20 | 2020-02-20 | Concrete crack segmentation method based on SCSEOCUnet |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111353396A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112418229A (en) * | 2020-11-03 | 2021-02-26 | 上海交通大学 | Unmanned ship marine scene image real-time segmentation method based on deep learning |
| CN112489001A (en) * | 2020-11-23 | 2021-03-12 | 石家庄铁路职业技术学院 | Tunnel water seepage detection method based on improved deep learning |
| CN112927240A (en) * | 2021-03-08 | 2021-06-08 | 重庆邮电大学 | CT image segmentation method based on improved AU-Net network |
| CN114596266A (en) * | 2022-02-25 | 2022-06-07 | 烟台大学 | Concrete crack detection method based on ConcreteCrackSegNet model |
| CN115471734A (en) * | 2022-09-23 | 2022-12-13 | 中国农业大学 | Method, device and server for identifying wooden package IPPC (Internet protocol personal computer) identification |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110208859A (en) * | 2019-05-07 | 2019-09-06 | 长江大学 | Oil-base mud well crack quantitative parameter intelligence computation method based on ultrasonic imaging |
| CN110415233A (en) * | 2019-07-26 | 2019-11-05 | 东南大学 | Pavement crack rapid extracting method based on two step convolutional neural networks |
| CN110472666A (en) * | 2019-07-18 | 2019-11-19 | 广东工业大学 | A kind of distress in concrete recognition methods based on convolutional neural networks |
-
2020
- 2020-02-20 CN CN202010106077.2A patent/CN111353396A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110208859A (en) * | 2019-05-07 | 2019-09-06 | 长江大学 | Oil-base mud well crack quantitative parameter intelligence computation method based on ultrasonic imaging |
| CN110472666A (en) * | 2019-07-18 | 2019-11-19 | 广东工业大学 | A kind of distress in concrete recognition methods based on convolutional neural networks |
| CN110415233A (en) * | 2019-07-26 | 2019-11-05 | 东南大学 | Pavement crack rapid extracting method based on two step convolutional neural networks |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112418229A (en) * | 2020-11-03 | 2021-02-26 | 上海交通大学 | Unmanned ship marine scene image real-time segmentation method based on deep learning |
| CN112489001A (en) * | 2020-11-23 | 2021-03-12 | 石家庄铁路职业技术学院 | Tunnel water seepage detection method based on improved deep learning |
| CN112489001B (en) * | 2020-11-23 | 2023-07-25 | 石家庄铁路职业技术学院 | Tunnel water seepage detection method based on improved deep learning |
| CN112927240A (en) * | 2021-03-08 | 2021-06-08 | 重庆邮电大学 | CT image segmentation method based on improved AU-Net network |
| CN112927240B (en) * | 2021-03-08 | 2022-04-05 | 重庆邮电大学 | CT image segmentation method based on improved AU-Net network |
| CN114596266A (en) * | 2022-02-25 | 2022-06-07 | 烟台大学 | Concrete crack detection method based on ConcreteCrackSegNet model |
| CN115471734A (en) * | 2022-09-23 | 2022-12-13 | 中国农业大学 | Method, device and server for identifying wooden package IPPC (Internet protocol personal computer) identification |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN111353396A (en) | Concrete crack segmentation method based on SCSEOCUnet | |
| CN110427937B (en) | Inclined license plate correction and indefinite-length license plate identification method based on deep learning | |
| CN108121991B (en) | Deep learning ship target detection method based on edge candidate region extraction | |
| CN110569738B (en) | Natural scene text detection method, equipment and medium based on densely connected network | |
| CN111325203A (en) | American license plate recognition method and system based on image correction | |
| CN110796009A (en) | Method and system for detecting marine vessel based on multi-scale convolution neural network model | |
| CN111257341B (en) | Underwater building crack detection method based on multi-scale features and stacked full convolution network | |
| CN112200143A (en) | Road disease detection method based on candidate area network and machine vision | |
| CN110717493B (en) | License plate recognition method containing stacked characters based on deep learning | |
| CN113435407B (en) | Small target identification method and device for power transmission system | |
| CN112232351A (en) | License plate recognition system based on deep neural network | |
| CN113989604B (en) | Tire DOT information identification method based on end-to-end deep learning | |
| CN110909657A (en) | Method for identifying apparent tunnel disease image | |
| CN112766170B (en) | Self-adaptive segmentation detection method and device based on cluster unmanned aerial vehicle image | |
| CN110717886A (en) | Pavement pool detection method based on machine vision in complex environment | |
| CN113326846B (en) | Rapid bridge apparent disease detection method based on machine vision | |
| CN109325407B (en) | Optical remote sensing video target detection method based on F-SSD network filtering | |
| CN111985552A (en) | Method for detecting diseases of thin strip-shaped structure of airport pavement under complex background | |
| CN113159215A (en) | Small target detection and identification method based on fast Rcnn | |
| CN116206112A (en) | Remote sensing image semantic segmentation method based on multi-scale feature fusion and SAM | |
| CN116342536A (en) | Aluminum strip surface defect detection method, system and equipment based on lightweight model | |
| CN110991374B (en) | Fingerprint singular point detection method based on RCNN | |
| CN113610024B (en) | Multi-strategy deep learning remote sensing image small target detection method | |
| CN109615610B (en) | Medical band-aid flaw detection method based on YOLO v2-tiny |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20200630 |
|
| WD01 | Invention patent application deemed withdrawn after publication |