CN114694051A - Electromagnetic method pipeline disease identification and positioning method based on improved CapsNet network - Google Patents
Electromagnetic method pipeline disease identification and positioning method based on improved CapsNet network Download PDFInfo
- Publication number
- CN114694051A CN114694051A CN202210261292.9A CN202210261292A CN114694051A CN 114694051 A CN114694051 A CN 114694051A CN 202210261292 A CN202210261292 A CN 202210261292A CN 114694051 A CN114694051 A CN 114694051A
- Authority
- CN
- China
- Prior art keywords
- model
- improved
- electromagnetic
- ground penetrating
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 59
- 201000010099 disease Diseases 0.000 title claims abstract description 34
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 title claims abstract description 34
- 239000002775 capsule Substances 0.000 claims abstract description 35
- 230000000149 penetrating effect Effects 0.000 claims abstract description 34
- 238000001514 detection method Methods 0.000 claims abstract description 33
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000012360 testing method Methods 0.000 claims abstract description 24
- 238000001228 spectrum Methods 0.000 claims abstract description 17
- 238000011156 evaluation Methods 0.000 claims abstract description 9
- 238000000605 extraction Methods 0.000 claims abstract description 9
- 238000012545 processing Methods 0.000 claims abstract description 9
- 239000011800 void material Substances 0.000 claims description 10
- 239000000284 extract Substances 0.000 claims description 8
- 230000008859 change Effects 0.000 claims description 7
- 238000004088 simulation Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 claims description 5
- 238000012216 screening Methods 0.000 claims description 5
- 210000005036 nerve Anatomy 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 2
- 230000000694 effects Effects 0.000 abstract description 5
- 238000013135 deep learning Methods 0.000 abstract description 3
- 230000008569 process Effects 0.000 description 7
- 239000002131 composite material Substances 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 229910001018 Cast iron Inorganic materials 0.000 description 1
- XEEYBQQBJWHFJM-UHFFFAOYSA-N Iron Chemical compound [Fe] XEEYBQQBJWHFJM-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction 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
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 230000008676 import Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000009440 infrastructure construction Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000002184 metal Substances 0.000 description 1
- 229910052751 metal Inorganic materials 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 229910052755 nonmetal Inorganic materials 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 238000011426 transformation method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
- G01V3/00—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation
- G01V3/12—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation operating with electromagnetic waves
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Remote Sensing (AREA)
- Electromagnetism (AREA)
- Environmental & Geological Engineering (AREA)
- Geology (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geophysics (AREA)
- Geophysics And Detection Of Objects (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses an electromagnetic method pipeline disease identification and positioning method based on an improved CapsNet network, which comprises the following steps: s1 establishing an electromagnetic spectrum data set; s2 data set processing; s3, constructing and training a ground penetrating radar map detection model based on the improved CapsNet network; s4 model parameter adjustment; s5, performing model test, outputting each evaluation value index, storing an optimal model, and outputting the recognition result of the ground penetrating radar map; and S6, according to the recognition result, positioning the pipeline diseases by the ground penetrating radar A-scan data at the corresponding position. The method adopts deep learning, carries out model training based on electromagnetic map big data, adds a convolution layer for extracting the characteristics at different positions in the picture again and optimizes the characteristic extraction effect; the original dynamic routing algorithm of the capsule network is improved, so that the parameter quantity is greatly reduced, the scale of the input picture which is allowed to be accepted is increased, the accuracy of model detection is improved, and the intelligent and efficient identification of the pipeline diseases is realized.
Description
Technical Field
The invention relates to the technical field of deep learning and geophysical exploration, in particular to an electromagnetic method pipeline disease identification and positioning method based on an improved CapsNet network.
Background
The ground penetrating radar is an important electromagnetic nondestructive detection technology, can accurately and quickly image the superficial earth surface, obtains the space and geometric information of buried objects, and plays an important role in geophysical prospecting applications such as pipeline general survey, highway hidden disease detection and the like. In the detection process, the radar echo signal characteristic of the underground pipeline can be deduced to be close to a hyperbola according to the electromagnetic wave theory and the GPR principle; the characteristics of the void are that the energy of the reflected signal is strong, the frequency, amplitude and phase change is abnormal and obvious, the lower multi-time reflected wave is obvious, and the boundary can be accompanied with diffraction phenomenon; the loose characteristic is that the energy change of the reflected signal is large, the same phase axis is discontinuous, and the waveform is complex and discontinuous. The research on the analysis method of the radar map features is mainly divided into two types: manual method, automatic detection method. In the manual method, a large amount of time is spent by professional personnel for manually judging the picture, the subjective experience of the personnel is excessively relied on, the speed is low, and the automation degree is low; the automatic detection method mainly comprises a Hough (Hough) transformation method, a Least Square (LS) method, a machine learning method and the like, most input features need to be identified manually, and the quality of the features directly determines the detection and classification results. After the data volume increases, efficiency and accuracy are difficult to guarantee.
In the process of city development, underground pipelines are an important component in infrastructure construction, and normal operation of city functions and improvement of life quality of people cannot be separated from the underground pipelines. Due to the fact that underground space information is incomplete and not timely updated in time of planning design and construction, inevitable loss in the operation and maintenance process occurs, materials of a pipe network are aged, the pipe network is changed, excavation errors occur, and the problems of void, looseness, hidden cracks and the like occur in highway engineering. These problems not only adversely affect normal urban functions, but also may lead to the occurrence of serious accidents such as fire and explosion. Therefore, when the underground space is comprehensively surveyed, recorded and systematically managed, how to improve the precision and speed of detection and adopt a reasonable and practical detection and identification method becomes the important part of the detection work, which is also a urgent need to be solved at present.
Disclosure of Invention
In order to solve the problems in the prior art, the invention provides an electromagnetic method pipeline disease identification and positioning method based on an improved Capsule Net network, an improved and optimized Capsule Net detection algorithm is adopted, a convolution layer is additionally arranged on the basis of the Capsule Net for extracting the features at different positions in a picture again, and the feature extraction effect is optimized.
In order to achieve the purpose, the invention provides the following technical scheme: an electromagnetic method pipeline disease identification and positioning method based on an improved CapsNet network comprises the following steps:
s1, establishing an electromagnetic spectrum data set;
s2, processing a data set;
s3, constructing and training a ground penetrating radar map detection model based on the improved CapsNet network;
s4, model parameter adjustment;
s5, performing model test, outputting each evaluation value index, storing an optimal model, and outputting the recognition result of the ground penetrating radar map;
and S6, according to the identification result in the step S5, positioning the pipeline diseases by the ground penetrating radar A-scan data at the corresponding position, and acquiring the horizontal position and the buried depth of the pipeline diseases.
Preferably, the step S1 of establishing the electromagnetic spectrum data set specifically includes: collecting a ground penetrating radar map, screening and preprocessing the map, and establishing an electromagnetic map data set;
the electromagnetic spectrum data set comprises a real radar spectrum, a model test radar spectrum and an FDTD simulation spectrum;
the screening is as follows: images with less noise interference, good underground pipelines, and good characteristics of void and looseness are reserved;
the pretreatment is as follows: the image processing program is used for directly cutting and deleting the blank around the image and the additional useless information.
Preferably, the processing of the data set in step S2 is to expand data by using a data enhancement technique, label the data to generate a label file, and divide the data set in proportion;
the data enhancement technology comprises reverse mirror image, translation clipping, fuzzy processing and contrast adjustment;
the marking refers to marking underground pipelines, void and loose features in the ground penetrating radar map by using a labelme software marking program, wherein the pixels of a background area of a picture are marked as 0, the pixels of a characteristic area of the underground pipelines are marked as 1, the pixels of the feature area of the void are marked as 2, and the pixels of the feature area of the loose are marked as 3;
the step of dividing the data set in proportion refers to: according to the weight ratio of 8: the ratio of 2 is randomly divided into a training set and a testing set.
Preferably, the modified CapsNet network in step S3 includes a cons 1 layer, a Primary capsule layer, a cons 2 layer, a Digit capsule layer and a Fully connected layer.
Preferably, the Convs1 layer extracts local features of the pavement diseases by two convolution operations;
the Primary capsule layer adopts a nerve capsule to extract image information with different characteristics;
the Convs2 layer refines the extraction process of the features when performing convolution operation, one filter extracts one feature, and the training dimensionality is reduced by setting a small number of filters;
the vector output of the Primary capsule layer capsule is obtained by the Digit capsule layer by adopting a Dynamic Routing algorithm;
the full connected layer adopts full connection operation, improves the spatial correlation among image pixels, and accurately extracts a characteristic region.
Preferably, the model training in step S3 is to initialize a network model hyper-parameter, and then import a training set into the model to perform model training; the hyper-parameters comprise an initial learning rate, total iteration times, weight and momentum coefficients; and when the change curve of the loss value function reaches a stable state, the model converges and the training is stopped.
Preferably, the model parameter adjustment in step S4 specifically refers to adjusting the hyper-parameters according to an error loss value between an actual output and a target output of the model when different hyper-parameters are set; the basis of the adjustment is that the error loss value does not decrease any more and tends to be stable.
Preferably, the performing of the model test in step S5 is inputting the test set into the model for testing, and the condition for storing the optimal model is determining whether each of the output evaluation value indicators satisfies a preset value.
Preferably, the evaluation numerical index mainly comprises accuracy, detection efficiency and prediction error.
Preferably, the ground penetrating radar a-scan data in step S6 specifically refers to the amplitude and the two-way travel time of the ground penetrating radar a-scan data.
The invention has the beneficial effects that:
1) according to the method, a composite data set of the real radar map, the model test radar map and the FDTD simulation map is established, the actual detection use scene is oriented, the real radar map accounts for most of the total number of images of the data set, the rest images are used for enriching the data set, and the robustness and the generalization capability of the model are improved.
2) The invention adopts an improved and optimized detection algorithm of the CapsNet, and a convolution layer is additionally arranged on the basis of the CapsNet and is used for extracting the features at different positions in the picture again and optimizing the effect of feature extraction; the original dynamic routing algorithm of the capsule network is improved, so that the parameter quantity is greatly reduced, the scale of the input picture which is allowed to be accepted is increased, the accuracy of model detection is improved, and the intelligent and efficient identification of the pipeline diseases is realized. The method is simple and easy to implement. The interference of the size of the image, the rotation angle and the like is small, the detection precision of the ground penetrating radar map can be well improved, the operation efficiency of the algorithm is high, and the feature extraction capability is strong. And then, on the basis of identification, the positioning of the pipeline diseases is realized according to the ground penetrating radar A-scan data of the corresponding position, and the intelligent level of the engineering detection industry and the detection capability and efficiency are effectively improved.
Drawings
FIG. 1 is a schematic flow chart of the process steps of the present invention;
FIG. 2 is a schematic structural diagram of an improved CapsNet model of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
Referring to fig. 1-2, the present invention provides a technical solution: an electromagnetic method pipeline disease identification and positioning method based on an improved CapsNet network is shown in figure 1 and comprises the following steps:
the method comprises the following steps: creating electromagnetic spectrum data set
Collecting a ground penetrating radar map, screening and preprocessing the map, keeping the map with obvious characteristics, and establishing an underground pipeline data set of the ground penetrating radar, wherein the specific process comprises the following steps:
1.1, arranging the maps to establish a ground penetrating radar map data set, wherein the ground penetrating radar map data set comprises a real radar map, a model test radar map and an FDTD simulation map;
1.2 the radar map is screened by experienced professionals, and images with good underground pipelines, void and loose characteristics are reserved;
1.3 using image processing program, the unnecessary information added to the blank around the image is directly cut and deleted.
Specifically, the radar map is derived from:
1. the method comprises the steps of detecting an underground space by using a real radar map and an actual measurement ground penetrating radar B-Scan profile image of the urban municipal road surface, wherein the radar adopts an SIR-4000 ground penetrating radar to arrange a measuring line on the municipal road surface with a certain length.
2. According to the model test radar map, fully-dried sand is buried in an excavated foundation pit, concrete pipes, PVC plastic pipes, cast iron metal pipes, large-size cuboid boxes (simulated hollowing) and a plurality of gathered small-size cuboid boxes (simulated loosening) are buried at different buried positions and depths, and a ground penetrating radar is used for detecting for multiple times to obtain the radar map.
3. The FDTD simulation atlas is based on research on the field intensity of underground electromagnetic waves, pipelines of various material sizes, hollowing and loosening are designed at different depths in media with different electromagnetic characteristics, and the field intensity of the electromagnetic waves received by an antenna in each direction is calculated according to the field intensity of each point, so that the simulation radar atlas is generated.
According to the method, a composite data set of the real radar map, the model test radar map and the FDTD simulation map is established, the actual detection use scene is oriented, the real radar map accounts for most of the total number of images of the data set, the rest images are used for enriching the data set, and the robustness and the generalization capability of the model are improved.
The main bases for selecting the map are as follows:
1. the maps should be versatile, including single line, multiple lines, metal/non-metal lines, void, loose, etc. different situations.
2. The characteristics in the map are representative and clear and identifiable;
step two: data set processing
Data enhancement is carried out to increase data quantity, label is carried out to generate a label file, and the label file is divided into a training set, a verification set and a test set according to a certain proportion; the specific process is as follows:
2.1, expanding the data volume by adopting a data enhancement technology;
2.2, marking underground pipelines, voids and loose features in the ground penetrating radar map by using a labelme software marking program;
2.3 marking the pixel of the background area of the picture as 0, the pixel of the underground pipeline characteristic area as 1, the pixel of the void characteristic area as 2 and the pixel of the loose characteristic area as 3;
2.4 radar spectra database by 8: the ratio of 2 is randomly divided into a training set and a testing set.
Specifically, the main methods for enhancing the data in 2.1 include reverse mirror image, translational cropping, fuzzy processing, and contrast adjustment.
Specifically, the images of the training set and the test set in 2.4 are not overlapped in a cross mode, and the robustness and the generalization capability of the model can be tested.
Specifically, label files are made for the images of the data set through labelme software, and the labeling process is manually completed by experienced professionals. The characteristic parts of 3 underground targets including pipelines, voids and porosity are selected by a surrounding frame, the surrounding frame needs to just completely surround the characteristic area of the underground pipeline, and the surrounding frame is overlapped with the edge of the characteristic so as to ensure the accuracy and the integrity of the underground target area.
Step three: model training
Constructing a ground penetrating radar map detection model based on optimized CapsNet, initializing network model hyper-parameters, importing a training set into the model, and training the model;
specifically, an improved and optimized CapsNet model framework is established, and as shown in fig. 2, the framework comprises a cons 1 layer, a Primary capsule layer, a cons 2 layer, a Digit capsule layer and a fullly connected layer.
Specifically, the Convs1 layer extracts local features of the pavement diseases by two convolution operations. And the two convolution layers both use 256 convolution kernels, the size of each convolution kernel is 9 x 9, the set depth is 1, and the step size is 2, the convolution operation is carried out, and the result after the convolution is transmitted to a nonlinear activation function ReLU to obtain the output of the Convs layer.
Specifically, the Primary capsule layer adopts a nerve capsule to extract image information with different characteristics. It is a CapsNet version of the convolutional layer used to generate primary capsules containing a large amount of low-level image information. And extracting disease image information by using a neuron group (nerve capsule). This layer first performs a convolution operation on the output of the Convs layer. The layer uses 256 convolution kernels, each with a size of 9 x 9 and set to a depth of 1 step of 2, and repeats 8 parallel convolution operations, followed by reshape reconstruction, to finally output a primary capsule with a dimension of 8 x 1.
Specifically, the Convs2 layer refines the feature extraction process when performing convolution operation, and can set a smaller number of filters, so that one filter extracts one feature, and after the number of filters is reduced, the number of features is reduced, the training dimensionality is reduced, and the time is saved.
Specifically, the vector output of the Primary capsule layer capsule is obtained by the Digit capsule layer by adopting a Dynamic Routing algorithm; a vector of high-level features is stored, which contains two types of parameters. One of them is a weight vector between the Primary Caps layer, and the other is a coupling coefficient, which determines the closeness between each capsule in the layer and each capsule in the previous layer, and the parameters of this part depend on a Dynamic Routing (Dynamic Routing) algorithm.
Specifically, the full connected layer adopts full connection operation, improves spatial correlation between image pixels, and accurately extracts a characteristic region.
Specifically, the model hyper-parameters of the initial learning rate, the total iteration number, the weight, and the momentum coefficient are set to 0.0001, 30000, 0.00015, and 0.95, respectively, and training is performed. And (4) referring to the function change curve of the loss value, converging the model when the model is stable, and stopping training at the moment.
Specifically, the model training includes forward propagation and backward propagation. And the forward propagation puts the training set in the constructed data set into the optimized capsNet in batches to obtain the output of the model. The parameters of the Convs1 layer, the Primary capsule layer and the Convs2 layer are updated by a BP back propagation algorithm, and the update of the Digit capsule layer parameters is determined by a Dynamic Routing algorithm.
Step four: model parameter adjustment
Adjusting each hyper-parameter according to the training effect and performance of the step three, comparing the loss value and the change curve of the accuracy rate in the model training process, searching the optimal hyper-parameter value, and continuously adjusting the optimal hyper-parameter value;
the main basis for adjusting the hyper-parameters in the fourth step is that when different hyper-parameters are set, the performance of the model is represented, namely, the error loss value between the actual output and the target output of the model is calculated, so that the error loss value is not reduced any more and tends to be stable, the model is considered to be converged, and meanwhile, the overfitting is avoided.
Specifically, the main hyper-parameters adjusted in the fourth step are initial learning rate, total iteration times, momentum coefficient and weight. After multiple tuning, the optimal hyper-parameter of the final model is set as: the initial learning rate is 0.0001, the total iteration number is 40000, the weight attenuation value is 0.00015, and the momentum coefficient is 0.9.
Step five: model testing
And (4) introducing a test set to test the model according to the optimal performance model obtained by training in the step four, outputting various evaluation value indexes, judging whether the expected value is reached, and storing the optimal model.
Specifically, in the fifth step, the evaluation numerical index mainly includes accuracy, detection efficiency, and prediction error. Judging whether preset values are met (the accuracy rate is more than 90%, the detection efficiency is 24fps, and the false detection rate and the missing detection rate in the prediction error are all less than 5%); if not, returning to the fourth step; if so, saving the model and each weight parameter, and outputting the recognition result of the ground penetrating radar map.
Specifically, in the sixth step, the main basis for obtaining the position information of the pipeline disease is as follows: amplitude and two-way travel time of the ground penetrating radar A-scan data.
Specifically, in this embodiment, correct and incorrectly identified underground target information should be counted during testing, and it is determined whether the detection and positioning method achieves an accuracy of more than 90%, so as to ensure the practical application capability of the present invention. If the accuracy rate does not meet the expected requirement, errors should be counted and summarized, an error map is searched, the error map is processed according to the method of the invention, and then the error map is added into a ground penetrating radar map database, retrained and adjusted and optimized until all indexes meet the expected requirement.
According to the method, an intelligent algorithm suitable for identifying and positioning the electromagnetic spectrum of the pipeline disease is researched and developed by adopting a deep learning method, the optimized and improved CapsNet model is trained on the basis of the big data of the electromagnetic spectrum, a convolutional layer Convs2 layer is additionally arranged on the model, and the convolutional layer Convs2 layer is positioned behind a Primary capsule layer and used for extracting the features positioned at different positions in the picture again and optimizing the effect of feature extraction; the original dynamic routing algorithm of the capsule network is improved, so that the parameter quantity is greatly reduced, the scale of the input picture which is allowed to be accepted is increased, the accuracy of model detection is improved, and the intelligent and efficient identification of the pipeline diseases is realized. The method is simple and easy to implement, is less interfered by image size, rotation angle and the like, can well improve the detection precision of the ground penetrating radar map, and has high algorithm operation efficiency and strong feature extraction capability.
And then according to the recognition result of the optimized CapsNet model, acquiring the position information of the pipeline diseases by the amplitude change of the echo in the ground penetrating radar A-scan data at the corresponding position and the two-way walking, and assisting the relative dielectric constant of the local geological environment, so as to realize the positioning of the pipeline diseases, form recognition and positioning, and effectively improve the intelligentization level of the engineering detection industry and the detection capability and efficiency.
Although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that various changes in the embodiments and/or modifications of the invention can be made, and equivalents and modifications of some features of the invention can be made without departing from the spirit and scope of the invention.
Claims (10)
1. An electromagnetic method pipeline disease identification and positioning method based on an improved CapsNet network is characterized by comprising the following steps:
s1, establishing an electromagnetic spectrum data set;
s2, processing a data set;
s3, constructing and training a ground penetrating radar map detection model based on the improved CapsNet network;
s4, model parameter adjustment;
s5, performing model test, outputting each evaluation value index, storing an optimal model, and outputting the recognition result of the ground penetrating radar map;
and S6, according to the identification result in the step S5, positioning the pipeline diseases by the ground penetrating radar A-scan data at the corresponding position, and acquiring the horizontal position and the buried depth of the pipeline diseases.
2. The electromagnetic method pipeline disease identification and positioning method based on the improved CapsNet network of claim 1, wherein: the step S1 of establishing the electromagnetic spectrum data set specifically includes: collecting a ground penetrating radar map, screening and preprocessing the map, and establishing an electromagnetic map data set;
the electromagnetic spectrum data set comprises a real radar spectrum, a model test radar spectrum and an FDTD simulation spectrum;
the screening is as follows: images with less noise interference, good underground pipelines, emptiness and loosening characteristics are reserved;
the pretreatment is as follows: the image processing program is used for directly cutting and deleting the blank around the image and the additional useless information.
3. The electromagnetic method pipeline disease identification and positioning method based on the improved CapsNet network of claim 1, wherein: the processing of the data set in the step S2 is to expand the data by adopting a data enhancement technology, label the data to generate a label file, and divide the data set in proportion;
the data enhancement technology comprises reverse mirror image, translation clipping, fuzzy processing and contrast adjustment;
the marking refers to marking underground pipelines, void and loose features in the ground penetrating radar map by using a labelme software marking program, wherein the pixels of a background area of a picture are marked as 0, the pixels of a characteristic area of the underground pipelines are marked as 1, the pixels of the feature area of the void are marked as 2, and the pixels of the feature area of the loose are marked as 3;
the step of dividing the data set in proportion refers to: according to the following steps of 8: the ratio of 2 is randomly divided into a training set and a testing set.
4. The electromagnetic method pipeline disease identification and positioning method based on the improved CapsNet network of claim 1, wherein: the modified CapsNet network in the step S3 includes a cons 1 layer, a Primary capsule layer, a cons 2 layer, a Digit capsule layer, and a Fully connected layer.
5. The electromagnetic method pipeline disease identification and positioning method based on the improved CapsNet network of claim 4, wherein: the Convs1 layer adopts two convolution operations to extract local characteristics of the pavement diseases;
the Primary capsule layer adopts a nerve capsule to extract image information with different characteristics;
the Convs2 layer refines the extraction process of the features when performing convolution operation, one filter extracts one feature, and the training dimensionality is reduced by setting a small number of filters;
the vector output of the Primary capsule layer capsule is obtained by the Digit capsule layer by adopting a Dynamic Routing algorithm;
the full connected layer adopts full connection operation, improves the spatial correlation among image pixels, and accurately extracts a characteristic region.
6. The electromagnetic method pipeline disease identification and positioning method based on the improved CapsNet network of claim 1, wherein: in the step S3, model training refers to initializing a network model hyper-parameter, then importing a training set into a model, and performing model training; the hyper-parameters comprise an initial learning rate, total iteration times, weight and momentum coefficients; and when the change curve of the loss value function reaches a stable state, the model converges and the training is stopped.
7. The electromagnetic method pipeline disease identification and positioning method based on the improved CapsNet network of claim 1, wherein: the model parameter adjustment in the step S4 is specifically to adjust the hyper-parameters according to the error loss value between the actual output and the target output of the model when different hyper-parameters are set; the basis of the adjustment is that the error loss value does not decrease any more and tends to be stable.
8. The electromagnetic method pipeline disease identification and positioning method based on the improved CapsNet network of claim 1, wherein: the step S5 of performing model test is to input the test set into the model for testing, and the condition for storing the optimal model is to determine whether each output evaluation value index meets a preset value.
9. The electromagnetic method pipe disease identification and location method based on the improved CapsNet network of claim 8, wherein: the evaluation numerical index mainly comprises accuracy, detection efficiency and prediction error.
10. The electromagnetic method pipe damage identification and positioning method based on the improved CapsNet network of claim 1, wherein: the ground penetrating radar A-scan data in the step S6 specifically refers to the amplitude and the two-way travel time of the ground penetrating radar A-scan data.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210261292.9A CN114694051A (en) | 2022-03-16 | 2022-03-16 | Electromagnetic method pipeline disease identification and positioning method based on improved CapsNet network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210261292.9A CN114694051A (en) | 2022-03-16 | 2022-03-16 | Electromagnetic method pipeline disease identification and positioning method based on improved CapsNet network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114694051A true CN114694051A (en) | 2022-07-01 |
Family
ID=82139285
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210261292.9A Pending CN114694051A (en) | 2022-03-16 | 2022-03-16 | Electromagnetic method pipeline disease identification and positioning method based on improved CapsNet network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114694051A (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110321859A (en) * | 2019-07-09 | 2019-10-11 | 中国矿业大学 | A kind of optical remote sensing scene classification method based on the twin capsule network of depth |
CN111833449A (en) * | 2020-06-30 | 2020-10-27 | 南京航空航天大学 | Three-dimensional reconstruction and intelligent disease identification method for internal environment of subway tunnel |
CN112231477A (en) * | 2020-10-20 | 2021-01-15 | 淮阴工学院 | Text classification method based on improved capsule network |
US20210097377A1 (en) * | 2018-05-14 | 2021-04-01 | Nokia Technologies Oy | Method and apparatus for image recognition |
CN113205137A (en) * | 2021-04-30 | 2021-08-03 | 中国人民大学 | Image identification method and system based on capsule parameter optimization |
CN113256562A (en) * | 2021-04-22 | 2021-08-13 | 深圳安德空间技术有限公司 | Road underground hidden danger detection method and system based on radar images and artificial intelligence |
CN113392705A (en) * | 2021-05-13 | 2021-09-14 | 电子科技大学 | Method for identifying pipeline leakage target in desert area based on convolutional neural network |
CN113424055A (en) * | 2019-10-09 | 2021-09-21 | 山东大学 | Multi-scale tunnel structure disease detection and intelligent diagnosis system and method |
-
2022
- 2022-03-16 CN CN202210261292.9A patent/CN114694051A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20210097377A1 (en) * | 2018-05-14 | 2021-04-01 | Nokia Technologies Oy | Method and apparatus for image recognition |
CN110321859A (en) * | 2019-07-09 | 2019-10-11 | 中国矿业大学 | A kind of optical remote sensing scene classification method based on the twin capsule network of depth |
CN113424055A (en) * | 2019-10-09 | 2021-09-21 | 山东大学 | Multi-scale tunnel structure disease detection and intelligent diagnosis system and method |
CN111833449A (en) * | 2020-06-30 | 2020-10-27 | 南京航空航天大学 | Three-dimensional reconstruction and intelligent disease identification method for internal environment of subway tunnel |
CN112231477A (en) * | 2020-10-20 | 2021-01-15 | 淮阴工学院 | Text classification method based on improved capsule network |
CN113256562A (en) * | 2021-04-22 | 2021-08-13 | 深圳安德空间技术有限公司 | Road underground hidden danger detection method and system based on radar images and artificial intelligence |
CN113205137A (en) * | 2021-04-30 | 2021-08-03 | 中国人民大学 | Image identification method and system based on capsule parameter optimization |
CN113392705A (en) * | 2021-05-13 | 2021-09-14 | 电子科技大学 | Method for identifying pipeline leakage target in desert area based on convolutional neural network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Liu et al. | Detection and localization of rebar in concrete by deep learning using ground penetrating radar | |
Liu et al. | Automatic recognition and localization of underground pipelines in GPR B-scans using a deep learning model | |
CN113256562B (en) | Road underground hidden danger detection method and system based on radar images and artificial intelligence | |
Shi et al. | Nonparametric and data-driven interpolation of subsurface soil stratigraphy from limited data using multiple point statistics | |
CN102918423B (en) | Method for earthquake hydrocarbon system anlysis | |
CN115063525B (en) | Three-dimensional mapping method and device for urban road subgrade and pipeline | |
CN112731522B (en) | Intelligent recognition method, device and equipment for seismic stratum and storage medium | |
CN107462924A (en) | A kind of absolute wave impedance inversion method independent of well-log information | |
Liu et al. | Whale optimization algorithm-based point cloud data processing method for sewer pipeline inspection | |
KR20180055518A (en) | System and method of 3d mineral prospectivity | |
Chen et al. | A novel image-based approach for interactive characterization of rock fracture spacing in a tunnel face | |
CN116203559A (en) | Intelligent recognition and early warning system and method for underground rock and soil disease body | |
CN115983505A (en) | Solid mineral three-dimensional ore formation prediction method and device | |
Yue et al. | Automatic recognition of defects behind railway tunnel linings in GPR images using transfer learning | |
Zhu et al. | Evaluating the optimised YOLO-based defect detection method for subsurface diagnosis with ground penetrating radar | |
Zhao et al. | A Comprehensive Horizon‐Picking Method on Subbottom Profiles by Combining Envelope, Phase Attributes, and Texture Analysis | |
Hu et al. | Defects identification and location of underground space for ground penetrating radar based on deep learning | |
CN117351321A (en) | Single-stage lightweight subway lining cavity recognition method and related equipment | |
CN114694051A (en) | Electromagnetic method pipeline disease identification and positioning method based on improved CapsNet network | |
Guo et al. | Research on tunnel lining image target recognition method based on YOLOv3 | |
CN112285773B (en) | Quantitative characterization method for binary structure of buried fault | |
CN115099111A (en) | Shield construction ground surface settlement prediction method based on machine learning and finite element | |
Msaddek et al. | A fuzzy mathematical model for evaluation of rock-fracture and structural complexity: application for Southern Atlas in Tunisia | |
CN112800664A (en) | Method for estimating tree root diameter based on ground penetrating radar A-scan data | |
Li et al. | Factor analysis of affecting the accuracy for intelligent picking of seismic first arrivals with deep learning model |
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 |