CN110197477A - The method, apparatus and system of pavement crack detection - Google Patents
The method, apparatus and system of pavement crack detection Download PDFInfo
- Publication number
- CN110197477A CN110197477A CN201910374004.9A CN201910374004A CN110197477A CN 110197477 A CN110197477 A CN 110197477A CN 201910374004 A CN201910374004 A CN 201910374004A CN 110197477 A CN110197477 A CN 110197477A
- Authority
- CN
- China
- Prior art keywords
- error
- network model
- image
- characteristic pattern
- pavement
- 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
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
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
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)
Abstract
The present invention provides the method, apparatus and system of a kind of pavement crack detection, this method, comprising: obtain pavement image to be detected；By the pavement image input coding device to be detected, multiple characteristic patterns are obtained；Using the characteristic pattern as the input of target detection network model, wherein the target detection network model refers to: according to the amplification characteristic figure of the characteristic pattern, reducing characteristic pattern, calculate the corresponding total losses error of the characteristic pattern；The corresponding target crack image of the pavement image to be detected is exported by the target detection network model.By that will divide network and the problems such as reconstructed network combines, avoids imbalanced training sets, the efficiency and accuracy of pavement crack detection are improved, error rate is reduced, especially there is preferable detection effect to minute crack.
Description
Technical field
The present invention relates to the method, apparatus that computer image recognition technology field more particularly to a kind of pavement crack are detected
And system.
Background technique
Road surface damage detection is the pith of road management, for obtaining the road condition information of maintenance.Crack is normal
The road surface seen type in danger, if the service life of road can shorten, and driving safety also will receive shadow without handling crack in time
It rings.Efficient pavement crack detection is road improvement situation, extends Road Service Life, reduces the effective of road maintenance cost
Approach.
Currently, traditional image processing method can not summarize the task of Crack Detection, i.e., can not become in environmental condition
Enough notable features are obtained in the case where change, from pavement image to distinguish crack image and non-crack image.
Learning model welcome in recent years is mainly used for solving pavement crack classification in a manner of its powerful feature extraction
The problems such as.Therefore the crack image for obtaining in mixed and disorderly background and significantly distinguishing, the efficiency of pavement crack detection still be cannot achieve
Lowly, accuracy is bad.
Summary of the invention
The present invention provides the method, apparatus and system of a kind of pavement crack detection, to improve the efficiency of pavement crack detection
And accuracy, the error rate of detection is reduced, especially there is preferable detection effect to minute crack.
In a first aspect, a kind of method of pavement crack detection provided in an embodiment of the present invention, comprising:
Obtain pavement image to be detected；
By the pavement image input coding device to be detected, multiple characteristic patterns are obtained；
Using the characteristic pattern as the input of target detection network model, wherein the target detection network model refers to:
According to the amplification characteristic figure of the characteristic pattern, characteristic pattern is reduced, calculates the corresponding total losses error of the characteristic pattern；
The corresponding target crack image of the pavement image to be detected is exported by the target detection network model.
In a kind of possible design, by the pavement image input coding device to be detected, multiple characteristic patterns are obtained, are wrapped
It includes:
Process of convolution is carried out to the pavement image to be detected by the encoder, and the image of process of convolution is inputted
Normalizing layer and active coating obtain multiple characteristic patterns；Wherein, the encoder uses multiple down-sampling grades, and each down-sampling grade
Including multiple convolutional layers.
In a kind of possible design, it is corresponding that the pavement image to be detected is exported by the target detection network model
Target crack image before, further includes:
Initial detecting network model is constructed, the building initial detecting network model includes decoding branch and segmentation branch；
The decoding branch includes multiple cascade up-sampling grades, wherein up-sampling grade includes: 2D transposition convolutional layer, normalizing layer, activation
Layer, wherein the 2D transposition convolutional layer is for up-sampling the characteristic pattern of input to obtain the amplification characteristic figure, it is described to return
The size of one layer of amplification characteristic figure for default output, the active coating for exporting there is Nonlinear Mapping to enhance table
The amplification characteristic figure reached；The decoding branch is used to export the corresponding amplification characteristic figure of the characteristic pattern, and according to described
Reconstruct loss error is calculated in pavement image to be detected；The segmentation branch includes the continuous down-sampling grade of at least two, for defeated
The corresponding diminution characteristic pattern of the characteristic pattern out, and segmentation loss error is calculated according to preset standard characteristic pattern；
Using total losses error as evaluation goal, by the training dataset training initial detecting network model, institute is obtained
State target detection network model.
In a kind of possible design, the total losses error includes:
Reconstruct loss error rec_loss and segmentation loss error seg_loss；
The total losses error hinge loss=a*rec_loss+b*seg_loss；Wherein: a indicates that reconstruct loss misses
The penalty values of difference account for the ratio of total losses value；B indicates that the penalty values of segmentation loss error account for the ratio of total losses value.
In a kind of possible design, it is corresponding that the pavement image to be detected is exported by the target detection network model
Target crack image, comprising:
If detecting the penalty values convergence of the total losses error or being less than preset threshold, the target detection network mould
Type is exported using the corresponding diminution characteristic pattern of segmentation loss error as target crack image.
Second aspect, a kind of device of pavement crack detection provided in an embodiment of the present invention, comprising:
Module is obtained, for obtaining pavement image to be detected；
Coding module, for obtaining multiple characteristic patterns for the pavement image input coding device to be detected；
Detection module, for using the characteristic pattern as the input of target detection network model, wherein the target detection
Network model refers to: according to the amplification characteristic figure of the characteristic pattern, reducing characteristic pattern, calculates the corresponding total losses of the characteristic pattern
Error；
Output module, for exporting the corresponding target of the pavement image to be detected by the target detection network model
Crack image.
In a kind of possible design, the coding module is specifically used for:
Process of convolution is carried out to the pavement image to be detected by the encoder, and the image of process of convolution is inputted
Normalizing layer and active coating obtain multiple characteristic patterns；Wherein, the encoder lower uses grade, and each down-sampling grade using multiple
Including multiple convolutional layers.
In a kind of possible design, it is corresponding that the pavement image to be detected is exported by the target detection network model
Target crack image before, further includes:
Initial detecting network model is constructed, the building initial detecting network model includes decoding branch and segmentation branch；
The decoding branch includes multiple cascade up-sampling grades；Wherein up-sampling grade includes: 2D transposition convolutional layer, normalizing layer, activation
Layer, wherein the 2D transposition convolutional layer is for up-sampling the characteristic pattern of input to obtain the amplification characteristic figure, it is described to return
The size of one layer of amplification characteristic figure for default output, the active coating for exporting there is Nonlinear Mapping to enhance table
The amplification characteristic figure reached；The decoding branch is used to export the corresponding amplification characteristic figure of the characteristic pattern, and according to described
Reconstruct loss error is calculated in pavement image to be detected；The segmentation branch includes the continuous down-sampling grade of at least two, for defeated
The corresponding diminution characteristic pattern of the characteristic pattern out, and segmentation loss error is calculated according to preset standard characteristic pattern；
Using total losses error as evaluation goal, by the training dataset training initial detecting network model, institute is obtained
State target detection network model.
In a kind of possible design, the total losses error includes:
Reconstruct loss error rec_loss and segmentation loss error seg_loss；
The total losses error hinge loss=a*rec_loss+b*seg_loss；Wherein: a indicates that reconstruct loss misses
The penalty values of difference account for the ratio of total losses value；B indicates that the penalty values of segmentation loss error account for the ratio of total losses value.
In a kind of possible design, the output module is specifically used for:
If detecting the penalty values convergence of the total losses error or being less than preset threshold, the target detection network mould
Type is exported using the corresponding diminution characteristic pattern of segmentation loss error as target crack image.
The third aspect, a kind of system of pavement crack detection provided in an embodiment of the present invention, including memory and processor,
The executable instruction of the processor is stored in the memory, wherein the processor is configured to hold via described in execution
The method that row instructs to execute the detection of pavement crack described in any one of first aspect.
A kind of fourth aspect, computer readable storage medium provided in an embodiment of the present invention, is stored thereon with computer journey
Sequence realizes the method for pavement crack detection described in any one of first aspect when the program is executed by processor.
The present invention provides the method, apparatus and system of a kind of pavement crack detection, this method, comprising: obtain to be detected
Pavement image obtains pavement image to be detected；By the pavement image input coding device to be detected, multiple characteristic patterns are obtained；By institute
Input of the characteristic pattern as target detection network model is stated, wherein the target detection network model refers to: according to the feature
The amplification characteristic figure of figure reduces characteristic pattern, calculates the corresponding total losses error of the characteristic pattern；Pass through the target detection network
Model exports the corresponding target crack image of the pavement image to be detected.By the way that network will be divided and reconstructed network combines,
The problems such as avoiding imbalanced training sets, realizes the efficiency for efficiently carrying out improving pavement crack detection and accuracy road surface is split
Seam, reduces the error rate of detection, especially has preferable detection effect to minute crack.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is this hair
Bright some embodiments for those of ordinary skill in the art without any creative labor, can be with
It obtains other drawings based on these drawings.
Fig. 1 is the method flow diagram for the pavement crack detection that the embodiment of the present invention one provides；
Fig. 2 is the effect diagram for the pavement crack detection that the embodiment of the present invention one provides；
Fig. 3 is the method flow diagram of pavement crack provided by Embodiment 2 of the present invention detection；
Fig. 4 is the method flow schematic diagram for the pavement crack detection that the embodiment of the present invention three provides；
Fig. 5 is the structural schematic diagram of the device for the pavement crack detection that the embodiment of the present invention four provides；
Fig. 6 is the structural schematic diagram of the system for the pavement crack detection that the embodiment of the present invention five provides.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
Every other embodiment obtained without making creative work, shall fall within the protection scope of the present invention.
Description and claims of this specification and term " first ", " second ", " third " " in above-mentioned attached drawing
The (if present)s such as four " are to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should manage
The data that solution uses in this way are interchangeable under appropriate circumstances, so that the embodiment of the present invention described herein for example can be to remove
Sequence other than those of illustrating or describe herein is implemented.In addition, term " includes " and " having " and theirs is any
Deformation, it is intended that cover it is non-exclusive include, for example, containing the process, method of a series of steps or units, system, production
Product or equipment those of are not necessarily limited to be clearly listed step or unit, but may include be not clearly listed or for this
A little process, methods, the other step or units of product or equipment inherently.
How to be solved with technical solution of the specifically embodiment to technical solution of the present invention and the application below above-mentioned
Technical problem is described in detail.These specific embodiments can be combined with each other below, for the same or similar concept
Or process may repeat no more in certain embodiments.Below in conjunction with attached drawing, the embodiment of the present invention is described.
Increased rapidly by the volume of traffic, vehicle enlargement, overload are serious and traveling channelization etc. influences, pavement behavior is subjected to sternly
High test, along with the factors such as construction, weather or weather accelerate pavement damage, to influence pavement life, wherein crack
Class breakage is common one of road surface loss, if it is possible to find in time and detect its situation, it will help reduce pavement preservation
Cost and workload, improve drive safety.The present invention improves road surface by that will divide network and reconstructed network combines
The efficiency and accuracy of Crack Detection, reduce the error rate of detection, especially have preferable detection effect to minute crack.This hair
The system of bright executing subject pavement crack detection can be mounted or integrated into arbitrary terminal device, in a kind of optional implementation
In example, terminal device may include mobile phone, tablet computer etc..
Fig. 1 is the method flow diagram for the pavement crack detection that the embodiment of the present invention one provides, as shown in Figure 1, the present embodiment
Pavement crack detection method may include:
S101, pavement image to be detected is obtained.
Specifically, pavement image to be detected can be obtained by camera captured in real-time, or from the video of captured in real-time
The middle pavement image for obtaining a certain frame, can also obtain pavement image to be detected from existing image data base.This field skill
Art personnel can select approach appropriate to obtain pavement image to be detected according to the actual situation, to obtain better effect, this
It is not especially limited in embodiment.
S102, by pavement image input coding device to be detected, obtain multiple characteristic patterns.
Specifically, carrying out process of convolution to pavement image to be detected by encoder, and the image of process of convolution is inputted
Normalizing layer and active coating obtain multiple characteristic patterns；Wherein, encoder uses multiple down-sampling grades, and each sample stage can wrap
Include multiple convolutional layers.
In the present embodiment, encoder can use 3 down-sampling grades, and each sample stage may include 3 convolutional layers, can
The resolution ratio of pavement image to be detected is reduced to the 1/8 of original resolution sizes.Encoder to pavement image to be detected into
Row process of convolution, and by after the image of process of convolution input normalizing layer and active coating, it is compressed into multiple characteristic patterns.Wherein convolution
Processing, which refers to, operates each pixel in the pavement image to be detected using a convolution kernel, and convolution kernel is for mapping
As the matrix of processing, such as the matrix or pixel region of 3*3.By the image input normalizing layer of process of convolution to obtain uniform sizes
The characteristic pattern of size, active coating is for being added non-linear factor to obtain the characteristic pattern of Enhanced expressing effect.
S103, using characteristic pattern as the input of target detection network model, wherein target detection network model refers to: according to
The amplification characteristic figure of characteristic pattern reduces characteristic pattern, calculates the corresponding total losses error of characteristic pattern.
Specifically, total losses error may include: reconstruct loss error rec_loss and segmentation loss error seg_loss；
Total losses error hinge loss=a*rec_loss+b*seg_loss；Wherein: a indicates reconstruct loss error
Penalty values account for the ratio of total losses value；B indicates that the penalty values of segmentation loss error account for the ratio of total losses value.
In the present embodiment, target detection network model may include decoding branch and segmentation branch.Wherein decoding branch can
To include multiple cascade up-sampling grades, and the up-sampling grade may include: 2D transposition convolutional layer, normalizing layer and active coating, be somebody's turn to do
Reconstruct loss is calculated for exporting the corresponding amplification characteristic figure of characteristic pattern, and according to the pavement image to be detected in decoding branch
Error rec_loss.Dividing branch may include multiple continuous down-sampling grades, and for exporting the corresponding diminution feature of characteristic pattern
Figure, and segmentation loss error seg_loss is calculated according to preset standard characteristic pattern.In an alternative embodiment, divide
Loss error can use cross entropy, dice coefficient (likeness coefficient) etc..And then calculate total losses error.
S104, the corresponding target crack image of pavement image to be detected is exported by target detection network model.
Specifically, if detecting the penalty values convergence of total losses error or being less than preset threshold, target detection network mould
Type is exported using the corresponding diminution characteristic pattern of segmentation loss error as target crack image.
In the present embodiment, if detecting the penalty values convergence of total losses error or being less than preset threshold, by target detection
The corresponding diminution characteristic pattern of segmentation loss error is exported as target crack image in network model.Such as it is with reference to Fig. 2, Fig. 2
The effect diagram for the pavement crack detection that the embodiment of the present invention one provides, three optimization aim cracks as shown in Figure 2 image
The corresponding amplification characteristic figure of segmentation loss error as optimized in the present embodiment.Preset threshold is not limited in the present embodiment
Fixed, those skilled in the art can make specifically to limit to reach better effect according to the actual situation.
The efficiency and accuracy of pavement crack detection can be improved in the method for pavement crack detection in the present embodiment, reduces
The error rate of detection especially has preferable detection effect to minute crack.
In an alternative embodiment, the system of pavement crack detection is by building initial detecting network model, and leads to
It crosses training dataset training and obtains target detection network model.
Fig. 3 is the method flow diagram of pavement crack provided by Embodiment 2 of the present invention detection, as shown in figure 3, the present embodiment
In pavement crack detection method may include:
Step S201, initial detecting network model is constructed, building initial detecting network model includes decoding branch and segmentation
Branch；Decoding branch includes multiple cascade up-sampling grades；Wherein up-sampling grade includes: 2D transposition convolutional layer, normalizing layer, activation
Layer, wherein for being up-sampled to obtain the amplification characteristic figure to the characteristic pattern of input, normalizing layer is used for 2D transposition convolutional layer
The size of the amplification characteristic figure of default output, active coating are used to export the amplification characteristic figure with Nonlinear Mapping Enhanced expressing；
Decoding branch is calculated reconstruct loss according to pavement image to be detected and misses for exporting the corresponding amplification characteristic figure of characteristic pattern
Difference；Dividing branch includes the continuous down-sampling grade of at least two, for exporting the corresponding diminution characteristic pattern of characteristic pattern, and according to default
Segmentation loss error is calculated in standard feature figure.
Step S202, it using total losses error as evaluation goal, by training dataset training initial detecting network model, obtains
To target detection network model.
Specifically, by target detection network model export the corresponding target crack image of pavement image to be detected it
Before, it may include building initial detecting network model.
With reference to Fig. 4, Fig. 4 is the method flow schematic diagram for the pavement crack detection that the embodiment of the present invention three provides, such as Fig. 4 institute
Show, initial detecting network model includes decoding branch and segmentation branch.Wherein decoding branch may include 3 cascade up-samplings
Grade, in which: up-sampling grade includes 2D transposition convolutional layer, normalizing layer, active coating, and 2D transposition convolutional layer is used for the characteristic pattern to input
It is up-sampled to obtain the amplification characteristic figure, size of the normalizing layer for the amplification characteristic figure of default output, active coating is used for
Export the amplification characteristic figure with Nonlinear Mapping Enhanced expressing.Decoding branch is for exporting the corresponding amplification characteristic of characteristic pattern
Figure, and reconstruct loss error rec_loss can be calculated according to pavement image to be detected.Segmentation branch includes that at least two connects
Continuous down-sampling grade, the segmentation branch calculate for exporting the corresponding diminution characteristic pattern of characteristic pattern, and according to preset standard characteristic pattern
Obtain segmentation loss error seg_loss.And then using total losses error as evaluation goal, by training dataset training, this is initial
Network model is detected, target detection network model is obtained, in an alternative embodiment, total losses error hinge loss=
a*rec_loss+b*seg_loss；Wherein: a indicates that the penalty values of reconstruct loss error account for the ratio of total losses value；B indicates to divide
The penalty values for cutting loss error account for the ratio of total losses value.Reconstruct branch therein and segmentation branch are based on convolution, are training
The image of any stage can be supported to input in journey, training, using road surface normal region image and crack image-region as negative sample
Sheet and positive sample, and using the difference for maximizing road surface normal region and crack area based on hinge_loss, and then obtain
Target crack image, the problems such as avoiding imbalanced training sets, improve the efficiency of pavement crack detection, reduce the mistake of detection
Rate especially has preferable detection effect to minute crack.
In an alternative embodiment, the pavement image that training data is concentrated is cut into patch (subgraph), such as from original
The patch that 192 × 192 pixels are plucked out in beginning image obtains each patch corresponding to 6 × 6 sizes using down-sampling rate 1/32
Preset standard image.
Step S203, pavement image to be detected is obtained.
Step S204, by pavement image input coding device to be detected, multiple characteristic patterns are obtained.
Step S205, using characteristic pattern as the input of target detection network model, wherein target detection network model refers to:
According to the amplification characteristic figure of characteristic pattern, characteristic pattern is reduced, calculates the corresponding total losses error of characteristic pattern.
Step S206, the corresponding target crack image of pavement image to be detected is exported by target detection network model.
Step S203~step 206 specific implementation process and technical principle side shown in Figure 1 in the present embodiment
Associated description in method in step S101~step 104, details are not described herein again.
With reference to Fig. 2, the optimization aim crack image as obtained in three groups of optimizations experiment of Fig. 2 shows the target detection net
Network model can the more efficient inspection that must carry out the abnormal areas such as pavement crack compared to the model only containing segmentation network detection
It surveys, to reduce the wrong report of pavement crack detection.Road surface more efficiently can carried out just by autocoder in this implementation
The denoising in normal region；It is reconstructed by target detection network model, it is ensured that more effective, high quality normal region road surface
The detection in (region i.e. other than pavement crack)；The difference of road surface normal region and crack area is maximized based on hinge_loss
It is different, allow segmentation branch accurately to carry out the detection of minute crack, it can be with compared to the model only containing segmentation network detection
It is more efficient to carry out the detection of the abnormal areas such as pavement crack, to reduce the wrong report of pavement crack detection.
Fig. 5 is the structural schematic diagram of the device for the pavement crack detection that the embodiment of the present invention four provides, as shown in figure 5, this
The device of pavement crack detection may include: in embodiment
Module 31 is obtained, for obtaining pavement image to be detected；
Coding module 32, for obtaining multiple characteristic patterns for pavement image input coding device to be detected；
Detection module 33, for using characteristic pattern as the input of target detection network model, wherein target detection network mould
Type refers to: according to the amplification characteristic figure of characteristic pattern, reducing characteristic pattern, calculates the corresponding total losses error of characteristic pattern；
Output module 34, for exporting the corresponding target crack pattern of pavement image to be detected by target detection network model
Picture.
In an alternative embodiment, coding module 32 are specifically used for:
By encoder to pavement image to be detected carry out process of convolution, and by the image of process of convolution input normalizing layer and
Active coating obtains multiple characteristic patterns；Wherein, encoder uses multiple lower using grade, and each down-sampling grade includes multiple convolution
Layer.
In an alternative embodiment, the corresponding target of pavement image to be detected is exported by target detection network model
Before the image of crack, further includes:
Initial detecting network model is constructed, building initial detecting network model includes decoding branch and segmentation branch；Decoding
Branch includes multiple cascade up-sampling grades；Wherein: up-sampling grade includes: 2D transposition convolutional layer, normalizing layer, active coating, wherein
2D transposition convolutional layer is used to be up-sampled to obtain to the characteristic pattern of input the amplification characteristic figure, and normalizing layer is for default output
Amplification characteristic figure size, active coating be used for export have Nonlinear Mapping Enhanced expressing amplification characteristic figure；Decode branch
Reconstruct loss error is calculated for exporting the corresponding amplification characteristic figure of characteristic pattern, and according to pavement image to be detected；Segmentation
Branch includes the continuous down-sampling grade of at least two, for exporting the corresponding diminution characteristic pattern of characteristic pattern, and according to preset standard feature
Segmentation loss error is calculated in figure；
Using total losses error as evaluation goal, by training dataset training initial detecting network model, target inspection is obtained
Survey network model.
In an alternative embodiment, total losses error includes:
Reconstruct loss error rec_loss and segmentation loss error seg_loss；
Total losses error hinge loss=a*rec_loss+b*seg_loss；Wherein: a indicates reconstruct loss error
Penalty values account for the ratio of total losses value；B indicates that the penalty values of segmentation loss error account for the ratio of total losses value.
In an alternative embodiment, output module 34 are specifically used for:
If detecting the penalty values convergence of total losses error or being less than preset threshold, target detection network model will be divided
The corresponding diminution characteristic pattern of error is lost to export as target crack image.
The device of the pavement crack detection of the present embodiment can execute the technical solution in method shown in Fig. 2, Fig. 3, have
Body realizes the associated description of process and technical principle referring to fig. 2, in method shown in Fig. 3, and details are not described herein again.
Fig. 6 is the structural schematic diagram of the system for the pavement crack detection that the embodiment of the present invention five provides, as shown in fig. 6, this
The system 40 of the pavement crack detection of embodiment may include: processor 41 and memory 42.
Memory 42 (such as realizes application program, the function of the method for above-mentioned pavement crack detection for storing computer program
Can module etc.), computer instruction etc.；
Above-mentioned computer program, computer instruction etc. can be with partitioned storages in one or more memories 42.And
Above-mentioned computer program, computer instruction, data etc. can be called with device 41 processed.
Processor 41, for executing the computer program of the storage of memory 42, to realize method that above-described embodiment is related to
In each step.
It specifically may refer to the associated description in previous methods embodiment.
Processor 41 and memory 42 can be absolute construction, be also possible to the integrated morphology integrated.Work as processing
When device 41 and memory 42 are absolute construction, memory 42, processor 41 can be of coupled connections by bus 43.
The server of the present embodiment can execute the technical solution in method shown in Fig. 2, Fig. 3, specific implementation process and
Associated description of the technical principle referring to fig. 2, in method shown in Fig. 3, details are not described herein again.
In addition, the embodiment of the present application also provides a kind of computer readable storage medium, deposited in computer readable storage medium
Computer executed instructions are contained, when at least one processor of user equipment executes the computer executed instructions, user equipment
Execute above-mentioned various possible methods.
Wherein, computer-readable medium includes computer storage media and communication media, and wherein communication media includes being convenient for
From a place to any medium of another place transmission computer program.Storage medium can be general or specialized computer
Any usable medium that can be accessed.A kind of illustrative storage medium is coupled to processor, to enable a processor to from this
Read information, and information can be written to the storage medium.Certainly, storage medium is also possible to the composition portion of processor
Point.Pocessor and storage media can be located in ASIC.In addition, the ASIC can be located in user equipment.Certainly, processor and
Storage medium can also be used as discrete assembly and be present in communication equipment.
Those of ordinary skill in the art will appreciate that: realize that all or part of the steps of above-mentioned each method embodiment can lead to
The relevant hardware of program instruction is crossed to complete.Program above-mentioned can be stored in a computer readable storage medium.The journey
When being executed, execution includes the steps that above-mentioned each method embodiment to sequence；And storage medium above-mentioned include: ROM, RAM, magnetic disk or
The various media that can store program code such as person's CD.
Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Pipe present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that: its according to
So be possible to modify the technical solutions described in the foregoing embodiments, or to some or all of the technical features into
Row equivalent replacement；And these are modified or replaceed, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (12)
1. a kind of method of pavement crack detection characterized by comprising
Obtain pavement image to be detected；
By the pavement image input coding device to be detected, multiple characteristic patterns are obtained；
Using the characteristic pattern as the input of target detection network model, wherein the target detection network model refers to: according to
The amplification characteristic figure of the characteristic pattern reduces characteristic pattern, calculates the corresponding total losses error of the characteristic pattern；
The corresponding target crack image of the pavement image to be detected is exported by the target detection network model.
2. the method according to claim 1, wherein the pavement image input coding device to be detected is obtained
Multiple characteristic patterns, comprising:
Process of convolution is carried out to the pavement image to be detected by the encoder, and the image of process of convolution is inputted into normalizing
Layer and active coating, obtain multiple characteristic patterns；Wherein, the encoder uses multiple lower using grade, and each down-sampling grade includes
Multiple convolutional layers.
3. method according to claim 1 or 2, which is characterized in that by described in target detection network model output
Before the corresponding target crack image of pavement image to be detected, further includes:
Initial detecting network model is constructed, the building initial detecting network model includes decoding branch and segmentation branch；It is described
Decoding branch includes multiple cascade up-sampling grades；Wherein up-sampling grade includes: 2D transposition convolutional layer, normalizing layer, active coating,
In: the 2D transposition convolutional layer is for up-sampling the characteristic pattern of input to obtain the amplification characteristic figure, the normalizing layer
The size of the amplification characteristic figure for default output, the active coating are used to export with Nonlinear Mapping Enhanced expressing
The amplification characteristic figure；The decoding branch is used to export the corresponding amplification characteristic figure of the characteristic pattern, and according to described to be checked
It surveys pavement image and reconstruct loss error is calculated；The segmentation branch includes the continuous down-sampling grade of at least two, for exporting
The corresponding diminution characteristic pattern of characteristic pattern is stated, and segmentation loss error is calculated according to preset standard characteristic pattern；
Using total losses error as evaluation goal, by the training dataset training initial detecting network model, the mesh is obtained
Mark detection network model.
4. according to the method described in claim 3, it is characterized in that, the total losses error includes:
Reconstruct loss error rec_loss and segmentation loss error seg_loss；
The total losses error hinge loss=a*rec_loss+b*seg_loss；Wherein: a indicates reconstruct loss error
Penalty values account for the ratio of total losses value；B indicates that the penalty values of segmentation loss error account for the ratio of total losses value.
5. according to the method described in claim 4, it is characterized in that, being exported by the target detection network model described to be checked
Survey the corresponding target crack image of pavement image, comprising:
If detecting the penalty values convergence of the total losses error or being less than preset threshold, the target detection network model will
The segmentation is lost the corresponding diminution characteristic pattern of error and is exported as target crack image.
6. a kind of device of pavement crack detection characterized by comprising
Module is obtained, for obtaining pavement image to be detected；
Coding module, for obtaining multiple characteristic patterns for the pavement image input coding device to be detected；
Detection module, for using the characteristic pattern as the input of target detection network model, wherein the target detection network
Model refers to: according to the amplification characteristic figure of the characteristic pattern, reducing characteristic pattern, calculates the corresponding total losses of the characteristic pattern and miss
Difference；
Output module, for exporting the corresponding target crack of the pavement image to be detected by the target detection network model
Image.
7. device according to claim 6, which is characterized in that the coding module is specifically used for:
Process of convolution is carried out to the pavement image to be detected by the encoder, and the image of process of convolution is inputted into normalizing
Layer and active coating, obtain multiple characteristic patterns；Wherein, the encoder uses multiple lower using grade, and each down-sampling grade includes
Multiple convolutional layers.
8. device according to claim 6 or 7, which is characterized in that by described in target detection network model output
Before the corresponding target crack image of pavement image to be detected, further includes:
Initial detecting network model is constructed, the building initial detecting network model includes decoding branch and segmentation branch；It is described
Decoding branch includes multiple cascade up-sampling grades；Wherein up-sampling grade includes: 2D transposition convolutional layer, normalizing layer, active coating,
In: the 2D transposition convolutional layer is for up-sampling the characteristic pattern of input to obtain the amplification characteristic figure, the normalizing layer
The size of the amplification characteristic figure for default output, the active coating are used to export with Nonlinear Mapping Enhanced expressing
The amplification characteristic figure；The decoding branch is used to export the corresponding amplification characteristic figure of the characteristic pattern, and according to described to be checked
It surveys pavement image and reconstruct loss error is calculated；The segmentation branch includes the continuous down-sampling grade of at least two, for exporting
The corresponding diminution characteristic pattern of characteristic pattern is stated, and segmentation loss error is calculated according to preset standard characteristic pattern；
Using total losses error as evaluation goal, by the training dataset training initial detecting network model, the mesh is obtained
Mark detection network model.
9. device according to claim 8, which is characterized in that the total losses error includes:
Reconstruct loss error rec_loss and segmentation loss error seg_loss；
The total losses error hinge loss=a*rec_loss+b*seg_loss；Wherein: a indicates reconstruct loss error
Penalty values account for the ratio of total losses value；B indicates that the penalty values of segmentation loss error account for the ratio of total losses value.
10. device according to claim 9, which is characterized in that the output module is specifically used for:
If detecting the penalty values convergence of the total losses error or being less than preset threshold, the target detection network model will
The segmentation is lost the corresponding diminution characteristic pattern of error and is exported as target crack image.
11. a kind of system of pavement crack detection characterized by comprising memory and processor store in memory
State the executable instruction of processor；Wherein, the processor is configured to want via executing the executable instruction and carry out perform claim
The method for asking the described in any item pavement crack detections of 1-5.
12. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the program is by processor
The method of the described in any item pavement crack detections of claim 1-5 is realized when execution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910374004.9A CN110197477A (en) | 2019-05-07 | 2019-05-07 | The method, apparatus and system of pavement crack detection |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910374004.9A CN110197477A (en) | 2019-05-07 | 2019-05-07 | The method, apparatus and system of pavement crack detection |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110197477A true CN110197477A (en) | 2019-09-03 |
Family
ID=67752353
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910374004.9A Pending CN110197477A (en) | 2019-05-07 | 2019-05-07 | The method, apparatus and system of pavement crack detection |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110197477A (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106846267A (en) * | 2016-12-30 | 2017-06-13 | 四川迪派锐科技有限公司 | A kind of pipe welding line detecting multi-scale image contrast enhancement process |
| CN107403197A (en) * | 2017-07-31 | 2017-11-28 | 武汉大学 | A kind of crack identification method based on deep learning |
| JP2018128309A (en) * | 2017-02-07 | 2018-08-16 | 大成建設株式会社 | Crack detection method |
| CN108416307A (en) * | 2018-03-13 | 2018-08-17 | 北京理工大学 | A kind of Aerial Images road surface crack detection method, device and equipment |
| CN108520516A (en) * | 2018-04-09 | 2018-09-11 | 陕西师范大学 | A kind of bridge pavement Crack Detection and dividing method based on semantic segmentation |
| CN108876780A (en) * | 2018-06-26 | 2018-11-23 | 陕西师范大学 | Bridge Crack image crack detection method under a kind of complex background |
| CN108876781A (en) * | 2018-06-26 | 2018-11-23 | 广东工业大学 | Surface defect recognition method based on SSD algorithm |
| CN108961270A (en) * | 2018-06-26 | 2018-12-07 | 陕西师范大学 | A kind of Bridge Crack Image Segmentation Model based on semantic segmentation |
| CN109672885A (en) * | 2019-01-08 | 2019-04-23 | 中国矿业大学(北京) | A kind of video image encoding and decoding method for mine intelligent monitoring |
-
2019
- 2019-05-07 CN CN201910374004.9A patent/CN110197477A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106846267A (en) * | 2016-12-30 | 2017-06-13 | 四川迪派锐科技有限公司 | A kind of pipe welding line detecting multi-scale image contrast enhancement process |
| JP2018128309A (en) * | 2017-02-07 | 2018-08-16 | 大成建設株式会社 | Crack detection method |
| CN107403197A (en) * | 2017-07-31 | 2017-11-28 | 武汉大学 | A kind of crack identification method based on deep learning |
| CN108416307A (en) * | 2018-03-13 | 2018-08-17 | 北京理工大学 | A kind of Aerial Images road surface crack detection method, device and equipment |
| CN108520516A (en) * | 2018-04-09 | 2018-09-11 | 陕西师范大学 | A kind of bridge pavement Crack Detection and dividing method based on semantic segmentation |
| CN108876780A (en) * | 2018-06-26 | 2018-11-23 | 陕西师范大学 | Bridge Crack image crack detection method under a kind of complex background |
| CN108876781A (en) * | 2018-06-26 | 2018-11-23 | 广东工业大学 | Surface defect recognition method based on SSD algorithm |
| CN108961270A (en) * | 2018-06-26 | 2018-12-07 | 陕西师范大学 | A kind of Bridge Crack Image Segmentation Model based on semantic segmentation |
| CN109672885A (en) * | 2019-01-08 | 2019-04-23 | 中国矿业大学(北京) | A kind of video image encoding and decoding method for mine intelligent monitoring |
Non-Patent Citations (4)
| Title |
|---|
| LUO LING 等: "Reconstruction-based Robust Pavement Crack Detection", 《BEIJING:SCIENCEPAPER》 * |
| 司马海峰 等编著: "《遥感图像分类中的智能计算方法》", 31 January 2018, 吉林大学出版社 * |
| 沙爱民 等: "基于卷积神经网络的路表病害识别与测量", 《中国公路学报》 * |
| 郝志峰 等编著: "《数据科学与数学建模》", 31 January 2019, 华中科技大学出版社 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Li et al. | Localization of deep inpainting using high-pass fully convolutional network | |
| CN107403197B (en) | Crack identification method based on deep learning | |
| CN108664981A (en) | Specific image extracting method and device | |
| JP2017514365A5 (en) | ||
| CN111008961B (en) | Transmission line equipment defect detection method and system, equipment and medium thereof | |
| CN107273791A (en) | A kind of articles from the storeroom checking method based on unmanned plane image technique | |
| CN114549563A (en) | Real-time composite insulator segmentation method and system based on deep LabV3+ | |
| CN111709883A (en) | Image detection method, device and equipment | |
| CN113610772B (en) | Method, system, device and storage medium for detecting spraying code defect at bottom of pop can bottle | |
| CN112785578A (en) | Road crack detection method and system based on U-shaped codec neural network | |
| CN108932474B (en) | Remote sensing image cloud judgment method based on full convolution neural network composite characteristics | |
| CN110399760A (en) | A kind of batch two dimensional code localization method, device, electronic equipment and storage medium | |
| CN112183218A (en) | Method and device for monitoring industrial production index | |
| Doycheva et al. | GPU-enabled pavement distress image classification in real time | |
| CN115809982B (en) | Method, device and system for detecting cell crush injury | |
| CN116912257B (en) | Concrete pavement crack identification method based on deep learning and storage medium | |
| CN116030074A (en) | Identification method, re-identification method and related equipment for road diseases | |
| CN107729842A (en) | Oil-gas pipeline damage from third-party dangerous discernment method, apparatus and system based on machine vision | |
| CN114418937B (en) | Pavement crack detection method and related equipment | |
| CN111027564A (en) | Low-illumination imaging license plate recognition method and device based on deep learning integration | |
| CN110197477A (en) | The method, apparatus and system of pavement crack detection | |
| CN112070792B (en) | Edge growth connection method and device for image segmentation | |
| CN114648467B (en) | Image defogging method and device, terminal equipment and computer readable storage medium | |
| CN115330729A (en) | Multi-scale feature attention-fused light-weight strip steel surface defect detection method | |
| CN115578371A (en) | Detection method and device based on mildewed sundries before tobacco bale slicing |
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: 20190903 |
|
| WD01 | Invention patent application deemed withdrawn after publication |