CN109146849A - A kind of road surface crack detection method based on convolutional neural networks and image recognition - Google Patents
A kind of road surface crack detection method based on convolutional neural networks and image recognition Download PDFInfo
- Publication number
- CN109146849A CN109146849A CN201810834095.5A CN201810834095A CN109146849A CN 109146849 A CN109146849 A CN 109146849A CN 201810834095 A CN201810834095 A CN 201810834095A CN 109146849 A CN109146849 A CN 109146849A
- Authority
- CN
- China
- Prior art keywords
- image
- road surface
- neural networks
- convolutional neural
- network
- 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 41
- 238000001514 detection method Methods 0.000 title claims abstract description 35
- 238000000034 method Methods 0.000 claims abstract description 29
- 238000013528 artificial neural network Methods 0.000 claims abstract description 27
- 230000006870 function Effects 0.000 claims abstract description 20
- 230000000306 recurrent effect Effects 0.000 claims abstract description 18
- 238000010606 normalization Methods 0.000 claims abstract description 4
- 239000010410 layer Substances 0.000 claims description 42
- 230000007547 defect Effects 0.000 claims description 34
- 238000012549 training Methods 0.000 claims description 28
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000002950 deficient Effects 0.000 claims description 6
- 238000013461 design Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 3
- 238000005286 illumination Methods 0.000 claims description 3
- 239000011229 interlayer Substances 0.000 claims description 3
- 238000012804 iterative process Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 230000000644 propagated effect Effects 0.000 claims description 3
- 230000003252 repetitive effect Effects 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 4
- 238000010586 diagram Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 230000008439 repair process Effects 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
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of road surface crack detection method based on convolutional neural networks and image recognition, belongs to traffic detection field.This method is using convolutional neural networks chip and a recurrent neural network as core (convolutional neural networks chip is used to detect pavement crack, and recurrent neural network is used to generate the sentence of description types of fractures).Include: that road pavement crack original image carries out preliminary making, intensity normalization is carried out according to preliminary making result road pavement image and the pretreatment of pixel saturation, pretreated pavement image is input in convolutional neural networks (CNN) model is trained and (a variety of different road surfaces is used to be trained as sample to neural network), determine network structure and model parameter, trained network is used to detect complex road surface, and determination pavement crack type.Using road surface image modalities, image preprocess apparatus, memory, USB external device, microprocessor etc., the pavement crack detection system with deep learning function is constituted.
Description
Technical field
The present invention relates to a kind of road surface crack detection method based on convolutional neural networks and image recognition, belongs to traffic road
Face detection field.
Background technique
With the rapid development of highway transportation industry, the maintenance work of highway pavement is also increasingly heavy, highway administration portion
Door needs quickly to grasp highway pavement information in time.Traditional artificial detection method early has been unable to meet the basic of Road Development
Demand.Meanwhile face large-scale road surface breakage image, it is traditional based on the method for image recognition due to being made an uproar by pavement image
The influence for the factors such as sound, feature extracting method limit to, image data amount is big, is also no longer satisfied the application of large-scale data.
Currently, there are following problems for pavement detection: 1. pavement images 2. that can only be handled under identical conditions a variety of ought make an uproar
When sound is mixed in together, the effect of Crack Detection is unobvious.3. it is good to simple pavement crack detection effect, it is multiple for being distributed
Miscellaneous Crack Detection result fracture is serious.4. tiny for texture and dispersion pavement crack testing result is bad.5. cannot have
Effect judges road surface defect type.In order to improve the accuracy rate of Crack Detection, method proposes based on CNN detection pavement image
Crack.
Summary of the invention
The technical problem to be solved in the present invention is to provide a kind of pavement crack based on convolutional neural networks and image recognition
Detection method can extract reflection number using this deep learning method of convolutional neural networks from a large amount of training sample
According to the recessive character of essence, and it is able to detect pavement crack type, facilitates relevant departments that appropriate action repairing road surface is taken to split
Seam.The invention is safe and reliable, and detection efficiency is high, and operating cost is low.
The technical solution adopted by the present invention is that: a kind of pavement crack detection side based on convolutional neural networks and image recognition
Method, steps are as follows:
A, pavement crack picture is acquired using image modalities, and picture is pre-processed；
B, it designs convolutional neural networks structure and is trained using processed picture in step A；
C, it acquires road surface picture and judges road surface with the presence or absence of defect using trained neural network in step B；
If D, road surface existing defects, the picture that defect will be present is input in the multi-modal Recognition with Recurrent Neural Network trained,
The text of description road surface defect is obtained, so that it is determined that types of fractures.
Specifically, the image preprocessing process in the step A includes:
Firstly, carry out preliminary making to the picture of acquisition, then to the pavement image after preliminary making carry out intensity normalization and
The pretreatment of pixel saturation, to reduce the influence of uneven illumination etc., wherein the method for preliminary making is: by 2400 pixels × 3150 pictures
The pavement image of element is divided into 150 pixels × 150 pixel image blocks of 16 rows, 21 column, calculates the mean value and mark of each image block
It is quasi- poor, in this way, every width pavement image has 16 × 21 Mean Matrix MmWith standard deviation matrix STDm, then to MmMatrix difference
The image block that vertical and horizontal scanning search includes crack pixel is done, the purpose of preliminary making is the image block that will include crack pixel
Labeled as ' 1 ', the image block not comprising crack pixel is labeled as " 0 ".
Specifically, the design of the convolutional neural networks structure in the step B includes: with training process
B1. there are four convolutional layers, whole network to all employ an equal amount of 3*3 convolution kernel and 2*2 most for the network structure
Great Chiization；
B2. the training of network is divided into two stages: propagated forward and backpropagation；
The training step of CNN is as follows:
Step 1: net is inputted as training group by collected crack image and the corresponding ground truth image of image
Network；
Step 2: setting network weight and being biased to the random value close to 0, and by the parameter learning rate α of precision controlling
Setting
It is 0.001；
Step 3: it takes 1 sample to add to network from training set, then calculates its target output vector；
Step 4: middle layer output vector is calculated；
The error that step 5: calculating output vector and object vector is exported in middle layer；
Step 6: the sum of each inter-layer prediction result in output vector is calculated；
Step 7: the output vector of last fused layer and the output error of object vector are calculated；
Step 8: objective function is minimized using stochastic gradient descent；
Step 9: constantly updating weight and bias term, optimizes network model, and determines corresponding defect setting score.
Specifically, judge whether road surface is defective in the step C and include:
Pretreated picture to be measured is substituted into the network model that step B is determined, operation is carried out to convolutional neural networks,
If defect score, which is greater than defect, sets score, then it is assumed that road surface is defective, if the defect score is less than or equal to defect setting point
Number, then it is assumed that zero defect.
Specifically, the training of the multi-modal Recognition with Recurrent Neural Network in the step D needs to collect a large amount of pavement cracks and retouches
State the set of reflection crack type text；In the training stage, the feature of Recognition with Recurrent Neural Network input picture I and corresponding text sequence
Sequence vector (x1,...,xT), hidden state is then calculated by the iterative process (t=1 to T) of repetitive (2) and formula (3)
Sequence (h1,...,ht) and output value sequence (y1,...,yt)；
ht=f (Whxxt+Whhht-1+bh+ ∏ (t=1) ⊙ bv) (2)
yt=softmax (Wohht+bo) (3)
Wherein,For the output of neural network the last layer, bvFor image information, Whi, Whx, WhhAnd WohFor instruction
Weight during white silk, boAnd bhFor the biasing in training process.
Specifically, the output vector of the middle layer in the step 4 can be calculated according to formula (4)
Wherein WjkThe weight for needing to learn for j layers to k layer network；FjFor the characteristic pattern for being input to k layers；θkFor bias term.
Specifically, the error in the step 5 can be calculated according to formula (5)
Wherein αmFor adjustment factor；lsideIt is the loss function of intermediate tomographic image；W is the network layer parameter of all standards；w(m)For each layer of weight, it is represented by w=[w(1),···,w(M)]。
Specifically, the sum of prediction result in the step 6 can be calculated according to formula (6)
Wherein,For the weight merged to middle layer,σ () is sigmoid function,
The error exported for output vector and object vector in middle layer.
Specifically, the output vector in the step 7 and the output error of object vector can be calculated according to formula (7)
Wherein,For fusion forecasting figureThe distance between groundtruth label figure Y, by making
Penalty values size is calculated with cross entropy loss function.
Specifically, the minimum objective function in the step 8 can be calculated according to formula (8)
Wherein, the variate-value that arg min () is available when being minimized objective function, Lside(S, Y, W, w) is
The loss of each middle layer,For the loss of final prognostic chart Yu ground truth label figure.
Specifically, the continuous renewal weight in the step 9 and biasing can be carried out according to formula (9) and (10)
Wherein,WithThe weight not updated after respectively updating and,WithIt does not update after respectively updating and
Biasing, η is learning rate.
The beneficial effects of the present invention are: this method is safe and reliable, detection speed is fast, and detection efficiency is high, and rate of false alarm is low, will not
There is night work fatigue phenomenon, and can accurately determine highway defect type, aspect relevant departments take in time rationally to be arranged
Apply repairing road surface defect.
Detailed description of the invention
Fig. 1 is a kind of identification implemented based on convolutional neural networks and the road surface crack detection method of image recognition of the present invention
Structural block diagram.
Fig. 2 is the network frame figure of convolutional neural networks.
Fig. 3 is the PR curve assessed testing result.
Fig. 4 is that multi-mode recurrent neural network generates model.
Fig. 5 is pavement crack detection system block diagram.
Specific embodiment
With reference to the accompanying drawings and detailed description, the present invention is further described.
Embodiment 1: a kind of road surface crack detection method based on convolutional neural networks and image recognition, steps are as follows:
A, pavement crack picture is acquired using image modalities, and picture is pre-processed；
B, it designs convolutional neural networks structure and is trained using processed picture in step A and (use various differences
Pavement crack neural network is trained as sample)；
C, it acquires road surface picture and judges road surface with the presence or absence of defect using trained neural network in step B；
If D, road surface existing defects, the picture that defect will be present is input in the multi-modal Recognition with Recurrent Neural Network trained,
The text of description road surface defect is obtained, so that it is determined that types of fractures.
Further, the image preprocessing process in the step A includes:
Firstly, carry out preliminary making to the picture of acquisition, then to the pavement image after preliminary making carry out intensity normalization and
The pretreatment of pixel saturation, to reduce the influence of uneven illumination etc., wherein the method for preliminary making is: by 2400 pixels × 3150 pictures
The pavement image of element is divided into 150 pixels × 150 pixel image blocks of 16 rows, 21 column, calculates the mean value and mark of each image block
It is quasi- poor, in this way, every width pavement image has 16 × 21 Mean Matrix MmWith standard deviation matrix STDm, then to MmMatrix difference
The image block that vertical and horizontal scanning search includes crack pixel is done, the purpose of preliminary making is the image block that will include crack pixel
Labeled as ' 1 ', the image block not comprising crack pixel is labeled as " 0 ".
Further, the design of the convolutional neural networks structure in the step B includes: with training process
B1. there are four convolutional layers, whole network to all employ an equal amount of 3*3 convolution kernel and 2*2 most for the network structure
Great Chiization；
B2. the training of network is divided into two stages: propagated forward and backpropagation；
The training step of CNN is as follows:
Step 1: net is inputted as training group by collected crack image and the corresponding groundtruth image of image
Network；
Step 2: setting network weight and being biased to the random value close to 0, and by the parameter learning rate α of precision controlling
It is set as 0.001；
Step 3: it takes 1 sample to add to network from training set, then calculates its target output vector；
Step 4: middle layer output vector is calculated；
The error that step 5: calculating output vector and object vector is exported in middle layer；
Step 6: the sum of each inter-layer prediction result in output vector is calculated；
Step 7: the output vector of last fused layer and the output error of object vector are calculated；
Step 8: objective function is minimized using stochastic gradient descent；
Step 9: constantly updating weight and bias term, optimizes network model, and determines corresponding defect setting score.
Further, judge whether road surface is defective in the step C and include:
Pretreated picture to be measured is substituted into the network model that step B is determined, operation is carried out to convolutional neural networks,
If defect score, which is greater than defect, sets score, then it is assumed that road surface is defective, if the defect score is less than or equal to defect setting point
Number, then it is assumed that zero defect.
Further, the training of the multi-modal Recognition with Recurrent Neural Network in the step D need to collect a large amount of pavement cracks and
The set of reflection crack type text is described；In the training stage, the spy of Recognition with Recurrent Neural Network input picture I and corresponding text sequence
Levy sequence vector (x1,...,xT), hiding shape is then calculated by the iterative process (t=1 to T) of repetitive (2) and formula (3)
State sequence (h1,...,ht) and output value sequence (y1,...,yt)；
ht=f (Whxxt+Whhht-1+bh+ ∏ (t=1) ⊙ bv) (2)
yt=softmax (Wohht+bo) (3)
Wherein,For the output of neural network the last layer, bvFor image information, Whi, Whx, WhhAnd WohFor instruction
Weight during white silk, boAnd bhFor the biasing in training process.
Specifically, the output vector of the middle layer in the step 4 can be calculated according to formula (4)
Wherein WjkThe weight for needing to learn for j layers to k layer network；FjFor the characteristic pattern for being input to k layers；θkFor bias term.
Further, the error in the step 5 can be calculated according to formula (5)
Wherein αmFor adjustment factor；lsideIt is the loss function of intermediate tomographic image；W is the network layer parameter of all standards；w(m)For each layer of weight, it is represented by w=[w(1),···,w(M)]。
Further, the sum of prediction result in the step 6 can be calculated according to formula (6)
Wherein,For the weight merged to middle layer,σ () is sigmoid function,The error exported for output vector and object vector in middle layer.
Further, the output vector in the step 7 and the output error of object vector can be calculated according to formula (7)
Wherein,For fusion forecasting figureThe distance between groundtruth label figure Y, by using
Cross entropy loss function calculates penalty values size.
Further, the minimum objective function in the step 8 can be calculated according to formula (8)
Wherein, the variate-value that arg min () is available when being minimized objective function, Lside(S, Y, W, w) is
The loss of each middle layer,For the loss of final prognostic chart Yu ground truth label figure.
Further, the continuous renewal weight in the step 9 and biasing can be carried out according to formula (9) and (10)
Wherein,WithThe weight not updated after respectively updating and,WithIt does not update after respectively updating and
Biasing, η is learning rate.
As shown in Fig. 2, the neural network network frame figure of this method.There are four convolutional layer, entire nets for the network structure
Network all employs an equal amount of 3*3 convolution kernel and 2*2 maximum pond network structure.
It as shown in Fig. 3, is the PR curve graph assessed testing result.It can intuitively show convolutional Neural net
Recall ratio and precision ratio of the network in sample generally, if a PR curve surrounds another curve, illustrate the former performance compared with
It is good.
It as shown in Fig. 4, is that multi-mode recurrent neural network generates model, for generating the text envelope of description road surface defect
Breath, so that it is determined that defect type.It is corresponding with description that the training that recurrent neural network generates model needs to collect a large amount of pavement cracks
The set of types of fractures text.
Further, the training process of the Recognition with Recurrent Neural Network, RNN is according to a word (xt) and before context letter
Cease (ht-1) predict next word (yt).The trained first step is by introducing image information (bv) predicted value is had an impact.Instruction
Practice process and refer to attached drawing 4: setting h0For 0 vector, x1For a specific START vector, it is desirable that labely1Sequence is corresponded to
First word, and so on.In final step, XTIndicate the last one word, target labels yTThen it is set as one specifically
END mark.
It as shown in Fig. 5, is pavement crack detection system block diagram.With convolutional neural networks chip and multi-modal recurrent neural
Network is core, utilizes road surface image modalities, image preprocess apparatus, USB external device, memory, microprocessor, loudspeaker
With power supply etc., the pavement crack detection system with deep learning function is constituted.Wherein, USB external device is for placing mind
Through network chip.
Using convolutional neural networks chip and a recurrent neural network, as core, (convolutional neural networks chip is used to this method
Pavement crack is detected, recurrent neural network is used to generate the sentence of description types of fractures), there is small power consumption, integrated level is high, calculates
Speed is fast, configures the features such as flexible.The present invention uses this deep learning method of convolutional neural networks, can be from a large amount of training
The recessive character of reflection data essence is extracted in sample, and is able to detect pavement crack type, and relevant departments is facilitated to take
Appropriate action repairs pavement crack.The invention is safe and reliable, and detection efficiency is high, and operating cost is low.
Finally, it should be noted that the above content is only to illustrate the technical solution of this method, rather than the present invention is protected
The limitation of range, although the specific embodiment part explains this method in detail, those skilled in the art
It should be appreciated that can be modified or replaced equivalently to the technical solution of this method, without departing from this method technical solution
Spirit and scope.
Claims (10)
1. a kind of road surface crack detection method based on convolutional neural networks and image recognition, it is characterised in that: steps are as follows:
A, pavement crack picture is acquired using image modalities, and picture is pre-processed；
B, it designs convolutional neural networks structure and is trained using processed picture in step A；
C, it acquires road surface picture and judges road surface with the presence or absence of defect using trained neural network in step B；
If D, road surface existing defects, the picture that defect will be present is input in the multi-modal Recognition with Recurrent Neural Network trained, and is obtained
The text for describing road surface defect, so that it is determined that types of fractures.
2. the road surface crack detection method according to claim 1 based on convolutional neural networks and image recognition, feature
Be: the image preprocessing process in the step A includes:
Firstly, the picture to acquisition carries out preliminary making, intensity normalization and pixel then are carried out to the pavement image after preliminary making
Saturation pretreatment, to reduce the influence of uneven illumination etc., wherein the method for preliminary making is: by 2400 pixels × 3150 pixels
Pavement image is divided into 150 pixels × 150 pixel image blocks of 16 rows, 21 column, calculates the mean value and standard deviation of each image block,
In this way, every width pavement image has 16 × 21 Mean Matrix MmWith standard deviation matrix STDm, then to matrix MmLongitudinal direction is done respectively
The image block comprising crack pixel is searched with transversal scanning, the purpose of preliminary making is to be labeled as the image block comprising crack pixel
' 1 ', the image block not comprising crack pixel is labeled as " 0 ".
3. the road surface crack detection method according to claim 1 based on convolutional neural networks and image recognition, feature
Be: the design of the convolutional neural networks structure in the step B includes: with training process
B1. there are four convolutional layers, whole network to all employ an equal amount of 3*3 convolution kernel and 2*2 maximum pond for the network structure
Change；
B2. the training of network is divided into two stages: propagated forward and backpropagation；
The training step of CNN is as follows:
Step 1: network is inputted as training group by collected crack image and the corresponding ground truth image of image；
Step 2: being arranged network weight and is biased to the random value close to 0, and the parameter learning rate α of precision controlling is arranged
It is 0.001；
Step 3: it takes 1 sample to add to network from training set, then calculates its target output vector；
Step 4: middle layer output vector is calculated；
The error that step 5: calculating output vector and object vector is exported in middle layer；
Step 6: the sum of each inter-layer prediction result in output vector is calculated；
Step 7: the output vector of last fused layer and the output error of object vector are calculated；
Step 8: objective function is minimized using stochastic gradient descent；
Step 9: constantly updating weight and bias term, optimizes network model, and determines corresponding defect setting score.
4. the road surface crack detection method according to claim 1 based on convolutional neural networks and image recognition, feature
It is: judges whether road surface is defective in the step C and include:
Pretreated picture to be measured is substituted into the network model that step B is determined, operation is carried out to convolutional neural networks, if lacking
It falling into score and is greater than defect setting score, then it is assumed that road surface is defective, if the defect score is less than or equal to defect and sets score,
Think zero defect.
5. the road surface crack detection method according to claim 1 based on convolutional neural networks and image recognition, feature
Be: the training of the multi-modal Recognition with Recurrent Neural Network in the step D needs to collect a large amount of pavement cracks and description reflection crack
The set of type text；In the training stage, the characteristic vector sequence of Recognition with Recurrent Neural Network input picture I and corresponding text sequence
(x1,...,xT), hidden state sequence is then calculated by the iterative process (t=1 to T) of repetitive (2) and formula (3)
(h1,...,ht) and output value sequence (y1,...,yt)；
ht=f (Whxxt+Whhht-1+bh+ ∏ (t=1) ⊙ bv) (2)
yt=softmax (Wohht+bo) (3)
Wherein,For the output of neural network the last layer, bvFor image information, Whi, Whx, WhhAnd WohTo train
Weight in journey, boAnd bhFor the biasing in training process.
6. the road surface crack detection method according to claim 3 based on convolutional neural networks and image recognition, feature
Be: the output vector of the middle layer in the step 4 can be calculated according to formula (4)
Wherein WjkThe weight for needing to learn for j layers to k layer network；FjFor the characteristic pattern for being input to k layers；θkFor bias term.
7. the road surface crack detection method according to claim 3 based on convolutional neural networks and image recognition, feature
Be: the error in the step 5 can be calculated according to formula (5)
Wherein αmFor adjustment factor；lsideIt is the loss function of intermediate tomographic image；W is the network layer parameter of all standards；w(m)For
Each layer of weight is represented by w=[w(1),…,w(M)]。
8. the road surface crack detection method according to claim 3 based on convolutional neural networks and image recognition, feature
Be: the sum of prediction result in the step 6 can be calculated according to formula (6)
Wherein,For the weight merged to middle layer,σ () is sigmoid function,It is defeated
The error that outgoing vector and object vector are exported in middle layer.
9. the road surface crack detection method according to claim 3 based on convolutional neural networks and image recognition, feature
Be: the output error of output vector and object vector in the step 7 can be calculated according to formula (7)
Wherein,For fusion forecasting figureThe distance between groundtruth label figure Y, by using intersection
Entropy loss function calculates penalty values size.
10. the road surface crack detection method according to claim 3 based on convolutional neural networks and image recognition, feature
Be: the minimum objective function in the step 8 can be calculated according to formula (8)
Wherein, the variate-value that arg min () is available when being minimized objective function, Lside(S, Y, W, w) is in each
The loss of interbed,For the loss of final prognostic chart Yu groundtruth label figure.
Continuous renewal weight and biasing in the step 9 can be carried out according to formula (9) and (10)
Wherein,WithThe weight not updated after respectively updating and,WithIt does not update after respectively updating and
Biasing, η is learning rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810834095.5A CN109146849A (en) | 2018-07-26 | 2018-07-26 | A kind of road surface crack detection method based on convolutional neural networks and image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810834095.5A CN109146849A (en) | 2018-07-26 | 2018-07-26 | A kind of road surface crack detection method based on convolutional neural networks and image recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109146849A true CN109146849A (en) | 2019-01-04 |
Family
ID=64797921
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810834095.5A Pending CN109146849A (en) | 2018-07-26 | 2018-07-26 | A kind of road surface crack detection method based on convolutional neural networks and image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109146849A (en) |
Cited By (24)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872318A (en) * | 2019-02-22 | 2019-06-11 | 中国石油大学(华东) | A kind of geology for deep learning is appeared crack data set production method |
| CN109949290A (en) * | 2019-03-18 | 2019-06-28 | 北京邮电大学 | Pavement crack detection method, device, equipment and storage medium |
| CN109961030A (en) * | 2019-03-18 | 2019-07-02 | 北京邮电大学 | Pavement patching information detecting method, device, equipment and storage medium |
| CN109978141A (en) * | 2019-03-28 | 2019-07-05 | 腾讯科技（深圳）有限公司 | Neural network model training method and device, natural language processing method and apparatus |
| CN110020652A (en) * | 2019-01-07 | 2019-07-16 | 新而锐电子科技(上海)有限公司 | The dividing method of Tunnel Lining Cracks image |
| CN110046656A (en) * | 2019-03-28 | 2019-07-23 | 南京邮电大学 | Multi-modal scene recognition method based on deep learning |
| CN110059804A (en) * | 2019-04-15 | 2019-07-26 | 北京迈格威科技有限公司 | Network training method, data processing method and device to be searched |
| CN110070520A (en) * | 2019-03-19 | 2019-07-30 | 长安大学 | The building of pavement crack detection model and detection method based on deep neural network |
| CN110300257A (en) * | 2019-06-13 | 2019-10-01 | 广东技术师范大学天河学院 | A kind of face tracking safety defense monitoring system and its application method based on deep learning |
| CN110349119A (en) * | 2019-05-27 | 2019-10-18 | 北京邮电大学 | Pavement disease detection method and device based on edge detection neural network |
| CN110503637A (en) * | 2019-08-13 | 2019-11-26 | 中山大学 | A kind of crack on road automatic testing method based on convolutional neural networks |
| CN110738131A (en) * | 2019-09-20 | 2020-01-31 | 广州游艺云物联网技术有限公司 | Garbage classification management method and device based on deep learning neural network |
| CN110992314A (en) * | 2019-11-15 | 2020-04-10 | 广东华路交通科技有限公司 | Pavement defect detection method and device and storage medium |
| CN111126505A (en) * | 2019-12-28 | 2020-05-08 | 北京工业大学 | Pavement crack rapid identification method based on deep learning |
| CN111222546A (en) * | 2019-12-27 | 2020-06-02 | 中国科学院计算技术研究所 | Multi-scale fusion food image classification model training and image classification method |
| CN111583229A (en) * | 2020-05-09 | 2020-08-25 | 江苏野马软件科技有限公司 | Road surface fault detection method based on convolutional neural network |
| CN111862042A (en) * | 2020-07-21 | 2020-10-30 | 北京航空航天大学 | Pipeline contour detection method based on full convolution neural network |
| WO2020248371A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技（深圳）有限公司 | Road damage detection method and apparatus, computer device, and storage medium |
| CN112986950A (en) * | 2020-12-25 | 2021-06-18 | 南京理工大学 | Single-pulse laser radar echo feature extraction method based on deep learning |
| CN113015887A (en) * | 2019-10-15 | 2021-06-22 | 谷歌有限责任公司 | Navigation directions based on weather and road surface type |
| CN114049356A (en) * | 2022-01-17 | 2022-02-15 | 湖南大学 | Method, device and system for detecting structure apparent crack |
| CN115984221A (en) * | 2023-01-03 | 2023-04-18 | 广州新粤交通技术有限公司 | Road marking repairing and identifying method, device, equipment and storage medium thereof |
| CN116664582A (en) * | 2023-08-02 | 2023-08-29 | 四川公路桥梁建设集团有限公司 | Road surface detection method and device based on neural vision network |
| CN117115147A (en) * | 2023-10-19 | 2023-11-24 | 山东华盛创新纺织科技有限公司 | Textile detection method and system based on machine vision |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107609460A (en) * | 2017-05-24 | 2018-01-19 | 南京邮电大学 | A kind of Human bodys' response method for merging space-time dual-network stream and attention mechanism |
| CN107862331A (en) * | 2017-10-31 | 2018-03-30 | 华中科技大学 | It is a kind of based on time series and CNN unsafe acts recognition methods and system |
-
2018
- 2018-07-26 CN CN201810834095.5A patent/CN109146849A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107609460A (en) * | 2017-05-24 | 2018-01-19 | 南京邮电大学 | A kind of Human bodys' response method for merging space-time dual-network stream and attention mechanism |
| CN107862331A (en) * | 2017-10-31 | 2018-03-30 | 华中科技大学 | It is a kind of based on time series and CNN unsafe acts recognition methods and system |
Non-Patent Citations (6)
| Title |
|---|
| JIANG WANG ET AL.: "CNN-RNN: A Unified Framework for Multi-label Image Classification", 《CPVR》 * |
| KASTHURIRANGAN GOPALAKRISHNAN: "Deep Learning in Data-Driven Pavement Image Analysis and Automated Distress Detection:A Review", 《DATA》 * |
| MARKUS EISENBACH ET AL.: "How to Get Pavement Distress Detection Ready for Deep Learning? A Systematic Approach", 《2017 IEEE》 * |
| XIANGLONG WANG ET AL.: "Grid-based Pavement Crack Analysis Using Deep Learning", 《2017 4TH INTERNATIONAL CONFERENCE ON TRANSPORTATION INFORMATION AND SAFETY (ICTIS)》 * |
| 李楠等: "基于深度学习框架 Caffe 的路面裂缝识别研究", 《工程技术与应用》 * |
| 赵珊珊等: "基于卷积神经网络的路面裂缝检测", 《传感器与微系统》 * |
Cited By (36)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110020652A (en) * | 2019-01-07 | 2019-07-16 | 新而锐电子科技(上海)有限公司 | The dividing method of Tunnel Lining Cracks image |
| CN109872318A (en) * | 2019-02-22 | 2019-06-11 | 中国石油大学(华东) | A kind of geology for deep learning is appeared crack data set production method |
| CN109949290A (en) * | 2019-03-18 | 2019-06-28 | 北京邮电大学 | Pavement crack detection method, device, equipment and storage medium |
| CN109961030A (en) * | 2019-03-18 | 2019-07-02 | 北京邮电大学 | Pavement patching information detecting method, device, equipment and storage medium |
| CN110070520A (en) * | 2019-03-19 | 2019-07-30 | 长安大学 | The building of pavement crack detection model and detection method based on deep neural network |
| CN110070520B (en) * | 2019-03-19 | 2022-09-30 | 长安大学 | Pavement crack detection model construction and detection method based on deep neural network |
| CN110046656B (en) * | 2019-03-28 | 2023-07-11 | 南京邮电大学 | Multi-mode scene recognition method based on deep learning |
| CN109978141A (en) * | 2019-03-28 | 2019-07-05 | 腾讯科技（深圳）有限公司 | Neural network model training method and device, natural language processing method and apparatus |
| CN110046656A (en) * | 2019-03-28 | 2019-07-23 | 南京邮电大学 | Multi-modal scene recognition method based on deep learning |
| CN109978141B (en) * | 2019-03-28 | 2022-11-25 | 腾讯科技（深圳）有限公司 | Neural network model training method and device, and natural language processing method and device |
| CN110059804A (en) * | 2019-04-15 | 2019-07-26 | 北京迈格威科技有限公司 | Network training method, data processing method and device to be searched |
| CN110349119A (en) * | 2019-05-27 | 2019-10-18 | 北京邮电大学 | Pavement disease detection method and device based on edge detection neural network |
| CN110300257A (en) * | 2019-06-13 | 2019-10-01 | 广东技术师范大学天河学院 | A kind of face tracking safety defense monitoring system and its application method based on deep learning |
| CN110300257B (en) * | 2019-06-13 | 2020-12-08 | 广东技术师范大学天河学院 | Face tracking security monitoring system based on deep learning and use method thereof |
| WO2020248371A1 (en) * | 2019-06-14 | 2020-12-17 | 平安科技（深圳）有限公司 | Road damage detection method and apparatus, computer device, and storage medium |
| CN110503637A (en) * | 2019-08-13 | 2019-11-26 | 中山大学 | A kind of crack on road automatic testing method based on convolutional neural networks |
| CN110503637B (en) * | 2019-08-13 | 2022-12-06 | 中山大学 | Road crack automatic detection method based on convolutional neural network |
| CN110738131A (en) * | 2019-09-20 | 2020-01-31 | 广州游艺云物联网技术有限公司 | Garbage classification management method and device based on deep learning neural network |
| CN113015887A (en) * | 2019-10-15 | 2021-06-22 | 谷歌有限责任公司 | Navigation directions based on weather and road surface type |
| CN110992314A (en) * | 2019-11-15 | 2020-04-10 | 广东华路交通科技有限公司 | Pavement defect detection method and device and storage medium |
| CN111222546A (en) * | 2019-12-27 | 2020-06-02 | 中国科学院计算技术研究所 | Multi-scale fusion food image classification model training and image classification method |
| CN111222546B (en) * | 2019-12-27 | 2023-04-07 | 中国科学院计算技术研究所 | Multi-scale fusion food image classification model training and image classification method |
| CN111126505A (en) * | 2019-12-28 | 2020-05-08 | 北京工业大学 | Pavement crack rapid identification method based on deep learning |
| CN111583229B (en) * | 2020-05-09 | 2024-01-05 | 江苏野马软件科技有限公司 | Road surface fault detection method based on convolutional neural network |
| CN111583229A (en) * | 2020-05-09 | 2020-08-25 | 江苏野马软件科技有限公司 | Road surface fault detection method based on convolutional neural network |
| CN111862042B (en) * | 2020-07-21 | 2023-05-23 | 北京航空航天大学 | Pipeline contour detection method based on full convolution neural network |
| CN111862042A (en) * | 2020-07-21 | 2020-10-30 | 北京航空航天大学 | Pipeline contour detection method based on full convolution neural network |
| CN112986950A (en) * | 2020-12-25 | 2021-06-18 | 南京理工大学 | Single-pulse laser radar echo feature extraction method based on deep learning |
| CN114049356A (en) * | 2022-01-17 | 2022-02-15 | 湖南大学 | Method, device and system for detecting structure apparent crack |
| CN114049356B (en) * | 2022-01-17 | 2022-04-12 | 湖南大学 | Method, device and system for detecting structure apparent crack |
| CN115984221A (en) * | 2023-01-03 | 2023-04-18 | 广州新粤交通技术有限公司 | Road marking repairing and identifying method, device, equipment and storage medium thereof |
| CN115984221B (en) * | 2023-01-03 | 2023-08-04 | 广州新粤交通技术有限公司 | Road marking restoration and identification method, device, equipment and storage medium thereof |
| CN116664582A (en) * | 2023-08-02 | 2023-08-29 | 四川公路桥梁建设集团有限公司 | Road surface detection method and device based on neural vision network |
| CN116664582B (en) * | 2023-08-02 | 2023-10-27 | 四川公路桥梁建设集团有限公司 | Road surface detection method and device based on neural vision network |
| CN117115147A (en) * | 2023-10-19 | 2023-11-24 | 山东华盛创新纺织科技有限公司 | Textile detection method and system based on machine vision |
| CN117115147B (en) * | 2023-10-19 | 2024-01-26 | 山东华盛创新纺织科技有限公司 | Textile detection method and system based on machine vision |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109146849A (en) | A kind of road surface crack detection method based on convolutional neural networks and image recognition | |
| Kou et al. | Development of a YOLO-V3-based model for detecting defects on steel strip surface | |
| CN110070536A (en) | A kind of pcb board component detection method based on deep learning | |
| CN111127449B (en) | Automatic crack detection method based on encoder-decoder | |
| CN110059734A (en) | A kind of training method, object identification method, device, robot and the medium of target identification disaggregated model | |
| CN110569508A (en) | Method and system for classifying emotional tendencies by fusing part-of-speech and self-attention mechanism | |
| EP3349048A1 (en) | Inspection devices and methods for detecting a firearm in a luggage | |
| CN111739075A (en) | Deep network lung texture recognition method combining multi-scale attention | |
| CN108596226A (en) | A kind of defects of display panel training method and system based on deep learning | |
| CN109859163A (en) | A kind of LCD defect inspection method based on feature pyramid convolutional neural networks | |
| CN104992167A (en) | Convolution neural network based face detection method and apparatus | |
| CN110097145A (en) | One kind being based on CNN and the pyramidal traffic contraband recognition methods of feature | |
| CN110020652A (en) | The dividing method of Tunnel Lining Cracks image | |
| CN107066445A (en) | The deep learning method of one attribute emotion word vector | |
| CN105260734A (en) | Commercial oil surface laser code recognition method with self modeling function | |
| US20220315243A1 (en) | Method for identification and recognition of aircraft take-off and landing runway based on pspnet network | |
| CN110032969A (en) | For text filed method, apparatus, equipment and the medium in detection image | |
| CN116863274A (en) | Semi-supervised learning-based steel plate surface defect detection method and system | |
| CN105404865A (en) | Probability state restricted Boltzmann machine cascade based face detection method | |
| CN110991359A (en) | Satellite image target detection method based on multi-scale depth convolution neural network | |
| CN107545301A (en) | Page display method and device | |
| CN108416270A (en) | A kind of traffic sign recognition method based on more attribute union features | |
| CN110458794A (en) | Fittings quality detection method and device for track train | |
| CN110503098A (en) | A kind of object detection method and equipment of quick real-time lightweight | |
| Ni et al. | Toward high-precision crack detection in concrete bridges using deep learning |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190104 |