CN108182681A - A kind of distress in concrete detection method for the learning network that transfinited based on multilayer - Google Patents
A kind of distress in concrete detection method for the learning network that transfinited based on multilayer Download PDFInfo
- Publication number
- CN108182681A CN108182681A CN201810031254.8A CN201810031254A CN108182681A CN 108182681 A CN108182681 A CN 108182681A CN 201810031254 A CN201810031254 A CN 201810031254A CN 108182681 A CN108182681 A CN 108182681A
- Authority
- CN
- China
- Prior art keywords
- image
- default
- network
- layer
- learning
- 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
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30132—Masonry; Concrete
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The present invention is suitable for technical field of image processing, provides a kind of training method, training device and the terminal device of distress in concrete detector, including：Sample image is obtained, and the sample image is pre-processed；Feature extraction is carried out to pretreated sample image based on default Feature Selection Model, obtains characteristic；Default detector is trained based on the characteristic, the default detector after being trained；The detection of distress in concrete is carried out to image to be detected using the default detector after the training；Effectively increase the efficiency and precision of distress in concrete detection.
Description
Technical field
The invention belongs to technical field of image processing more particularly to a kind of coagulation soil crackings for the learning network that transfinited based on multilayer
Stitch detection method.
Background technology
Crack is one of most common security risk for influencing mixed mud structural strength, accurately and timely finds coagulation soil cracking
Seam is to safeguarding that concrete mechanism safety and service life are significant.
Common distress in concrete detection mode has artificial detection, ultrasound examination, the detection method based on image procossing.
Existing distress in concrete detection mode accuracy of detection is relatively low, and detection efficiency is low.
Invention content
In view of this, an embodiment of the present invention provides a kind of distress in concrete detection sides for the learning network that transfinited based on multilayer
Method, to solve the problems, such as that the precision of distress in concrete detection in the prior art is low, efficiency is low.
The first aspect of the embodiment of the present invention provides a kind of distress in concrete detection for the learning network that transfinites based on multilayer
Method, including：
Sample image is obtained, and the sample image is pre-processed；
Feature extraction is carried out to pretreated sample image based on default Feature Selection Model, obtains characteristic；
Default detector is trained based on the characteristic, the default detector after being trained；
The detection of distress in concrete is carried out to image to be detected using the default detector after the training.
The second aspect of the embodiment of the present invention provides a kind of distress in concrete detection for the learning network that transfinites based on multilayer
Device, including：
Pretreatment unit for obtaining sample image, and pre-processes the sample image；
Feature extraction unit puies forward pretreated sample image progress feature for being based on default Feature Selection Model
It takes, obtains characteristic；
Training unit is trained default detector for being based on the characteristic, the default inspection after being trained
Survey device；
Detection unit, for carrying out the inspection of distress in concrete to image to be detected using the default detector after the training
It surveys.
The third aspect of the embodiment of the present invention provides a kind of terminal device, including memory, processor and is stored in
In the memory and the computer program that can run on the processor, when the processor performs the computer program
The step of realizing the method that first aspect of the embodiment of the present invention provides.
The fourth aspect of the embodiment of the present invention provides a kind of computer readable storage medium, the computer-readable storage
Media storage has computer program, and the embodiment of the present invention the is realized when the computer program is executed by one or more processors
On the one hand the step of the method provided.
Existing advantageous effect is the embodiment of the present invention compared with prior art：
The embodiment of the present invention is pre-processed by obtaining sample image, and to the sample image；Based on default feature
Extraction model carries out feature extraction to pretreated sample image, obtains characteristic；Based on the characteristic to default
Detector is trained, the default detector after being trained；Using the default detector after the training to image to be detected
Carry out the detection of distress in concrete；Effectively increase the efficiency and precision of distress in concrete detection.
Description of the drawings
It to describe the technical solutions in the embodiments of the present invention more clearly, below will be to embodiment or description of the prior art
Needed in attached drawing be briefly described, it should be apparent that, the accompanying drawings in the following description be only the present invention some
Embodiment, for those of ordinary skill in the art, without having to pay creative labor, can also be according to these
Attached drawing obtains other attached drawings.
Fig. 1 be it is provided in an embodiment of the present invention based on multilayer transfinite learning network distress in concrete detection method realization
Flow diagram；
Fig. 2 be it is provided in an embodiment of the present invention based on multilayer transfinite learning network distress in concrete detection device signal
Figure；
Fig. 3 is the schematic diagram of terminal device provided in an embodiment of the present invention；
Fig. 4 be it is provided in an embodiment of the present invention based on multilayer transfinite learning network distress in concrete detection method signal
Block diagram；
Fig. 5 be it is provided in an embodiment of the present invention based on multilayer transfinite learning network distress in concrete detection training process
Schematic diagram；
Fig. 6 is provided in an embodiment of the present invention to be obtained using based on the transfinite distress in concrete detection method of learning network of multilayer
The experimental result picture arrived.
Specific embodiment
In being described below, in order to illustrate rather than in order to limit, it is proposed that such as tool of particular system structure, technology etc
Body details, to understand thoroughly the embodiment of the present invention.However, it will be clear to one skilled in the art that there is no these specifically
The present invention can also be realized in the other embodiments of details.In other situations, it omits to well-known system, device, electricity
Road and the detailed description of method, in case unnecessary details interferes description of the invention.
It it should be appreciated that ought be special described by the instruction of term " comprising " use in this specification and in the appended claims
Sign, entirety, step, operation, the presence of element and/or component, but be not precluded one or more of the other feature, entirety, step,
Operation, element, component and/or its presence or addition gathered.
It is also understood that the term used in this description of the invention is merely for the sake of the mesh for describing specific embodiment
And be not intended to limit the present invention.As description of the invention and it is used in the attached claims, unless on
Other situations are hereafter clearly indicated, otherwise " one " of singulative, "one" and "the" are intended to include plural form.
It will be further appreciated that the term "and/or" used in description of the invention and the appended claims is
Refer to any combinations and all possible combinations of one or more of the associated item listed, and including these combinations.
As used in this specification and in the appended claims, term " if " can be according to context quilt
Be construed to " when ... " or " once " or " in response to determining " or " in response to detecting ".Similarly, phrase " if it is determined that " or
" if detecting [described condition or event] " can be interpreted to mean according to context " once it is determined that " or " in response to true
It is fixed " or " once detecting [described condition or event] " or " in response to detecting [described condition or event] ".
In order to illustrate technical solutions according to the invention, illustrated below by specific embodiment.
Fig. 1 be it is provided in an embodiment of the present invention based on multilayer transfinite learning network distress in concrete detection realization flow
Figure, Fig. 4 be it is provided in an embodiment of the present invention based on multilayer transfinite learning network distress in concrete detection method schematic block diagram,
Fig. 5 be it is provided in an embodiment of the present invention based on multilayer transfinite learning network distress in concrete detection training process schematic diagram；
With reference to Fig. 4 and Fig. 5, the method that can be become apparent from, be visually known in the present invention；As shown in Figure 1, the method may include with
Lower step：
Step S101 obtains sample image, and the sample image is pre-processed.
Optionally, it is described that the sample image is pre-processed, including：
The sample image is converted into gray level image；
The gray level image is standardized to obtain standardized images；
The label data of the standardized images is obtained, and utilizes the standardized images and label data structure sample
This image library.
In practical applications, can sample coloured image be converted by gray level image by single channel transformation approach.It is exemplary
, Y=0.3R+0.596G+0.11B, wherein, G represents the green component of coloured image, and R represents the red component of coloured image,
B represents the blue component of coloured image, and Y represents transformed gray value.
The label data of standardized images can be by manually being demarcated, and label data can be divided into crack and non-split
Seam, will be in every standardized images label data deposit sample image library corresponding with its.
Optionally, it is described that the gray level image is standardized to obtain standardized images, including：
Piecemeal processing is carried out to the gray level image, obtains at least one image block；
It carries out the rotation of predetermined angle respectively to described image block, obtains postrotational image block；
The size of the size of image block before rotation and the postrotational image block is adjusted to pre-set dimension, is obtained
Standardized images.
Illustratively, it is Z nonoverlapping image block areas according to 2D planar meshs by gray level image, this Z are not
The image block areas of overlapping can be divided into crannied image block areas and two class of image block areas without crack, thus obtain
Crack image block and non-crack image block.Obtained image block is further carried out to the image rotation of 90 degree (predetermined angle), by
This obtains the image block of more multi-mode；In order to adapt to the image block of different sizes, the ruler of each image block that will obtain
It is very little to be adjusted to 50*50 (pre-set dimension).It is it should be noted that above-mentioned only to how to be standardized to gray level image
To an example of standardized images, number, predetermined angle, the pre-set dimension of the image block in example are an example, and
Logarithm is not specifically limited.
Step S102 carries out feature extraction to pretreated sample image based on default Feature Selection Model, obtains spy
Levy data.
Optionally, it is described that feature extraction is carried out to pretreated sample image based on default Feature Selection Model, it obtains
Characteristic, including：
At least two standardized images are randomly selected from the sample image library, obtain training image；
Vectorization processing is carried out respectively to the training image, obtains image training sample data；
Described image training sample data are input to the default Feature Selection Model, obtain characteristic.
Illustratively, N/2 crack image block and N/2 non-crack image blocks, warp are randomly selected from sample image library
After crossing standardization, the image block of each 50*50 sizes is subjected to vectorization, the image vector of 1*2500 is obtained, most owns at last
Crack image vector is in parallel with non-crack image vector to obtain image training sample data matrixNeed what is illustrated
It is that wherein N is used only to represent the total number of crack image block and non-crack image block, concrete numerical value does not limit.
Optionally, it is described that described image training sample data are input to the default Feature Selection Model, obtain feature
Data, including：
The first layer that described image training sample data are input in the default Feature Selection Model is transfinited, and it is special to learn
Sign extraction network, obtains the first output weight；
The described first output weight is optimized based on the first preset formula, the first output weight after being optimized,
And according to after the optimization first output weight calculation described in first layer transfinite learning characteristic extraction network output, obtain institute
State the fisrt feature data of image training sample data；
The second layer that the fisrt feature data are input in the default Feature Selection Model learning characteristic that transfinites is carried
Network is taken, obtains the second output weight；
The described second output weight is optimized based on the first preset formula, the second output weight after being optimized,
And according to after the optimization second output weight calculation described in the second layer transfinite learning characteristic extraction network output, obtain institute
State the second feature data of image training sample data；
Wherein, first preset formula is：
In formula, the β1To export weight, the H1=G (W1,b1, X) for transfinite learning characteristic extract network hidden layer it is defeated
Go out matrix, the X is the input of learning network of transfiniting, (the W1,b1) for transfinite learning characteristic extract network input hidden layer section
Point parameter, the μ are regularization parameter；When being optimized to the first output weight, the β1For the first output weight, the H1
=G (W1,b1, X) for first layer transfinite learning characteristic extract network hidden layer output matrix, the X be described image training sample
The data, (W1,b1) transfinite the input hidden node parameter of feature extraction learning network for first layer；To the second output weight
When optimizing, the β1For the second output weight, the H1=G (W1,b1, X) for the second layer transfinite learning characteristic extract network
Hidden layer output matrix, the X be the fisrt feature data, (the W1,b1) for the second layer transfinite learning characteristic extract network
Input hidden node parameter.
Optionally, first layer described image training sample data being input in the default Feature Selection Model
The learning characteristic that transfinites extracts network, obtains the first output weight, including：
Obtain the first layer transfinite learning characteristic extraction network hidden node number and input hidden node parameter；
According to described image training sample data, the number of the hidden node and the input hidden node parameter generation
The first layer transfinite learning characteristic extraction network hidden layer output matrix；
Calculating the first layer based on the hidden layer output matrix transfinites the output of learning characteristic extraction network, and according to institute
It states hidden layer output matrix and the transfinite output of learning characteristic extraction network of the first layer calculates the first output weight.
In practical applications, default Feature Selection Model can be two layers of the learning network that transfinites, specifically, can be two
The study autoencoder network (ELM autoencoder networks) that transfinites of layer.The hidden node of the ELM autoencoder networks of first layer and the second layer
Number could be provided as 1000.In addition, the fisrt feature data are input in the default Feature Selection Model
The second layer transfinites learning characteristic extraction network, obtains the method for the second output weight, and described image training sample data are defeated
Enter to the first layer in the default Feature Selection Model transfinite learning characteristic extraction network, obtain the method for the first output weight
Identical, details are not described herein.
It is using 1 norm and defeated to first based on the first preset formula in order to obtain more sparse, compact structure characteristic
Go out weight to optimize.After the first output weight after being optimized, fisrt feature data X is calculated1=X β1 T, wherein, X is figure
As training sample data, β1For the first output weight, X1For fisrt feature data；Similar, based on the first preset formula to second
Output weight optimizes the second output weight after being optimized, and calculates second feature dataWherein, X1For
Fisrt feature data, β1For the second output weight, X2For second feature data.It should be noted that β1In subscript 1 be
For distinguishing the β in the first preset formula and the second preset formula, in other words, β1What is represented is in default Feature Selection Model
Output weight, and β2What is represented is the output weight in default detector；In the first preset formula, weighed when to the first output
When optimizing again, the β1Represent the first output weight, when being optimized to the second output weight, the β1Represent second
Export weight.
Step S103 is trained default detector based on the characteristic, the default detector after being trained.
Optionally, it is described that default detector is trained based on the characteristic, the default detection after being trained
Device, including：
90% characteristic is extracted out at random from the second feature data as feature training sample data；
The feature training sample data are input to the default detector, obtain third output weight, it is described default
Detector includes：Transfinite learning machine grader；
Third output weight is optimized based on the second preset formula, the third output weight after being optimized；
Wherein, second preset formula is：
Minimize:{||β2||2+λ||H2β2-T||2}
In formula, the β2Weight, the H are exported for the third2=G (W2,b2, X ') and it is the hidden of the default detector
Layer the output matrix, (W2,b2) be the default detector input hidden node parameter, the X ＇ for the feature training
Sample data, the T are the corresponding label matrix of the feature training sample data, and the λ is regularization parameter.
In practical applications, default detector can be the learning machine that transfinites, and the learning machine that transfinites is a kind of single hidden layer feedforward god
Through network.90% characteristic is extracted out at random from the second feature data as feature training sample data, feature instruction
Practice sample data and be used for the default detector of training；Remaining 10% characteristic is used as inspection data in second feature data
It tests to the default detector after training, to verify whether default detector meets required precision.It should be noted that
90%, 10% in the present embodiment is a preferable example, it is also an option that other ratios, are not specifically limited herein.
Step S104 carries out image to be detected using the default detector after the training detection of distress in concrete.
In practical applications, multilayer transfinites learning network including presetting Feature Selection Model and default detector, such as Fig. 4 institutes
Show, preset Feature Selection Model and include two layers of ELM autoencoder network (the sparse ELM own codings feature extraction in such as Fig. 4), preset
Detector transfinites learning machine grader (ELM crack areas detection) in such as Fig. 4 for individual layer.When need to image to be detected carry out
During the detection of distress in concrete, it is still necessary to first pre-process image to be detected, then pretreated image is input to more
Layer transfinites in learning network, and the process for feature extraction and the detection of learning network of transfiniting by multilayer obtains final detection knot
Fruit.
Fig. 6 is provided in an embodiment of the present invention to be obtained using based on the transfinite distress in concrete detection method of learning network of multilayer
The experimental result picture arrived.In figure, Fig. 6 (a) is actual photographed image, and Fig. 6 (b) is handmarking's image, and Fig. 6 (c) is according to this
The utilization that inventive embodiments provide is transfinited the distress in concrete detection method obtained crack area of learning network based on multilayer
Testing result figure, it is basically identical with the crack area of handmarking.
The present embodiment is pre-processed by obtaining sample image, and to the sample image；Based on default feature extraction
Model carries out feature extraction to pretreated sample image, obtains characteristic；Based on the characteristic to default detection
Device is trained, the default detector after being trained；Image to be detected is carried out using the default detector after the training
The detection of distress in concrete；Effectively increase the efficiency and precision of distress in concrete detection.
It should be understood that the size of the serial number of each step is not meant to the priority of execution sequence, each process in above-described embodiment
Execution sequence should determine that the implementation process without coping with the embodiment of the present invention forms any limit with its function and internal logic
It is fixed.
Fig. 2 be it is provided in an embodiment of the present invention based on multilayer transfinite learning network distress in concrete detection device signal
Figure, for convenience of description, only shows and the relevant part of the embodiment of the present invention.
It is described to be included based on the transfinite distress in concrete detection device 2 of learning network of multilayer：
Pretreatment unit 21 for obtaining sample image, and pre-processes the sample image.
Feature extraction unit 22 puies forward pretreated sample image progress feature for being based on default Feature Selection Model
It takes, obtains characteristic.
Training unit 23 is trained default detector for being based on the characteristic, default after being trained
Detector.
Detection unit 24, for carrying out distress in concrete to image to be detected using the default detector after the training
Detection.
Optionally, the pretreatment unit 21 includes：
Conversion subunit, for the sample image to be converted into gray level image.
Normalizer unit, for being standardized to obtain standardized images to the gray level image.
Subelement is built, for obtaining the label data of the standardized images, and utilizes the standardized images and institute
State label data structure sample image library.
Optionally, the normalizer unit includes：
Piecemeal module for carrying out piecemeal processing to the gray level image, obtains at least one image block.
Rotary module for carrying out the rotation of predetermined angle respectively to described image block, obtains postrotational image block.
Module is adjusted, for the size of image block before rotating and the size of the postrotational image block to be adjusted to
Pre-set dimension obtains standardized images.
Optionally, the feature extraction unit 22 includes：
First extracts subelement, for randomly selecting at least two standardized images from the sample image library, obtains
Training image.
Vectorization subelement for carrying out vectorization processing respectively to the training image, obtains image number of training
According to.
First input subelement, for described image training sample data to be input to the default Feature Selection Model,
Obtain characteristic.
Optionally, the first input subelement includes：
First input module, for described image training sample data to be input in the default Feature Selection Model
First layer transfinite learning characteristic extraction network, obtain the first output weight.
First optimization module optimizes the described first output weight for being based on the first preset formula, is optimized
Afterwards first output weight, and according to after the optimization first output weight calculation described in first layer transfinite learning characteristic extraction
The output of network obtains the fisrt feature data of described image training sample data.
Second input module, for the fisrt feature data to be input to second in the default Feature Selection Model
Layer transfinite learning characteristic extraction network, obtain the second output weight.
Second optimization module optimizes the described second output weight for being based on the first preset formula, is optimized
Afterwards second output weight, and according to after the optimization second output weight calculation described in the second layer transfinite learning characteristic extraction
The output of network obtains the second feature data of described image training sample data.
Wherein, first preset formula is：
In formula, the β1To export weight, the H1=G (W1,b1, X) and it is the hidden layer output matrix of learning network of transfiniting,
The X is the input of learning network of transfiniting, (the W1,b1) it is the input hidden node parameter of learning network of transfiniting, the μ is
Regularization parameter；When being optimized to the first output weight, the β1For the first output weight, the H1=G (W1,b1, X) be
First layer transfinite learning characteristic extraction network hidden layer output matrix, the X be described image training sample data, (the W1,
b1) for first layer transfinite learning characteristic extract network input hidden node parameter；When being optimized to the second output weight, institute
State β1For the second output weight, the H1=G (W1,b1, X) for the second layer transfinite learning characteristic extract network hidden layer export square
Battle array, the X be the fisrt feature data, (the W1,b1) for the second layer transfinite learning characteristic extract network input hidden layer section
Point parameter.
Optionally, first input module includes：
Acquisition submodule transfinites the number of hidden node of learning characteristic extraction network and defeated for obtaining the first layer
Enter hidden node parameter.
Submodule is generated, for the number according to described image training sample data, the hidden node and the input
Hidden node parameter, which generates the first layer, to transfinite the hidden layer output matrix of learning characteristic extraction network.
Computational submodule, calculating the first layer for being based on the hidden layer output matrix transfinites learning characteristic extraction network
Output, and calculate described the according to the transfinite output of learning characteristic extraction network of the hidden layer output matrix and the first layer
One output weight.
Optionally, the training unit 23 includes：
Second extracts subelement, for randomly selecting 90% characteristic from the second feature data as feature
Training sample data.
Second input subelement, for the feature training sample data to be input to the default detector, obtains the
Three output weights, the default detector include：Transfinite learning machine grader.
Optimize subelement, third output weight is optimized for being based on the second preset formula, after obtaining optimization
Third output weight.
Wherein, second preset formula is：
Minimize:{||β2||2+λ||H2β2-T||2}
In formula, the β2Weight, the H are exported for the third2=G (W2,b2, X ') and it is the hidden of the default detector
Layer the output matrix, (W2,b2) be the default detector input hidden node parameter, the X ＇ for the feature training
Sample data, the T are the corresponding label matrix of the feature training sample data, and the λ is regularization parameter.
It is apparent to those skilled in the art that for convenience of description and succinctly, only with above-mentioned each work(
Can unit, module division progress for example, in practical application, can be as needed and by above-mentioned function distribution by different
Functional unit, module are completed, i.e., the internal structure of described device are divided into different functional units or module, more than completion
The all or part of function of description.Each functional unit, module in embodiment can be integrated in a processing unit, also may be used
To be that each unit is individually physically present, can also two or more units integrate in a unit, it is above-mentioned integrated
The form that hardware had both may be used in unit is realized, can also be realized in the form of SFU software functional unit.In addition, each function list
Member, the specific name of module are not limited to the protection domain of the application also only to facilitate mutually distinguish.Above system
The specific work process of middle unit, module can refer to the corresponding process in preceding method embodiment, and details are not described herein.
Fig. 3 is the schematic diagram of terminal device provided in an embodiment of the present invention.As shown in figure 3, the terminal device 3 of the embodiment
Including：Processor 30, memory 31 and it is stored in the calculating that can be run in the memory 31 and on the processor 30
Machine program 32.The processor 30 realizes the training of above-mentioned each distress in concrete detector when performing the computer program 32
Step in embodiment of the method, such as step S101 to S104 shown in FIG. 1.Alternatively, the processor 30 performs the calculating
The function of each module/unit in above-mentioned each device embodiment, such as the work(of module 21 to 24 shown in Fig. 2 are realized during machine program 32
Energy.
Illustratively, the computer program 32 can be divided into one or more module/units, it is one or
Multiple module/units are stored in the memory 31, and are performed by the processor 30, to complete the present invention.Described one
A or multiple module/units can be the series of computation machine program instruction section that can complete specific function, which is used for
Implementation procedure of the computer program 32 in the terminal device 3 is described.For example, the computer program 32 can be divided
Pretreatment unit, feature extraction unit, training unit, detection unit are cut into, each unit concrete function is as follows：
Pretreatment unit for obtaining sample image, and pre-processes the sample image.
Feature extraction unit puies forward pretreated sample image progress feature for being based on default Feature Selection Model
It takes, obtains characteristic.
Training unit is trained default detector for being based on the characteristic, the default inspection after being trained
Survey device.
Detection unit, for carrying out the inspection of distress in concrete to image to be detected using the default detector after the training
It surveys.
Optionally, the pretreatment unit includes：
Conversion subunit, for the sample image to be converted into gray level image.
Normalizer unit, for being standardized to obtain standardized images to the gray level image.
Subelement is built, for obtaining the label data of the standardized images, and utilizes the standardized images and institute
State label data structure sample image library.
Optionally, the normalizer unit includes：
Piecemeal module for carrying out piecemeal processing to the gray level image, obtains at least one image block.
Rotary module for carrying out the rotation of predetermined angle respectively to described image block, obtains postrotational image block.
Module is adjusted, for the size of image block before rotating and the size of the postrotational image block to be adjusted to
Pre-set dimension obtains standardized images.
Optionally, the feature extraction unit includes：
First extracts subelement, for randomly selecting at least two standardized images from the sample image library, obtains
Training image.
Vectorization subelement for carrying out vectorization processing respectively to the training image, obtains image number of training
According to.
First input subelement, for described image training sample data to be input to the default Feature Selection Model,
Obtain characteristic.
Optionally, the first input subelement includes：
First input module, for described image training sample data to be input in the default Feature Selection Model
First layer transfinite learning characteristic extraction network, obtain the first output weight.
First optimization module optimizes the described first output weight for being based on the first preset formula, is optimized
Afterwards first output weight, and according to after the optimization first output weight calculation described in first layer transfinite learning characteristic extraction
The output of network obtains the fisrt feature data of described image training sample data.
Second input module, for the fisrt feature data to be input to second in the default Feature Selection Model
Layer transfinite learning characteristic extraction network, obtain the second output weight.
Second optimization module optimizes the described second output weight for being based on the first preset formula, is optimized
Afterwards second output weight, and according to after the optimization second output weight calculation described in the second layer transfinite learning characteristic extraction
The output of network obtains the second feature data of described image training sample data.
Wherein, first preset formula is：
In formula, the β1To export weight, the H1=G (W1,b1, X) and it is the hidden layer output matrix of learning network of transfiniting,
The X is the input of learning network of transfiniting, (the W1,b1) it is the input hidden node parameter of learning network of transfiniting, the μ is
Regularization parameter；When being optimized to the first output weight, the β1For the first output weight, the H1=G (W1,b1, X) be
First layer transfinite learning characteristic extraction network hidden layer output matrix, the X be described image training sample data, (the W1,
b1) for first layer transfinite learning characteristic extract network input hidden node parameter；When being optimized to the second output weight, institute
State β1For the second output weight, the H1=G (W1,b1, X) for the second layer transfinite learning characteristic extract network hidden layer export square
Battle array, the X be the fisrt feature data, (the W1,b1) for the second layer transfinite learning characteristic extract network input hidden layer section
Point parameter.
Optionally, first input module includes：
Acquisition submodule transfinites the number of hidden node of learning characteristic extraction network and defeated for obtaining the first layer
Enter hidden node parameter.
Submodule is generated, for the number according to described image training sample data, the hidden node and the input
Hidden node parameter, which generates the first layer, to transfinite the hidden layer output matrix of learning characteristic extraction network.
Computational submodule, calculating the first layer for being based on the hidden layer output matrix transfinites learning characteristic extraction network
Output, and calculate described the according to the transfinite output of learning characteristic extraction network of the hidden layer output matrix and the first layer
One output weight.
Optionally, the training unit includes：
Second extracts subelement, for randomly selecting 90% characteristic from the second feature data as feature
Training sample data.
Second input subelement, for the feature training sample data to be input to the default detector, obtains the
Three output weights, the default detector include：Transfinite learning machine grader.
Optimize subelement, third output weight is optimized for being based on the second preset formula, after obtaining optimization
Third output weight.
Wherein, second preset formula is：
Minimize:{||β2||2+λ||H2β2-T||2}
In formula, the β2Weight, the H are exported for the third2=G (W2,b2, X ') and it is the hidden of the default detector
Layer the output matrix, (W2,b2) be the default detector input hidden node parameter, the X ＇ for the feature training
Sample data, the T are the corresponding label matrix of the feature training sample data, and the λ is regularization parameter.
The terminal device 3 can be that the calculating such as desktop PC, notebook, palm PC and cloud server are set
It is standby.The terminal device may include, but be not limited only to, processor 30, memory 31.It will be understood by those skilled in the art that Fig. 3
The only example of terminal device 3 does not form the restriction to terminal device 3, can include than illustrating more or fewer portions
Part either combines certain components or different components, such as the terminal device can also include input-output equipment, net
Network access device, bus etc..
Alleged processor 30 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), application-specific integrated circuit
(Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor can also be any conventional processor
Deng.
The memory 31 can be the internal storage unit of the terminal device 3, such as the hard disk of terminal device 3 or interior
It deposits.The memory 31 can also be the External memory equipment of the terminal device 3, such as be equipped on the terminal device 3
Plug-in type hard disk, intelligent memory card (Smart Media Card, SMC), secure digital (Secure Digital, SD) card dodge
Deposit card (Flash Card) etc..Further, the memory 31 can also both include the storage inside list of the terminal device 3
Member also includes External memory equipment.The memory 31 is used to store needed for the computer program and the terminal device
Other programs and data.The memory 31 can be also used for temporarily storing the data that has exported or will export.
In the above-described embodiments, it all emphasizes particularly on different fields to the description of each embodiment, is not described in detail or remembers in some embodiment
The part of load may refer to the associated description of other embodiments.
Those of ordinary skill in the art may realize that each exemplary lists described with reference to the embodiments described herein
Member and algorithm steps can be realized with the combination of electronic hardware or computer software and electronic hardware.These functions are actually
It is performed with hardware or software mode, specific application and design constraint depending on technical solution.Professional technician
Described function can be realized using distinct methods to each specific application, but this realization is it is not considered that exceed
The scope of the present invention.
In embodiment provided by the present invention, it should be understood that disclosed device/terminal device and method, it can be with
It realizes by another way.For example, device described above/terminal device embodiment is only schematical, for example, institute
The division of module or unit is stated, only a kind of division of logic function can have other dividing mode in actual implementation, such as
Multiple units or component may be combined or can be integrated into another system or some features can be ignored or does not perform.Separately
A bit, shown or discussed mutual coupling or direct-coupling or communication connection can be by some interfaces, device
Or the INDIRECT COUPLING of unit or communication connection, can be electrical, machinery or other forms.
The unit illustrated as separating component may or may not be physically separate, be shown as unit
The component shown may or may not be physical unit, you can be located at a place or can also be distributed to multiple
In network element.Some or all of unit therein can be selected according to the actual needs to realize the mesh of this embodiment scheme
's.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, it can also
That each unit is individually physically present, can also two or more units integrate in a unit.Above-mentioned integrated list
The form that hardware had both may be used in member is realized, can also be realized in the form of SFU software functional unit.
If the integrated module/unit realized in the form of SFU software functional unit and be independent product sale or
In use, it can be stored in a computer read/write memory medium.Based on such understanding, the present invention realizes above-mentioned implementation
All or part of flow in example method, can also instruct relevant hardware to complete, the meter by computer program
Calculation machine program can be stored in a computer readable storage medium, the computer program when being executed by processor, it can be achieved that on
The step of stating each embodiment of the method.Wherein, the computer program includes computer program code, the computer program generation
Code can be source code form, object identification code form, executable file or certain intermediate forms etc..The computer-readable medium
It can include：Any entity of the computer program code or device, recording medium, USB flash disk, mobile hard disk, magnetic can be carried
Dish, CD, computer storage, read-only memory (ROM, Read-Only Memory), random access memory (RAM,
Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium etc..It is it should be noted that described
The content that computer-readable medium includes can carry out appropriate increasing according to legislation in jurisdiction and the requirement of patent practice
Subtract, such as in certain jurisdictions, according to legislation and patent practice, computer-readable medium do not include be electric carrier signal and
Telecommunication signal.
Embodiment described above is merely illustrative of the technical solution of the present invention, rather than its limitations；Although with reference to aforementioned reality
Example is applied the present invention is described in detail, it will be understood by those of ordinary skill in the art that：It still can be to aforementioned each
Technical solution recorded in embodiment modifies or carries out equivalent replacement to which part technical characteristic；And these are changed
Or replace, the spirit and scope for various embodiments of the present invention technical solution that it does not separate the essence of the corresponding technical solution should all
It is included within protection scope of the present invention.
Claims (10)
1. a kind of distress in concrete detection method for the learning network that transfinited based on multilayer, which is characterized in that including：
Sample image is obtained, and the sample image is pre-processed；
Feature extraction is carried out to pretreated sample image based on default Feature Selection Model, obtains characteristic；
Default detector is trained based on the characteristic, the default detector after being trained；
The detection of distress in concrete is carried out to image to be detected using the default detector after the training.
2. the distress in concrete detection method for the learning network that transfinited as described in claim 1 based on multilayer, which is characterized in that institute
It states and the sample image is pre-processed, including：
The sample image is converted into gray level image；
The gray level image is standardized to obtain standardized images；
The label data of the standardized images is obtained, and utilizes the standardized images and label data structure sample graph
As library.
3. the distress in concrete detection method for the learning network that transfinited as claimed in claim 2 based on multilayer, which is characterized in that institute
It states and the gray level image is standardized to obtain standardized images, including：
Piecemeal processing is carried out to the gray level image, obtains at least one image block；
It carries out the rotation of predetermined angle respectively to described image block, obtains postrotational image block；
The size of the size of image block before rotation and the postrotational image block is adjusted to pre-set dimension, obtains standard
Change image.
4. the distress in concrete detection method for the learning network that transfinited as claimed in claim 3 based on multilayer, which is characterized in that institute
It states and feature extraction is carried out to pretreated sample image based on default Feature Selection Model, obtain characteristic, including：
At least two standardized images are randomly selected from the sample image library, obtain training image；
Vectorization processing is carried out respectively to the training image, obtains image training sample data；
Described image training sample data are input to the default Feature Selection Model, obtain characteristic.
5. the distress in concrete detection method for the learning network that transfinited as claimed in claim 4 based on multilayer, which is characterized in that institute
It states and described image training sample data is input to the default Feature Selection Model, obtain characteristic, including：
The first layer that described image training sample data are input in the default Feature Selection Model learning characteristic that transfinites is carried
Network is taken, obtains the first output weight；
The described first output weight is optimized based on the first preset formula, the first output weight after being optimized, and root
According to after the optimization first output weight calculation described in first layer transfinite learning characteristic extraction network output, obtain the figure
As the fisrt feature data of training sample data；
By the second layer that the fisrt feature data are input in the default Feature Selection Model transfinite learning characteristic extraction net
Network obtains the second output weight；
The described second output weight is optimized based on the first preset formula, the second output weight after being optimized, and root
According to after the optimization second output weight calculation described in the second layer transfinite learning characteristic extraction network output, obtain the figure
As the second feature data of training sample data；
Wherein, first preset formula is：
In formula, the β1To export weight, the H1=G (W1,b1, X) and it is the hidden layer output matrix of learning network of transfiniting, the X
For the input for the learning network that the transfinites, (W1,b1) it is the input hidden node parameter of learning network of transfiniting, the μ is regularization
Parameter；When being optimized to the first output weight, the β1For the first output weight, the H1=G (W1,b1, X) and it is first layer
The learning characteristic that transfinites extract network hidden layer output matrix, the X be described image training sample data, (the W1,b1) it is the
The input hidden node parameter of the one layer of learning characteristic extraction network that transfinites；When being optimized to the second output weight, the β1For
Second output weight, the H1=G (W1,b1, X) for the second layer transfinite learning characteristic extract network hidden layer output matrix, it is described
X be the fisrt feature data, (the W1,b1) for the second layer transfinite learning characteristic extract network input hidden node parameter.
6. the distress in concrete detection method for the learning network that transfinited as claimed in claim 5 based on multilayer, which is characterized in that institute
State by the first layer that described image training sample data are input in the default Feature Selection Model transfinite learning characteristic extraction
Network obtains the first output weight, including：
Obtain the first layer transfinite learning characteristic extraction network hidden node number and input hidden node parameter；
According to described image training sample data, the number of the hidden node and the input hidden node parameter generation
First layer transfinite learning characteristic extraction network hidden layer output matrix；
Calculating the first layer based on the hidden layer output matrix transfinites the output of learning characteristic extraction network, and according to described hidden
Layer output matrix and the first layer transfinite learning characteristic extraction network output calculate it is described first output weight.
7. the distress in concrete detection method for the learning network that transfinited as claimed in claim 5 based on multilayer, which is characterized in that institute
It states and default detector is trained based on the characteristic, the default detector after being trained, including：
90% characteristic is randomly selected from the second feature data as feature training sample data；
The feature training sample data are input to the default detector, obtain third output weight, the default detection
Device includes：Transfinite learning machine grader；
Third output weight is optimized based on the second preset formula, the third output weight after being optimized；
Wherein, second preset formula is：
Minimize:{||β2||2+λ||H2β2-T||2}
In formula, the β2Weight, the H are exported for the third2=G (W2,b2, X ') for the default detector hidden layer export
The matrix, (W2,b2) be the default detector input hidden node parameter, the X ＇ be the feature number of training
According to the T is the corresponding label matrix of the feature training sample data, and the λ is regularization parameter.
8. a kind of distress in concrete detection device for the learning network that transfinited based on multilayer, which is characterized in that including：
Pretreatment unit for obtaining sample image, and pre-processes the sample image；
Feature extraction unit carries out feature extraction to pretreated sample image for being based on default Feature Selection Model, obtains
To characteristic；
Training unit is trained default detector for being based on the characteristic, the default detector after being trained；
Detection unit, for carrying out the detection of distress in concrete to image to be detected using the default detector after the training.
9. a kind of terminal device, including memory, processor and it is stored in the memory and can be on the processor
The computer program of operation, which is characterized in that the processor realizes such as claim 1 to 7 when performing the computer program
The step of any one the method.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists
In when the computer program is executed by processor the step of realization such as any one of claim 1 to 7 the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810031254.8A CN108182681A (en) | 2018-01-12 | 2018-01-12 | A kind of distress in concrete detection method for the learning network that transfinited based on multilayer |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810031254.8A CN108182681A (en) | 2018-01-12 | 2018-01-12 | A kind of distress in concrete detection method for the learning network that transfinited based on multilayer |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108182681A true CN108182681A (en) | 2018-06-19 |
Family
ID=62550430
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810031254.8A Pending CN108182681A (en) | 2018-01-12 | 2018-01-12 | A kind of distress in concrete detection method for the learning network that transfinited based on multilayer |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108182681A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109358185A (en) * | 2018-10-10 | 2019-02-19 | 青岛科技大学 | Rubber compounding performance prediction model and prediction technique based on extreme learning machine |
| CN111257341A (en) * | 2020-03-30 | 2020-06-09 | 河海大学常州校区 | Underwater building crack detection method based on multi-scale features and stacked full convolution network |
| CN115346127A (en) * | 2022-10-20 | 2022-11-15 | 成都大汇物联科技有限公司 | Dam safety detection method and system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897737A (en) * | 2017-01-24 | 2017-06-27 | 北京理工大学 | A kind of high-spectrum remote sensing terrain classification method based on the learning machine that transfinites |
| CN107480611A (en) * | 2017-07-31 | 2017-12-15 | 浙江大学 | A kind of crack identification method based on deep learning convolutional neural networks |
-
2018
- 2018-01-12 CN CN201810031254.8A patent/CN108182681A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897737A (en) * | 2017-01-24 | 2017-06-27 | 北京理工大学 | A kind of high-spectrum remote sensing terrain classification method based on the learning machine that transfinites |
| CN107480611A (en) * | 2017-07-31 | 2017-12-15 | 浙江大学 | A kind of crack identification method based on deep learning convolutional neural networks |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109358185A (en) * | 2018-10-10 | 2019-02-19 | 青岛科技大学 | Rubber compounding performance prediction model and prediction technique based on extreme learning machine |
| CN111257341A (en) * | 2020-03-30 | 2020-06-09 | 河海大学常州校区 | Underwater building crack detection method based on multi-scale features and stacked full convolution network |
| CN115346127A (en) * | 2022-10-20 | 2022-11-15 | 成都大汇物联科技有限公司 | Dam safety detection method and system |
| CN115346127B (en) * | 2022-10-20 | 2023-01-24 | 成都大汇物联科技有限公司 | Dam safety detection method and system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109765462A (en) | Fault detection method, device and the terminal device of transmission line of electricity | |
| CN109493347A (en) | The method and system that the object of sparse distribution is split in the picture | |
| CN107704625A (en) | Fields match method and apparatus | |
| CN109785334A (en) | Cardiac magnetic resonance images dividing method, device, terminal device and storage medium | |
| CN108182681A (en) | A kind of distress in concrete detection method for the learning network that transfinited based on multilayer | |
| CN105488515A (en) | Method for training convolutional neural network classifier and image processing device | |
| CN107895311A (en) | A kind of method and apparatus of order matching distribution | |
| CN110111885B (en) | Attribute prediction method, attribute prediction device, computer equipment and computer readable storage medium | |
| CN107480702A (en) | Towards the feature selecting and Feature fusion of the identification of HCC pathological images | |
| CN109300119A (en) | Detection method, detection device and the terminal device in steel structure surface corrosion region | |
| CN108564569B (en) | A kind of distress in concrete detection method and device based on multicore classification learning | |
| WO2009035737A2 (en) | Adaptive refinement tools for tetrahedral unstructured grids | |
| CN113468382A (en) | Multi-party loop detection method, device and related equipment based on knowledge federation | |
| Wang et al. | ProMutator: Detecting vulnerable price oracles in DeFi by mutated transactions | |
| CN108447047A (en) | Acid-fast bacilli detection method and device | |
| CN109117475A (en) | A kind of method and relevant device of text rewriting | |
| Weber et al. | Detecting critical regions in scalar fields | |
| CN114781650A (en) | Data processing method, device, equipment and storage medium | |
| CN109559206A (en) | A kind of regional enterprises Credit Evaluation System method, apparatus and terminal device | |
| CN108764301B (en) | A kind of distress in concrete detection method based on reversed rarefaction representation | |
| CN107391872A (en) | The method and device that PDMS threedimensional models use in software | |
| CN108198173A (en) | A kind of online test method, device and the terminal device in distress in concrete region | |
| CN107633506A (en) | A kind of image symmetrical characteristic detection method, device and terminal device | |
| CN107730707A (en) | A kind of false distinguishing method and device of valuable document | |
| CN109918237A (en) | Abnormal network layer determines method and Related product |
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 |
Application publication date: 20180619 |
|
| RJ01 | Rejection of invention patent application after publication |