CN110349134A - A kind of piping disease image classification method based on multi-tag convolutional neural networks - Google Patents
A kind of piping disease image classification method based on multi-tag convolutional neural networks Download PDFInfo
- Publication number
- CN110349134A CN110349134A CN201910565360.9A CN201910565360A CN110349134A CN 110349134 A CN110349134 A CN 110349134A CN 201910565360 A CN201910565360 A CN 201910565360A CN 110349134 A CN110349134 A CN 110349134A
- Authority
- CN
- China
- Prior art keywords
- neural networks
- convolutional neural
- tag
- layer
- image
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
- Endoscopes (AREA)
Abstract
The present invention discloses a kind of piping disease image classification method based on multi-tag convolutional neural networks, and steps are as follows for the piping disease image classification method: step 1: peeping detection video in collection conduit, extract the picture frame in video；Step 2: calculating the timestamp feature of each image；Step 3: a part of picture frame that step 1 is collected into being sent into multi-tag convolutional neural networks model and is trained, the multi-tag convolutional neural networks model that can correctly sort out piping disease type is obtained；Step 4: with trained multi-tag convolutional neural networks model inspection pipe endoscopic image to be detected, then multi-tag convolutional neural networks model can export one-hot coding, existing piping disease type is determined according to one-hot coding, multi-tag classification layer is increased on the basis of existing Inception-ResNet-v2 network, realizes the classification feature of pipelines disease geo-radar image.
Description
Technical field
The present invention relates to computer digital image processing and deep learning algorithm fields based on convolutional neural networks, especially
It is related to a kind of piping disease image classification method based on multi-tag convolutional neural networks.
Background technique
There are two types of the acquisition modes of currently used pipe endoscopic image: one is be based on closed-circuit television detection technique (Cl
OseCircuitTelevisionInspection, CCTV) detecting robot of pipe technology, one is quickly regarded based on pipeline
The pipeline periscope technology of figure detection technique (PipeQuickViewInspection, QV).
Processing method after obtaining image in the prior art is at present still based on artificial observation, then carries out Classification and Identification.
Artificial observation will appear following difficult points:
The increase of pipe-line construction length will lead to the generation that frequency and image data are pried through in huge, with detecting robot of pipe
For, operation maximum speed at least needs to shoot 5 minutes videos in 2.5-3m/s, the pipeline for being 1000m for length,
It is shot according to the frame per second of 30fps, then can generate 9000 pictures, so huge data volume is inevitable by manual inspection screening
Testing result can be caused inaccurate due to factors such as work fatigue, subjective consciousnesses；So just needing to utilize computer digital image
Processing technique realizes the automatic Classification and Identification of disease of pipe endoscopic image.
Pipeline is relative complex throughout the disease species that wide, type is more, generates, but pipe detection is that service conduit is normally transported
The important means of battalion, from artificial dive inspection, to the detection of the special instruments such as Ground Penetrating Radar, sonar, developing deeply is to current using pipe
Pipeline robot or pipeline periscope obtain the mode that pipe endoscopic image is observed, and safety and universality there has been larger
It is promoted, and digital image processing techniques successfully come true the automatic identification of several piping diseases with monitoring.
There are two ways to classifying currently with computer digital image processing technique to pipe endoscopic image is main
Scheme:
1. subjectively obtains the characteristics of image of specific diseases type, classified using traditional Machine learning classifiers:
<1>is directed to the rupture occurred in pipeline, mismatch, disconnection disease geo-radar image, extracts horizontal and diagonal Wavelet Component, figure
As entropy, symbiosis correlation are characterized, and are combined into feature vector；
<2>is directed to the dirt deposition occurred in pipeline, tree root disease geo-radar image, extracts profile circularity, compact degree, camber
It is characterized, and is combined into feature vector；
<3>extracts angular second moment, image entropy, phase for disease geo-radar images such as blocking, the accesses of illegal pipeline occurred in pipeline
Guan Xing, diversity, contrast, uniformity are characterized, and are combined into feature vector；
<4>is trained in features described above vector feeding support vector machines, random forest, full Connection Neural Network,
Realize the classification of disease species.
Classify 2. objectively extracts image abstraction feature using convolutional neural networks:
<1>is by all pipe endoscopic image normalizations taken at the resolution ratio having a size of 256 × 256；
<2>ruptures three kinds of diseases for tree root invasion common in piping disease, dirt deposition, tube wall, by above-mentioned picture
It is sent into convolutional neural networks as shown in Figure 3, is trained；Input layer × 3 indicate that the image of input is color image in figure,
Shared three channels of red, green, blue, the middle size Expressing of each convolutional layer, by convolution algorithm, the size and port number of original image.
Such as 128 × 128 × 32 indicate pass through first convolution algorithm, original image become resolution ratio be 128 × 128, and have 32
The image of a data channel, this image are known as characteristic pattern.
<3>characteristic pattern is converted to the vector of particular dimension, such as 1024 by the digital representation in the full articulamentum of, and being exactly will
Characteristic pattern becomes the vector of 1024 dimensions.The 3 of output layer indicate categorization vector, are identified using one-hot coding；
<4>for example inputs one there are the disease geo-radar image of tree root invasion, then output layer will be exported shaped like [0,0,1]
Vector；Similarly dirt deposition will export [0,1,0], and tube wall rupture will export [1,0,0].
But existing computer digital image processing technique needs subjectively to obtain the corresponding feature of disease geo-radar image, excessively
Rely on the quality of shooting image；For dark moist, the inner wall of the pipe of environment complexity, the image for obtaining high quality exists centainly
It is difficult；
And the differentiation susceptibility of different disease species characteristics of image is different, select which type of feature as classification according to
According to certain pipe detection working experience is needed, the development of image detecting technique is limited；
In addition existing convolutional neural networks due to its model it is simple, classification type is few, can not cope with complicated pipeline disease
Evil type classification problem, and classification accuracy is lower.
Summary of the invention
The present invention increases multi-tag classification layer on the basis of existing Inception-ResNet-v2 network, realizes more
The classification feature of kind piping disease image.
A kind of piping disease image classification method based on multi-tag convolutional neural networks, the piping disease image classification
Method is as follows:
Step 1: peeping detection video in collection conduit, extract the picture frame in pipe endoscopic detection video；
Step 2: calculating the timestamp feature of each image；
Step 3: a part of picture frame that step 1 is collected into being sent into multi-tag convolutional neural networks model and is instructed
Practice, obtains the multi-tag convolutional neural networks model that can correctly sort out piping disease type；
Step 4: then to mark with trained multi-tag convolutional neural networks model inspection pipe endoscopic image to be detected more
Label convolutional neural networks model can export one-hot coding, determine existing piping disease type according to one-hot coding.
Preferably, the multi-tag convolutional neural networks model includes upper layer Inception-ResNet-v2 network structure
With lower layer's multi-tag classification layer, the upper layer Inception-ResNet-v2 network structure further includes random deactivating layer.
Preferably, the processing step of lower layer's multi-tag classification layer are as follows:
Step 1: the feature vector that the random deactivating layer in the Inception-ResNet-v2 network structure of upper layer exports
One timestamp feature TimeFeature of middle addition,
Step 2: the feature vector for being added to timestamp feature in the first step is carried out dimensionality reduction activation processing, middle latitude is obtained
Angle value,
Step 3: the middle latitude value in second step is continued dimension-reduction treatment, one-hot coding is finally exported.
Preferably, the calculation formula of the timestamp feature TimeFeature are as follows:
Wherein Current_Frame_Index is detected to pipe endoscopic image to be detected in entire pipe endoscopic to be described
Serial number the location of in video, All_Frames_Num are the totalframes that entire pipe endoscopic detects video.
Preferably, the piping disease type include: hollow water, tree root, deposition, sundries, fouling, closure, mild corrosion,
Moderate corrosion, the disconnection of severe burn into, rupture, mismatch, deformation, invasion, slight leakage, moderate leakage and severe leakage.
Beneficial effect obtained by the present invention is:
1. uses the convolutional neural networks of model complexity, it is not necessary to subjectively extract feature, improve the accurate of disease classification
Rate；
2. combines the timestamp feature of image to be detected, the type of characteristic of division is enriched, is had using timestamp feature
Help will test non-disease geo-radar image frame in video and unrelated images frame (such as pipe detection periscope or pipe detection machine
The unrelated images that device people takes when being put into well or recycled by staff) it distinguishes；
3. can distinguish 17 kinds of piping disease types, all disease species of current urban duct kind are almost enumerated, are answered
It is wider with range.
Detailed description of the invention
From following description with reference to the accompanying drawings it will be further appreciated that the present invention.Component in figure is not drawn necessarily to scale,
But it focuses on and shows in the principle of embodiment；In different views, identical appended drawing reference specifies corresponding part.
Fig. 1 is network structure block diagram of the invention；
Fig. 2 is the process flow diagram of lower layer's multi-tag classification layer of the invention；
Fig. 3 is convolutional neural networks structure chart in the prior art.
Specific embodiment
In order to enable the objectives, technical solutions, and advantages of the present invention are more clearly understood, below in conjunction with embodiment, to this
Invention is further elaborated；It should be appreciated that described herein, the specific embodiments are only for explaining the present invention, and does not have to
It is of the invention in limiting.To those skilled in the art, after access is described in detail below, other systems of the present embodiment
System, method and/or feature will become obvious.All such additional systems, method, feature and advantage are intended to be included in
It in this specification, is included within the scope of the invention, and by the protection of the appended claims.In description described in detail below
The other feature of the disclosed embodiments, and these characteristic roots will be apparent according to described in detail below.
The same or similar label correspond to the same or similar components in the attached drawing of the embodiment of the present invention；It is retouched in of the invention
In stating, it is to be understood that if the orientation or positional relationship for having the instructions such as term " on ", "lower", "left", "right" is based on attached drawing
Shown in orientation or positional relationship, be merely for convenience of description of the present invention and simplification of the description, rather than indication or suggestion is signified
Device or component must have a particular orientation, be constructed and operated in a specific orientation, therefore positional relationship is described in attached drawing
Term only for illustration, should not be understood as the limitation to this patent, for the ordinary skill in the art, can
To understand the concrete meaning of above-mentioned term as the case may be.
Embodiment one:
A kind of piping disease image classification method based on multi-tag convolutional neural networks, the piping disease image classification
Method and step is as follows:
Step 1: peeping detection video in collection conduit, extract the picture frame in pipe endoscopic detection video；
Step 2: calculating the timestamp feature of each image；
Step 3: a part of picture frame that step 1 is collected into being sent into multi-tag convolutional neural networks model and is instructed
Practice, obtains the multi-tag convolutional neural networks model that can correctly sort out piping disease type；
Step 4: then to mark with trained multi-tag convolutional neural networks model inspection pipe endoscopic image to be detected more
Label convolutional neural networks model can export one-hot coding, determine existing piping disease type according to one-hot coding.
The multi-tag convolutional neural networks model includes that upper layer Inception-ResNet-v2 network structure and lower layer are more
Labeling layer, the upper layer Inception-ResNet-v2 network structure further include random deactivating layer.
The processing step of lower layer's multi-tag classification layer are as follows:
Step 1: the feature vector that the random deactivating layer in the Inception-ResNet-v2 network structure of upper layer exports
One timestamp feature TimeFeature of middle addition；
Step 2: the feature vector for being added to timestamp feature in the first step is carried out dimensionality reduction activation processing, middle latitude is obtained
Angle value；
Step 3: the middle latitude value in second step is continued dimension-reduction treatment, one-hot coding is finally exported.
The calculation formula of the timestamp feature TimeFeature are as follows:
Wherein Current_Frame_Index is detected to pipe endoscopic image to be detected in entire pipe endoscopic to be described
Serial number the location of in video, All_Frames_Num are the totalframes that entire pipe endoscopic detects video.
The piping disease type includes: hollow water, tree root, deposition, sundries, fouling, closure, mild corrosion, moderate corruption
Erosion, the disconnection of severe burn into, rupture, mismatch, deformation, invasion, slight leakage, moderate leakage and severe leakage.
Embodiment two:
The present embodiment is based on the above embodiments further detailed content, so that those skilled in the art is more
Realization of the invention is clearly understood, wherein the present invention increases on the basis of existing Inception-ResNet-v2 network
Multi-tag classification layer, realizes the classification feature of pipelines disease geo-radar image, the present invention replaces with SoftMax classifier therein
Multi-tag classification layer is answered so that Inception-ResNet-v2 has multi-tag classification feature to detect to the maximum extent
Some disease species, as shown in Figure 2.
The random deactivating layer of its Inception-ResNet-v2 network structure at the middle and upper levels can export the spy of 1792 dimensions
Vector is levied, corresponding to the X layer in Fig. 2.The present invention is added to one-dimensional characteristic again behind this vector, such as the light gray in Fig. 2
Shown in color frame, this feature is the timestamp feature of image to be detected, is calculated using formula 1,
Current_Frame_Index in formula is examined to pipe endoscopic image to be detected in entire pipe endoscopic to be described
Location serial number in video is surveyed, All_Frames_Num is the totalframes that entire pipe endoscopic detects video, can from formula
To see that its value range is 0~1.
H and C layers in Fig. 2 be exactly multi-tag classification layer chief component, wherein handled comprising 1024 activation for H layer
Unit, that is, circle；C layers include 17 disease species marks using one-hot coding mode output, are from left to right respectively as follows:
[hollow water, tree root, deposition, sundries, fouling, closure, mild corrosion, moderate corrosion, severe burn into disconnect, are broken C=
It splits, mismatch, deformation, invasion, slight leakage, moderate leakage, severe leakage].
For example, if there are tree root and deposition disease in image, C layers will output vector [0,1,1,0,0,0,0,0,
0,0,0,0,0,0,0,0,0], if there is no disease, will output vector [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
0,0]。
Embodiment three:
The present embodiment is based on the above embodiments further detailed content, so that those skilled in the art is more
Realization of the invention is clearly understood, wherein the present invention increases on the basis of existing Inception-ResNet-v2 network
Multi-tag classification layer, realizes the classification feature of pipelines disease geo-radar image, the present invention replaces with SoftMax classifier therein
Multi-tag classification layer is answered so that Inception-ResNet-v2 has multi-tag classification feature to detect to the maximum extent
Some disease species.
As shown in Fig. 2, wherein X layers be shown in random deactivating layer in the Inception-ResNet-v2 network structure of upper layer
Output vector.
The random deactivating layer of original I nception-ResNet-v2 network structure only exports preceding 1792 dimension information, the present invention
The temporal information of a dimension is increased on this basis.
The X layers of process converted to H layers, are the forward calculation processes of full Connection Neural Network, in order to by 1792+1
The vector of dimension suitably reduces dimension, reduces into 1024 dimensions.For convenience of description, several symbols are properly added in figure, wherein X1
~X1793Represent a newly-increased dimension of the output of the random deactivating layer of original I nception-ResNet-v2+herein；H1~
X1024It represents after H layers of activation, 1793 dimensional vectors is become to each dimension values of 1024 dimensional vectors, C1~C17Indicate warp
Cross the 17 dimension one-hot codings obtained after C layers of processing；Wx->HIt is a weight vectors, indicates the line for connecting X layers with H layers, dimension
It one shared 1973*1024, because X layers of all dimensions are all connected with H layers of all dimensions (crying full connection), then calculates
Process is as follows:
With H in H layers1Calculation method for, if formula 2 indicate,
Wherein Wxi->H1Indicate connection i-th of element of X layer and H layers of H1Connection weight, and sigmoid is such a letter
Number, real number can be mapped to the number between 0~1 by it, and function expression isIts image shaped like
S, codomain is between 0~1.H2~H1024Calculation it is similar with formula 1.
The H layers of process converted to C layers, it is similar with calculation method above, using formula 3, with C1Calculating process for,
All C have been calculatedi, according to the characteristic of sigmoid, at C layer, will export 17 dimensions and element
It is worth the vector all between 0~1, if the vector of output is C={ 0 (namely C1=0), 0.8 (namely C2=0.8, later
And so on), 0.9,0,0,0,0,0,0,0,0,0,0,0,0,0,0 }.
According to C=, [hollow water, tree root, deposition, sundries, fouling, closure, mild corrosion, moderate corrosion, severe burn into are de-
Section, rupture, mismatch, deformation, invasion, slightly leakage, moderate leakage, severe leakage], just have in the pipe endoscopic image of representative
There are tree root diseases for 80% probability, with the presence of 90% probability deposition disease.
Wherein original I nception-ResNet-v2 network structure described in the present invention is the prior art, by existing skill
It is residual that art can mainly be divided into the initial convolutional layer of input layer, Stem, 5 × Inception × resnet-A to the description of each step
Poor layer, Reduction-A reduction layer, 10 × Inception × resnet-A residual error layer, Reduction-B reduction layer, 5 ×
Inception × resnet-C residual error layer, global average pond layer, random deactivating layer and Softmax classifier, wherein the present invention
Described in upper layer include input layer, the initial convolutional layer of Stem, 5 × Inception × resnet-A residual error layer, Reduction-
A reduces layer, 10 × Inception × resnet-A residual error layer, Reduction-B and reduces layer, 5 × Inception × resnet-
C residual error layer, global average pond layer and random deactivating layer, the specific effect of each layer and relationship are already described in the prior art, this
Invention no longer describes one by one.
As shown in Fig. 2, wherein X layers be in Disclosure of invention Inception-ResNet-v2 network structure shown in Fig. 2 with
The output vector of machine deactivating layer.
The random deactivating layer of original I nception-ResNet-v2 network structure only exports preceding 1792 dimension information, the present invention
The temporal information of a dimension is increased on this basis.
The X layers of process converted to H layers, are the forward calculation processes of full Connection Neural Network, in order to by 1792+1
The vector of dimension suitably reduces dimension, reduces into 1024 dimensions.For convenience of description, several symbols are properly added in figure, wherein X1
~X1793Represent a newly-increased dimension of the output of the random deactivating layer of original I nception-ResNet-v2+herein；H1~
X1024It represents after H layers of activation, 1793 dimensional vectors is become to each dimension values of 1024 dimensional vectors, C1~C17Indicate warp
Cross the 17 dimension one-hot codings obtained after C layers of processing；Wx->HIt is a weight vectors, indicates the line for connecting X layers with H layers, dimension
It one shared 1973*1024, because X layers of all dimensions are all connected with H layers of all dimensions (crying full connection), then calculates
Process is as follows:
With H in H layers1Calculation method for, if formula 2 indicate,
Wherein Wxi->H1Indicate connection i-th of element of X layer and H layers of H1Connection weight, and sigmoid is such a letter
Number, real number can be mapped to the number between 0~1 by it, and function expression isIts image shaped like
S, codomain is between 0~1.H2~H1024Calculation it is similar with formula 1.
The H layers of process converted to C layers, it is similar with calculation method above, using formula 3, with C1Calculating process for,
All C have been calculatedi, according to the characteristic of sigmoid, at C layer, will export 17 dimensions and element
It is worth the vector all between 0~1, if the vector of output is
C={ 0 (namely C1=0), 0.8 (namely C2=0.8, later and so on), 0.9,0,0,0,0,0,0,0,
0,0,0,0,0,0,0 }
According to C=, [hollow water, tree root, deposition, sundries, fouling, closure, mild corrosion, moderate corrosion, severe burn into are de-
Section, rupture, mismatch, deformation, invasion, slightly leakage, moderate leakage, severe leakage], just have in the pipe endoscopic image of representative
There are tree root diseases for 80% probability, with the presence of 90% probability deposition disease.
Although describing the present invention by reference to various embodiments above, but it is to be understood that of the invention not departing from
In the case where range, many changes and modifications can be carried out.That is methods discussed above, system and equipment are examples.
Various configurations can be omitted suitably, replace or add various processes or component.For example, in alternative configuration, can with institute
The order in a different order of description executes method, and/or can add, and omits and/or combine various parts.Moreover, about certain
The features of a little configuration descriptions can be combined with various other configurations, can such as combine in a similar way the different aspect of configuration with
Element.In addition, can update as technology develops element therein, i.e., many elements are examples, are not intended to limit the disclosure or power
The range that benefit requires.
Give detail in the description to provide to the thorough understanding for including the exemplary configuration realized.However,
Configuration can be practiced without these specific details for example, having been illustrated with well-known circuit, and process is calculated
Method, structure and technology are without unnecessary details, to avoid fuzzy configuration.The description only provides example arrangement, and unlimited
The scope of the claims processed, applicability or configuration.It is used on the contrary, front will provide the description of configuration for those skilled in the art
Realize the enabled description of described technology.It, can be to the function of element without departing from the spirit or the scope of the present disclosure
It can and arrange and carry out various changes.
To sum up, be intended to foregoing detailed description be considered as it is illustrative and not restrictive, and it is to be understood that more than
These embodiments are interpreted as being merely to illustrate the present invention rather than limit the scope of the invention.Reading the present invention
Record content after, technical staff can make various changes or modifications the present invention, these equivalence changes and modification are same
Fall into the scope of the claims in the present invention.
Claims (5)
1. a kind of piping disease image classification method based on multi-tag convolutional neural networks, which is characterized in that the pipeline disease
Steps are as follows for evil image classification method:
Step 1: peeping detection video in collection conduit, extract the picture frame in pipe endoscopic detection video；
Step 2: calculating the timestamp feature of each image；
Step 3: a part of picture frame that step 1 is collected into being sent into multi-tag convolutional neural networks model and is trained, is obtained
Obtain the multi-tag convolutional neural networks model that can correctly sort out piping disease type；
Step 4: with trained multi-tag convolutional neural networks model inspection pipe endoscopic image to be detected, then multi-tag is rolled up
Product neural network model can export one-hot coding, determine existing piping disease type according to one-hot coding.
2. a kind of piping disease image classification method based on multi-tag convolutional neural networks as described in claim 1, special
Sign is that the multi-tag convolutional neural networks model includes that upper layer Inception-ResNet-v2 network structure and lower layer are more
Labeling layer, the upper layer Inception-ResNet-v2 network structure further include random deactivating layer.
3. a kind of piping disease image classification method based on multi-tag convolutional neural networks as claimed in claim 2, special
Sign is, the processing step of lower layer's multi-tag classification layer are as follows:
Step 1: adding in the feature vector of the random deactivating layer output in the Inception-ResNet-v2 network structure of upper layer
Add timestamp feature TimeFeature；
Step 2: the feature vector for being added to timestamp feature in the first step is carried out dimensionality reduction activation processing, middle latitude value is obtained；
Step 3: the middle latitude value in second step is continued dimension-reduction treatment, one-hot coding is finally exported.
4. a kind of piping disease image classification method based on multi-tag convolutional neural networks as claimed in claim 3, special
Sign is, the calculation formula of the timestamp feature TimeFeature are as follows:
Wherein Current_Frame_Index is described to be detected
Pipe endoscopic image serial number, All_Frames_Num the location of in entire pipe endoscopic detection video are in entire pipeline
Peep the totalframes of detection video.
5. a kind of piping disease image classification method based on multi-tag convolutional neural networks as described in claim 1, special
Sign is that the piping disease type includes: hollow water, tree root, deposition, sundries, fouling, closure, mild corrosion, moderate corruption
Erosion, the disconnection of severe burn into, rupture, mismatch, deformation, invasion, slight leakage, moderate leakage and severe leakage.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910565360.9A CN110349134B (en) | 2019-06-27 | 2019-06-27 | Pipeline disease image classification method based on multi-label convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910565360.9A CN110349134B (en) | 2019-06-27 | 2019-06-27 | Pipeline disease image classification method based on multi-label convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110349134A true CN110349134A (en) | 2019-10-18 |
| CN110349134B CN110349134B (en) | 2022-12-09 |
Family
ID=68183348
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910565360.9A Active CN110349134B (en) | 2019-06-27 | 2019-06-27 | Pipeline disease image classification method based on multi-label convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110349134B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111191593A (en) * | 2019-12-30 | 2020-05-22 | 成都云尚物联环境科技有限公司 | Image target detection method and device, storage medium and sewage pipeline detection device |
| CN111209794A (en) * | 2019-12-11 | 2020-05-29 | 浙江省交通运输科学研究院 | Underground pipeline identification method based on ground penetrating radar image |
| CN113963212A (en) * | 2021-10-25 | 2022-01-21 | 郑州大学 | Pipeline disease image classification method and device based on increment-Resnet neural network |
| CN114549402A (en) * | 2022-01-05 | 2022-05-27 | 江苏海洋大学 | Underwater image quality comparison method without reference image |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030023404A1 (en) * | 2000-11-22 | 2003-01-30 | Osama Moselhi | Method and apparatus for the automated detection and classification of defects in sewer pipes |
| CN108830287A (en) * | 2018-04-18 | 2018-11-16 | 哈尔滨理工大学 | The Chinese image, semantic of Inception network integration multilayer GRU based on residual error connection describes method |
| CN108898595A (en) * | 2018-06-27 | 2018-11-27 | 慧影医疗科技（北京）有限公司 | A kind of construction method of thoracopathy detection model and application |
| CN109716108A (en) * | 2016-12-30 | 2019-05-03 | 同济大学 | A kind of Asphalt Pavement Damage detection system based on binocular image analysis |
| CN109800824A (en) * | 2019-02-25 | 2019-05-24 | 中国矿业大学(北京) | A kind of defect of pipeline recognition methods based on computer vision and machine learning |
| CN109871906A (en) * | 2019-03-15 | 2019-06-11 | 西安获德图像技术有限公司 | A kind of classification method of the coil open defect based on depth convolutional neural networks |
-
2019
- 2019-06-27 CN CN201910565360.9A patent/CN110349134B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030023404A1 (en) * | 2000-11-22 | 2003-01-30 | Osama Moselhi | Method and apparatus for the automated detection and classification of defects in sewer pipes |
| CN109716108A (en) * | 2016-12-30 | 2019-05-03 | 同济大学 | A kind of Asphalt Pavement Damage detection system based on binocular image analysis |
| CN108830287A (en) * | 2018-04-18 | 2018-11-16 | 哈尔滨理工大学 | The Chinese image, semantic of Inception network integration multilayer GRU based on residual error connection describes method |
| CN108898595A (en) * | 2018-06-27 | 2018-11-27 | 慧影医疗科技（北京）有限公司 | A kind of construction method of thoracopathy detection model and application |
| CN109800824A (en) * | 2019-02-25 | 2019-05-24 | 中国矿业大学(北京) | A kind of defect of pipeline recognition methods based on computer vision and machine learning |
| CN109871906A (en) * | 2019-03-15 | 2019-06-11 | 西安获德图像技术有限公司 | A kind of classification method of the coil open defect based on depth convolutional neural networks |
Non-Patent Citations (4)
| Title |
|---|
| JACK C.P. CHENG,ET AL: "Automated detection of sewer pipe defects in closed-circuit television images using deep learning techniques", 《AUTOMATION IN CONSTRUCTION 95 (2018)》 * |
| 杜珂: "基于深度学习的图像分类关键技术研究", 《中国优秀硕士论文全文数据库（信息科技辑）》 * |
| 苏越: "基于卷积神经网络的多标签图像分类_-", 《中国优秀硕士论文全文数据库（信息科技辑）》 * |
| 陈智: "基于卷积神经网络的多标签场景分类", 《中国优秀硕士论文全文数据库（信息科技辑）》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111209794A (en) * | 2019-12-11 | 2020-05-29 | 浙江省交通运输科学研究院 | Underground pipeline identification method based on ground penetrating radar image |
| CN111191593A (en) * | 2019-12-30 | 2020-05-22 | 成都云尚物联环境科技有限公司 | Image target detection method and device, storage medium and sewage pipeline detection device |
| CN113963212A (en) * | 2021-10-25 | 2022-01-21 | 郑州大学 | Pipeline disease image classification method and device based on increment-Resnet neural network |
| CN114549402A (en) * | 2022-01-05 | 2022-05-27 | 江苏海洋大学 | Underwater image quality comparison method without reference image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110349134B (en) | 2022-12-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110349134A (en) | A kind of piping disease image classification method based on multi-tag convolutional neural networks | |
| CN112435221B (en) | Image anomaly detection method based on generated countermeasure network model | |
| CN107133948B (en) | Image blurring and noise evaluation method based on multitask convolution neural network | |
| CN109977921B (en) | Method for detecting hidden danger of power transmission line | |
| CN108564052A (en) | Multi-cam dynamic human face recognition system based on MTCNN and method | |
| CN113486865B (en) | Power transmission line suspended foreign object target detection method based on deep learning | |
| CN110992349A (en) | Underground pipeline abnormity automatic positioning and identification method based on deep learning | |
| CN111709265A (en) | Camera monitoring state classification method based on attention mechanism residual error network | |
| CN101563710A (en) | Method and apparatus for identifying properties of an object detected by a video surveillance camera | |
| Carrara et al. | Combining gans and autoencoders for efficient anomaly detection | |
| CN110084165A (en) | The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations | |
| Christlein et al. | A study on features for the detection of copy-move forgeries | |
| CN109919921B (en) | Environmental impact degree modeling method based on generation countermeasure network | |
| CN110348342B (en) | Pipeline disease image segmentation method based on full convolution network | |
| CN111563896B (en) | Image processing method for detecting abnormality of overhead line system | |
| CN112163572A (en) | Method and device for identifying object | |
| Zhou et al. | Deep multi-scale features learning for distorted image quality assessment | |
| CN113313107A (en) | Intelligent detection and identification method for multiple types of diseases on cable surface of cable-stayed bridge | |
| CN111783751A (en) | Rifle ball linkage and BIM-based breeding house piglet abnormity early warning method | |
| CN111767826A (en) | Timing fixed-point scene abnormity detection method | |
| CN114170686A (en) | Elbow bending behavior detection method based on human body key points | |
| CN113378672A (en) | Multi-target detection method for defects of power transmission line based on improved YOLOv3 | |
| CN106375756B (en) | It is a kind of to remove the detection method distorted for the single object of monitor video | |
| CN115331152B (en) | Fire fighting identification method and system | |
| CN113221603A (en) | Method and device for detecting shielding of monitoring equipment by foreign matters |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |