CN115114860A - Data modeling and amplifying method for concrete pipeline damage identification - Google Patents
Data modeling and amplifying method for concrete pipeline damage identification Download PDFInfo
- Publication number
- CN115114860A CN115114860A CN202210872779.0A CN202210872779A CN115114860A CN 115114860 A CN115114860 A CN 115114860A CN 202210872779 A CN202210872779 A CN 202210872779A CN 115114860 A CN115114860 A CN 115114860A
- Authority
- CN
- China
- Prior art keywords
- concrete pipeline
- data
- pipeline damage
- shapcontextnet
- model
- 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
- 238000000034 method Methods 0.000 title claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 31
- 238000012360 testing method Methods 0.000 claims abstract description 29
- 230000003321 amplification Effects 0.000 claims abstract description 24
- 238000003199 nucleic acid amplification method Methods 0.000 claims abstract description 24
- 238000001514 detection method Methods 0.000 claims abstract description 22
- 238000012545 processing Methods 0.000 claims abstract description 4
- 238000003062 neural network model Methods 0.000 claims description 31
- 238000011156 evaluation Methods 0.000 claims description 17
- 230000009466 transformation Effects 0.000 claims description 16
- 238000012795 verification Methods 0.000 claims description 15
- 201000010099 disease Diseases 0.000 claims description 11
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 claims description 11
- 238000011549 displacement method Methods 0.000 claims description 10
- 238000011176 pooling Methods 0.000 claims description 8
- 238000011423 initialization method Methods 0.000 claims description 6
- 230000008859 change Effects 0.000 claims description 5
- 230000006870 function Effects 0.000 claims description 4
- 238000010200 validation analysis Methods 0.000 claims description 4
- 238000002372 labelling Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 230000005653 Brownian motion process Effects 0.000 claims description 2
- 238000005537 brownian motion Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 abstract description 4
- 238000011897 real-time detection Methods 0.000 abstract 1
- 230000008569 process Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 5
- 230000002776 aggregation Effects 0.000 description 3
- 238000004220 aggregation Methods 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 239000011435 rock Substances 0.000 description 2
- 208000004350 Strabismus Diseases 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000003190 augmentative effect Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000007613 environmental effect Effects 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 230000003902 lesion Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 239000010865 sewage Substances 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000011179 visual inspection Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/13—Architectural design, e.g. computer-aided architectural design [CAAD] related to design of buildings, bridges, landscapes, production plants or roads
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2113/00—Details relating to the application field
- G06F2113/14—Pipes
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Geometry (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Civil Engineering (AREA)
- Computing Systems (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Architecture (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Structural Engineering (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a data modeling and amplification method for concrete pipeline damage identification. The method comprises the following steps: the method comprises the steps of concrete pipeline damage model building, data processing, ShapContextNet model building, model training, model parameter adjusting, model testing and real-time detection. The data modeling amplification method for concrete pipeline damage identification provided by the invention is used for researching and developing a ShapContextNet neural network algorithm suitable for concrete pipeline damage detection, and performing data amplification by using the data modeling method under the condition that the real concrete pipeline damage data quantity is rare, so that the robustness and generalization capability of the method are improved, and the automatic identification of the concrete pipeline damage is realized.
Description
Technical Field
The invention relates to the technical field of data modeling, in particular to a data modeling amplification method for concrete pipeline damage identification.
Background
With the development of urbanization, urban drainage systems are responsible for urban civilization construction and human healthy life. With the development of social economy and the continuous expansion of cities, the total length of drainage pipelines in China is rapidly increased, and accordingly, the maintenance work of drainage pipelines becomes important. Every year, municipal departments need to spend a large amount of funds and resources to treat road surface collapse caused by drainage pipelines, urban waterlogging and environmental problems caused by sewage, and the detection of drainage pipelines aims to find out the defects of the pipelines as soon as possible, which is of no great importance to the treatment of urban drainage pipelines.
At present, a detection method of a drainage pipeline is disease identification and detection which is manually participated, and a Closed Circuit Television (CCTV) is one of visual inspection systems, and is an existing detection method of a main drainage pipeline, and the detection method is transmitted to an external display through a closed circuit video by a pipeline robot carrying high-definition camera. Compared with the traditional detection method, the television detection method does not need to manually enter a pipeline for probing, and can provide a simpler and more obvious image result. However, this type of detection requires a large number of trained professionals to identify and evaluate the lesion over a long period of time, which can be time and resource consuming. In recent years, the use of three-dimensional information for structural health analysis has become a new trend. The three-dimensional laser scanning has strong anti-interference capability, can be used in dark light and fog environments, is more effective for detecting complex environments in drainage pipelines, and provides thinking for other detections due to the occurrence of three-dimensional information, such as calculating the depth, area and volume information of damage according to the three-dimensional information of the damaged part of the structure.
However, the existing drainage pipeline damage data is insufficient, and the intelligent detection progress of the drainage pipeline is seriously influenced. Therefore, it is necessary to provide a data modeling and amplification method for concrete pipe damage identification to solve the above problems.
Disclosure of Invention
The invention aims to provide a data modeling amplification method for concrete pipeline damage identification, which aims to solve the problem that the existing damage data of a drainage pipeline is insufficient and the intelligent detection progress of the drainage pipeline is seriously influenced.
The invention provides a data modeling and amplification method for concrete pipeline damage identification, which comprises the following steps:
establishing a damaged rough inner surface of the concrete pipeline based on a random midpoint displacement method in a fractional Brown motion theory, and combining the damaged rough inner surface of the concrete pipeline with a concrete pipeline model to obtain a concrete pipeline damage model;
combining real concrete pipeline damage data acquired by using a depth camera with the concrete pipeline damage model, classifying and labeling concrete pipeline damage data sets, amplifying the concrete pipeline damage data sets by using a data enhancement method, dividing the concrete pipeline damage data sets into a training set, a verification set and a test set according to a proportion, and constructing to obtain an amplified concrete pipeline damage data set;
building a ShapContextNet neural network model which mainly comprises a ShapContext layer, an MLP layer and a maximum pooling layer;
initializing the ShapContextNet neural network model by adopting a Zerewitinol initialization method, setting a hyper-parameter, importing a training set into the ShapContextNet neural network model, and training the model;
setting different hyper-parameters, introducing a verification set to verify the ShapContextNet neural network model, comparing the loss values of the ShapContextNet neural network model under the different hyper-parameters and the change curves of the accuracy, and searching the optimal hyper-parameter;
testing the ShapContextNet neural network model under the optimal hyper-parameter according to the test set, outputting each numerical value evaluation index, and judging whether the numerical value reaches an expected value;
and inputting an image obtained by shooting the damage disease of the concrete pipeline by using the pipeline robot into a ShapContextNet neural network model reaching an expected value, and judging whether the output evaluation index can meet the required requirement.
Further, based on a random midpoint displacement method in a fractional brownian motion theory, establishing a damaged rough inner surface of the concrete pipeline, and combining the damaged rough inner surface of the concrete pipeline with a concrete pipeline model to obtain the concrete pipeline damage model, wherein the method comprises the following steps:
according to a random midpoint displacement method, establishing a damaged rough inner surface of the concrete pipeline by using matlab;
building a concrete pipeline model by using comsol;
and embedding the generated point cloud of the damaged rough inner surface of the concrete pipeline into the concrete pipeline model.
Further, combining real concrete pipeline damage data acquired by using a depth camera with the concrete pipeline damage model, classifying and marking concrete pipeline damage data sets, amplifying the concrete pipeline damage data sets by using a data enhancement method, dividing the concrete pipeline damage data sets into a training set, a verification set and a test set according to a proportion, and constructing to obtain an amplified concrete pipeline damage data set, wherein the method comprises the following steps:
uniformly sampling the concrete pipeline damage model by using CloudCompare to generate three-dimensional point cloud data;
establishing a data set by using the concrete pipeline damage model point cloud and the concrete pipeline real damage point cloud in a ratio of 3: 1;
classifying and marking concrete pipeline damage into normal and damaged concrete pipeline damage by using a CloudCompare marking program;
amplifying the labeled data set by using a data enhancement method;
the data set was sorted by 6 using a written python sorter: 3: the scale of 1 is divided into a training set, a validation set and a test set.
Further, in the step of amplifying the labeled data set by using a data enhancement method, the data enhancement method comprises translation transformation, scaling, rotation transformation, symmetry transformation, stretching transformation and dithering transformation.
Further, the data set was sorted by 6: 3: the proportion of 1 is divided into a training set, a verification set and a test set, wherein images of the training set, the verification set and the test set do not overlap with each other.
Further, a ShapContextNet neural network model is built, the ShapContextNet neural network model mainly comprises a ShapContext layer, an MLP layer and a maximum pooling layer, the ShapNet layer is used for integrating local and overall information of the three-dimensional point cloud and mapping the features to a high dimension, the MLP layer is used for extracting and reducing the dimensions of the high-dimensional features, and the maximum pooling layer performs dimension reduction on the three-dimensional point cloud features on the basis that a symmetric function is used for ensuring that the features of the extracted three-dimensional point cloud are not changed.
And further, initializing the ShapContextNet neural network model by using a Zerewinder initialization method, setting a hyper-parameter, importing a training set into the ShapContextNet neural network model, and in the step of training the model, solving the problems that the parameter variance is reduced and the gradient disappears as the layer deepens by using a Zerewinder initialization method by using a statistical principle.
And further setting different hyper-parameters, introducing a verification set to verify the ShapContextNet neural network model, comparing the loss values of the ShapContextNet neural network model under the different hyper-parameters with the change curve of the accuracy, and searching the optimal hyper-parameter, wherein the hyper-parameters comprise the learning rate, the total iteration times, the number of three-dimensional point clouds and the attenuation rate.
And further, testing the ShapContextNet neural network model under the optimal hyper-parameter according to the test set, outputting various numerical evaluation indexes, and judging whether the numerical evaluation indexes reach expected values, wherein the numerical evaluation indexes comprise precision, accuracy and recall rate.
And further, inputting an image obtained by shooting the damage and the disease of the concrete pipeline by using the pipeline robot into a ShapContextNet neural network model reaching an expected value, and judging whether output evaluation indexes can meet the required requirements, wherein the evaluation indexes comprise accuracy, detection efficiency and IOU value.
The invention has the following beneficial effects: according to the data modeling and amplification method facing to concrete pipeline damage identification, a three-dimensional point cloud mode is adopted, damage shapes can be visually represented, more visual size information such as depth, area and volume information is provided for engineers, and information is provided for repairing a next drainage pipeline; the method of modeling and amplifying by combining the ShapContextNet neural network with data is adopted to automate the processes of detecting and segmenting the concrete pipeline damage, so that the efficiency of detecting the concrete pipeline damage is improved, and the robustness of a network model is improved.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious to those skilled in the art that other drawings can be obtained according to the drawings without any inventive exercise.
FIG. 1 is a flow chart of a data modeling amplification method for concrete pipeline damage identification according to the invention.
FIG. 2 is a schematic diagram of the random midpoint displacement method of the present invention.
FIG. 3 is a general structure diagram of a data modeling amplification method for concrete pipeline damage identification according to the present invention.
FIG. 4 is a schematic diagram of a processing result of the data modeling amplification method for concrete pipeline damage identification.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be clearly and completely described below with reference to the specific embodiments of the present invention and the accompanying drawings. It is to be understood that the described embodiments are merely exemplary of the invention, and not restrictive of the full scope of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention. The technical solutions provided by the embodiments of the present invention are described in detail below with reference to the accompanying drawings.
Referring to fig. 1, a data modeling and amplification method for identifying damage to a concrete pipe according to the present invention includes the following steps:
the method comprises the following steps: and (5) building a concrete pipeline damage model.
The method comprises the following steps of establishing a damaged rough inner surface of the concrete pipeline based on random midpoint displacement in a fractional Brown motion theory, and combining the damaged rough inner surface of the concrete pipeline to a concrete pipeline model, wherein the concrete process comprises the following steps:
1.1 establishing the damaged rough inner surface of the concrete pipeline by utilizing matlab according to a random mid-point displacement method.
1.2 use comsol to build concrete pipe models.
1.3, embedding the generated point cloud of the damaged rough inner surface of the concrete pipeline into a concrete pipeline model.
Specifically, as shown in fig. 2, the random midpoint displacement method is first established by 4 corner points a 0 ,B 0 ,C 0 And D 0 Determining a construction area of the fracture surface, the height value of which obeys a mean value of 0 and a variance of delta 2 Is a gaussian distribution random function N (0, δ) 2 ) Performing linear interpolation according to the average value of 4 angular points to obtain a central point A 1 Then, the middle point, B, of the adjacent angular point is interpolated according to the adjacent angular point 1 、C 1 、D 1 、E 1 While adding compliance
A random value. Wherein the variance is
In the formula of 2 H is the Hurst index and is more than or equal to 0 and less than or equal to 1. Repeating the above process, and obtaining a second interpolation result 2 、B 2 、C 2 、D 2 、E 2 So that n interpolations are performed, one for each result
A random value. Wherein the variance is
n is the interpolation degree.
Specifically, varying degrees of damage to the rough inner surface are generated by modifying the size control of H.
Specifically, a concrete drainage pipeline model with the diameter of 120cm is set for matching with real pipeline damage data, and a triangular, square and circular drainage pipeline damage shape is used for modeling, so that the robustness of the model is improved.
Step two: and (4) preprocessing data.
Acquiring real concrete pipeline damage data by using a depth camera, combining the real concrete pipeline damage data with the concrete pipeline damage model obtained in the step one, classifying and marking the concrete pipeline damage data set, amplifying the concrete pipeline damage data set by using a data enhancement method, dividing the concrete pipeline damage data set into a training set, a verification set and a test set according to a proportion, and constructing the concrete pipeline damage data set, wherein the concrete process comprises the following steps of:
2.1 using CloudCompare software to uniformly sample the concrete pipeline damage model to generate three-dimensional point cloud data.
2.2, establishing a data set by the concrete pipeline damage model point cloud and the concrete pipeline real damage point cloud according to the proportion of 3: 1.
2.3 Classification labelling of concrete pipe damage, normal and damaged, using the CloudCompare labelling program.
2.4 the annotated data set is augmented using data enhancement techniques.
2.5 Using the write python classifier, the dataset was sorted by 6: 3: the ratio of 1 is divided into a training set, a validation set and a test set.
Specifically, the acquired real damage image should have:
the images should be diverse and the concrete pipe data set should include images of light intensity, shadows, etc.
The image shooting distance has the universality, the shooting distances are different, and the generalization capability is improved.
The viewing angle of the image is diversified, and the image comprises a front-view image, a side-view image, a squint image and the like.
The disease target should be representative and have obvious disease characteristics.
Specifically, the main methods for data enhancement are translation transformation, scaling, rotation transformation, symmetry transformation, stretch transformation, and dithering transformation.
Specifically, images of the training set, the verification set and the test set do not overlap with each other, and robustness and generalization capability of the test model are favorably checked.
Step three: and (5) building a ShapContextNet model.
The model mainly comprises a ShapContext layer, an MLP layer and a maximum pooling layer.
Specifically, the ShapNet layer is used for integrating local and overall information of the three-dimensional point cloud and mapping the features to a high dimension, the MLP layer is used for extracting and reducing the dimensions of the high-dimensional features, and the maximum pooling layer performs dimension reduction on the three-dimensional point cloud features on the basis that symmetric functions are used for guaranteeing that the extracted three-dimensional point cloud features are unchanged.
Specifically, as shown in fig. 3, the ShapContext layer mainly includes three modules of selection, aggregation, and transformation, the selection module generates an affinity matrix according to the neighborhood information of the three-dimensional point cloud to represent the relationship between a certain point and other points in the three-dimensional point cloud, establishes a ShapContext kernel by means of spherical coordinates, and generates L affinity matrices from L bins introduced by the ShapContext kernel, wherein each bin corresponds to a specific bin. The result of the aggregation operation is a converged set of features, and thus the output tensor is N × L × D in shape. (N points, L is the number of bins, D is the vector dimension after each bin aggregation). The conversion operation is implemented by a convolutional layer having an [ L, 1] kernel that aggregates L sets of feature points and projects them to the output (high-dimensional) feature vector.
Step four: and (5) training a model.
Initializing the model by adopting a Zerewitinol initialization method, setting a hyper-parameter, importing a training set into a ShapContextNet neural network model, and training the model.
Specifically, the Zerewinder initialization applies a large number of statistical principles to solve the problems that the parameter variance is reduced and the gradient disappears along with the layer deepening, so that the training efficiency can be accelerated, and the detection precision can be improved.
Specifically, based on the high bandwidth and multithreading parallel computing characteristics of the GPU, the deep learning algorithm has to be performed on a high performance computer.
Step five: and (5) adjusting parameters of the model.
Setting different hyper-parameters, introducing a verification set data test model, comparing the loss values of the models under different hyper-parameters and the change curves of the accuracy, and searching the optimal hyper-parameter.
Specifically, the main hyper-parameters are learning rate, total iteration times, three-dimensional point cloud number, attenuation coefficient and the like.
Specifically, the learning rate controls the speed of updating the weight parameters. In the process of model training, the model misses the optimal solution due to the fact that an overlarge learning rate is set, and the model is over-fitted. Setting an excessively small learning rate slows down the learning speed. The learning rate needs to be set based on experience and constant experimentation.
Specifically, the total iteration number refers to the number of times that the whole training set is input to the neural network for training, and when the difference between the test error rate and the training error rate is small, the current iteration number can be considered to be appropriate. When the test error ratio is smaller and then larger, the iteration times are too large, the iteration times need to be reduced, otherwise, overfitting is easy to occur.
Specifically, the number of three-dimensional point clouds refers to the number of point clouds input into the training model, and is determined by counting the number of three-dimensional point clouds in the data set.
Specifically, the attenuation coefficient is set up to solve the problem that the learning rate is too large or too small, and the learning rate is continuously adjusted in the iteration process.
Specifically, the main basis for adjusting the hyper-parameters in this embodiment is that, when different hyper-parameters are set, the detection performance of the model on the verification set, that is, the efficiency of identification, the smoothness and the convergence of the maximum value of the accuracy rate rising curve and the loss value falling curve.
Step six: and (5) testing the model.
And testing the optimal model according to the test set data, outputting each numerical value evaluation index, and judging whether the expected value is reached.
Specifically, each numerical evaluation index in the sixth step includes precision, accuracy and recall rate.
Step seven: and (5) detecting in the field.
And shooting damage and disease of the concrete pipeline by using a robot of the pipeline, transmitting the damage and disease back to the high-performance computer terminal for detection, and judging whether the evaluation index can meet the required requirement.
Specifically, the Phoxi3DScanner dynamic 3D camera emits invisible infrared laser with specific wavelength as a light source, the emitted light is projected on an object through a certain code, the distortion of a returned code pattern is calculated through a certain algorithm to obtain the position and depth information of the object, the throughput of three-dimensional points of the instrument is 1600 ten thousand points per second, the resolution reaches 320 ten thousand pixels, and the detection requirement of an underground drainage pipeline is met.
Specifically, the evaluation index mainly includes accuracy, detection efficiency, and an IOU value.
Specifically, correct and wrong disease identification statistics should be performed, whether the accuracy rate of the statistics reaches more than 85% is calculated, if the accuracy rate does not meet the set requirements, the false detection disease data needs to be re-counted and retrained until all indexes reach the expected requirements.
The processing result of the method of the invention is shown in fig. 4, wherein fig. 4(a) is a rough inner surface generated by a random mid-point displacement method, fig. 4(b) is a concrete pipeline damage modeling diagram, and fig. 4(c) is a recognition result diagram of a ShapContext algorithm.
By using some fluid-solid coupling researches for reference, the fractal theory is used for representing the rock roughness, the rough surface of the rock is generated by modeling, and then three-dimensional model data is obtained, so that data support can be provided for a deep learning model. The invention combines data modeling and three-dimensional point cloud deep learning, provides a novel automatic detection method for concrete drainage pipeline damage, and can detect the damage under the condition that the number of the damaged three-dimensional point clouds of the concrete drainage pipeline is small.
The embodiments can show that the invention provides a data modeling amplification method facing to concrete pipeline damage identification, develops a ShapContextNet neural network algorithm suitable for concrete pipeline damage detection, performs data amplification by using the data modeling method under the condition that the real concrete pipeline damage data volume is rare, and improves the robustness and generalization capability of the method, thereby realizing automatic identification of concrete pipeline damage.
The same and similar parts in the various embodiments in this specification may be referred to each other. The above-described embodiments of the present invention should not be construed as limiting the scope of the present invention.
Claims (10)
1. A data modeling and amplification method for concrete pipeline damage identification is characterized by comprising the following steps:
establishing a damaged rough inner surface of the concrete pipeline based on a random midpoint displacement method in a fractional Brown motion theory, and combining the damaged rough inner surface of the concrete pipeline with a concrete pipeline model to obtain a concrete pipeline damage model;
combining real concrete pipeline damage data acquired by using a depth camera with the concrete pipeline damage model, classifying and labeling concrete pipeline damage data sets, amplifying the concrete pipeline damage data sets by using a data enhancement method, dividing the concrete pipeline damage data sets into a training set, a verification set and a test set according to a proportion, and constructing to obtain an amplified concrete pipeline damage data set;
building a ShapContextNet neural network model which mainly comprises a ShapContext layer, an MLP layer and a maximum pooling layer;
initializing the ShapContextNet neural network model by adopting a Zerewitinol initialization method, setting a hyper-parameter, importing a training set into the ShapContextNet neural network model, and training the model;
setting different hyper-parameters, introducing a verification set to verify the ShapContextNet neural network model, comparing the loss values of the ShapContextNet neural network model under the different hyper-parameters and the change curves of the accuracy, and searching the optimal hyper-parameter;
testing the ShapContextNet neural network model under the optimal hyper-parameter according to the test set, outputting each numerical value evaluation index, and judging whether the numerical value reaches an expected value;
and inputting an image obtained by shooting the damage disease of the concrete pipeline by using the pipeline robot into a ShapContextNet neural network model reaching an expected value, and judging whether the output evaluation index can meet the required requirement.
2. The data modeling and amplification method oriented to concrete pipeline damage identification according to claim 1, wherein the concrete pipeline damage rough inner surface is established based on a random midpoint displacement method in a fractional Brownian motion theory, and is combined with a concrete pipeline model to obtain the concrete pipeline damage model, and the method comprises the following steps:
according to a random midpoint displacement method, establishing a damaged rough inner surface of the concrete pipeline by using matlab;
building a concrete pipeline model by using comsol;
and embedding the generated point cloud of the damaged rough inner surface of the concrete pipeline into the concrete pipeline model.
3. The data modeling and amplification method for concrete pipeline damage recognition according to claim 1, wherein real concrete pipeline damage data collected by a depth camera is combined with the concrete pipeline damage model, a concrete pipeline damage data set is classified and labeled, the concrete pipeline damage data set is amplified by a data enhancement method, and is divided into a training set, a verification set and a test set according to a proportion, and an amplified concrete pipeline damage data set is constructed, and the method comprises the following steps:
uniformly sampling the concrete pipeline damage model by using CloudCompare to generate three-dimensional point cloud data;
establishing a data set by using the concrete pipeline damage model point cloud and the concrete pipeline real damage point cloud in a ratio of 3: 1;
classifying and marking concrete pipeline damage into normal and damaged concrete pipeline damage by using a CloudCompare marking program;
amplifying the labeled data set by using a data enhancement method;
the data set was sorted by 6 using a written python sorter: 3: the scale of 1 is divided into a training set, a validation set and a test set.
4. The concrete pipe damage identification-oriented data modeling and amplification method according to claim 3, wherein in the step of amplifying the labeled data set by using a data enhancement method, the data enhancement method comprises translation transformation, scaling, rotation transformation, symmetry transformation, stretching transformation and dithering transformation.
5. The data modeling and amplification method for concrete pipeline damage identification as claimed in claim 3, wherein a python classification program is written to classify the data set into 6: 3: the proportion of 1 is divided into a training set, a verification set and a test set, wherein images of the training set, the verification set and the test set do not overlap with each other.
6. The data modeling and amplification method for concrete pipeline damage identification as claimed in claim 1, wherein a ShapContextNet neural network model is built, the ShapContextNet neural network model mainly includes a ShapContext layer, an MLP layer and a maximum pooling layer, the ShapNet layer is used for integrating local and overall information of the three-dimensional point cloud and mapping the features to a high dimension, the MLP layer is used for extracting the high-dimensional features to reduce dimensions, and the maximum pooling layer uses a symmetric function to perform dimension reduction processing on the three-dimensional point cloud features on the basis that the extracted features of the three-dimensional point cloud are not changed.
7. The data modeling and amplification method oriented to concrete pipeline damage recognition is characterized in that a Zeval initialization method is adopted to initialize the ShapContextNet neural network model, the hyper-parameters are set, a training set is led into the ShapContextNet neural network model, and in the step of training the model, the Zeval initialization applies a statistical principle to solve the problems that parameter variance is reduced and gradient disappears as a layer deepens.
8. The data modeling and amplification method oriented to concrete pipeline damage recognition as claimed in claim 1, wherein different hyper-parameters are set, a validation set is introduced to validate the ShapContextNet neural network model, and optimal hyper-parameters are found by comparing the loss values and the change curves of the accuracy of the ShapContextNet neural network model under the different hyper-parameters, wherein the hyper-parameters include learning rate, total iteration number, three-dimensional point cloud number and attenuation rate.
9. The data modeling and amplification method for concrete pipeline damage identification as claimed in claim 1, wherein in the step of testing the ShapContextNet neural network model under the optimal hyper-parameter according to the test set, outputting various numerical evaluation indexes including accuracy, accuracy and recall rate, and judging whether the numerical evaluation indexes reach the expected values.
10. The data modeling and amplification method for concrete pipeline damage recognition according to claim 1, wherein the method includes the steps of inputting an image obtained by shooting a concrete pipeline damage disease by using a pipeline robot into a ShapContextNet neural network model with an expected value, and judging whether output evaluation indexes can meet required requirements, wherein the evaluation indexes include accuracy, detection efficiency and IOU value.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210872779.0A CN115114860B (en) | 2022-07-21 | 2022-07-21 | Data modeling amplification method for concrete pipeline damage identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210872779.0A CN115114860B (en) | 2022-07-21 | 2022-07-21 | Data modeling amplification method for concrete pipeline damage identification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115114860A true CN115114860A (en) | 2022-09-27 |
| CN115114860B CN115114860B (en) | 2024-03-01 |
Family
ID=83333415
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210872779.0A Active CN115114860B (en) | 2022-07-21 | 2022-07-21 | Data modeling amplification method for concrete pipeline damage identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115114860B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200175352A1 (en) * | 2017-03-14 | 2020-06-04 | University Of Manitoba | Structure defect detection using machine learning algorithms |
| WO2020119103A1 (en) * | 2018-12-13 | 2020-06-18 | 程琳 | Aero-engine hole detection image damage intelligent identification method based on deep learning |
| CN114758222A (en) * | 2022-03-09 | 2022-07-15 | 哈尔滨工业大学水资源国家工程研究中心有限公司 | Concrete pipeline damage identification and volume quantification method based on PointNet ++ neural network |
-
2022
- 2022-07-21 CN CN202210872779.0A patent/CN115114860B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200175352A1 (en) * | 2017-03-14 | 2020-06-04 | University Of Manitoba | Structure defect detection using machine learning algorithms |
| WO2020119103A1 (en) * | 2018-12-13 | 2020-06-18 | 程琳 | Aero-engine hole detection image damage intelligent identification method based on deep learning |
| CN114758222A (en) * | 2022-03-09 | 2022-07-15 | 哈尔滨工业大学水资源国家工程研究中心有限公司 | Concrete pipeline damage identification and volume quantification method based on PointNet ++ neural network |
Non-Patent Citations (1)
| Title |
|---|
| 谢坤明;: "基于随机森林算法的结构损伤识别", 福建交通科技, no. 05, 20 October 2020 (2020-10-20) * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115114860B (en) | 2024-03-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111091105B (en) | Remote sensing image target detection method based on new frame regression loss function | |
| Li et al. | Automatic pavement crack detection by multi-scale image fusion | |
| Wang et al. | Autonomous garbage detection for intelligent urban management | |
| Yuan et al. | Vision-based defect detection for mobile phone cover glass using deep neural networks | |
| CN111275688A (en) | Small target detection method based on context feature fusion screening of attention mechanism | |
| CN114638784B (en) | Method and device for detecting surface defects of copper pipe based on FE-YOLO | |
| CN112287788A (en) | Pedestrian detection method based on improved YOLOv3 and improved NMS | |
| Sun et al. | Autonomous crack and bughole detection for concrete surface image based on deep learning | |
| CN111191628B (en) | Remote sensing image earthquake damage building identification method based on decision tree and feature optimization | |
| CN114758222B (en) | Concrete pipeline damage identification and volume quantification method based on PointNet ++ neural network | |
| CN113420619A (en) | Remote sensing image building extraction method | |
| CN117876381B (en) | AI visual detection method and system for identifying and analyzing concrete structure cracks | |
| CN117011274A (en) | Automatic glass bottle detection system and method thereof | |
| Lee et al. | Deep learning based 3D defect detection system using photometric stereo illumination | |
| CN117036243A (en) | Method, device, equipment and storage medium for detecting surface defects of shaving board | |
| CN117949942B (en) | Target tracking method and system based on fusion of radar data and video data | |
| CN115424237A (en) | Forward vehicle identification and distance detection method based on deep learning | |
| CN115937518A (en) | Pavement disease identification method and system based on multi-source image fusion | |
| CN113496260B (en) | Grain depot personnel non-standard operation detection method based on improved YOLOv3 algorithm | |
| Pang et al. | Multi-Scale Feature Fusion Model for Bridge Appearance Defect Detection | |
| CN115114860A (en) | Data modeling and amplifying method for concrete pipeline damage identification | |
| CN116642897A (en) | Aviation spare part coating appearance defect detection system | |
| CN115311200A (en) | Concrete pipeline damage identification and volume quantification method based on SSA-PointNet ++ neural network | |
| Xiao et al. | Computer-aided industrial product design based on image enhancement algorithm and convolutional neural network | |
| Gong et al. | Pavement Image Data Set for Deep Learning: A Synthetic Approach |
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 |