CN115630492B - Tunnel lining defect change characteristic intelligent inversion method, system and storage medium - Google Patents
Tunnel lining defect change characteristic intelligent inversion method, system and storage medium Download PDFInfo
- Publication number
- CN115630492B CN115630492B CN202211246860.4A CN202211246860A CN115630492B CN 115630492 B CN115630492 B CN 115630492B CN 202211246860 A CN202211246860 A CN 202211246860A CN 115630492 B CN115630492 B CN 115630492B
- Authority
- CN
- China
- Prior art keywords
- scan data
- tunnel lining
- phase
- dimensional
- data
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 230000008859 change Effects 0.000 title claims abstract description 35
- 230000007547 defect Effects 0.000 title claims abstract description 33
- 238000003062 neural network model Methods 0.000 claims abstract description 45
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 20
- 238000012545 processing Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims description 35
- 238000005070 sampling Methods 0.000 claims description 20
- 238000000819 phase cycle Methods 0.000 claims description 18
- 238000004088 simulation Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 13
- 230000004044 response Effects 0.000 claims description 10
- 238000003384 imaging method Methods 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 238000002592 echocardiography Methods 0.000 claims description 4
- 230000001427 coherent effect Effects 0.000 claims description 3
- 230000004927 fusion Effects 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 claims description 2
- 238000010276 construction Methods 0.000 claims 1
- 230000003902 lesion Effects 0.000 claims 1
- 201000010099 disease Diseases 0.000 abstract description 16
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 abstract description 16
- 238000001514 detection method Methods 0.000 abstract description 8
- 238000013527 convolutional neural network Methods 0.000 description 12
- 238000010586 diagram Methods 0.000 description 5
- 230000009466 transformation Effects 0.000 description 5
- 230000006872 improvement Effects 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000007613 environmental effect Effects 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000009659 non-destructive testing Methods 0.000 description 2
- 230000000149 penetrating effect Effects 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000877 morphologic effect Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 230000003442 weekly effect Effects 0.000 description 1
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
-
- 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
-
- 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)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Geometry (AREA)
- Computer Hardware Design (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses an intelligent inversion method, system and storage medium for tunnel lining defect change characteristics, which comprises the following steps: geological radar data acquisition is carried out on the tunnel lining region to be tested in a plurality of different time periods, and B-scan data of the tunnel lining region to be tested in a plurality of time periods are obtained; respectively carrying out synthetic aperture focusing algorithm processing on the B-scan data of each time period to correspondingly obtain focusing B-scan data, and forming two-dimensional B-scan data pairs corresponding to each time period of the tunnel lining region to be tested; respectively inputting two-dimensional B-scan data pairs corresponding to each time period into a trained neural network model; the neural network model is formed by cascading a phase slice inversion module and a phase slice inversion module through a four-dimensional inversion module; and outputting a dielectric constant model corresponding to each time period by the neural network model to obtain the change characteristics of the defects of the tunnel lining region to be tested. The invention realizes the dynamic detection of the diseases in the region to be detected and can greatly improve the working efficiency and inversion stability.
Description
Technical Field
The invention belongs to the technical fields of civil engineering, geotechnical engineering and nondestructive testing, and particularly relates to an intelligent inversion method, system and storage medium for tunnel lining damage change characteristics.
Background
Geological radar (GPR) is widely applied to the fields of civil engineering, geotechnical engineering, nondestructive detection and the like by virtue of the characteristics of no damage, high efficiency, high anti-interference level, strong penetrating power and the like. The geological radar is used as a nondestructive testing tool capable of acquiring underground internal information, and the working flow is that after a region to be tested is selected, test parameters such as survey line layout, grid size, measuring point spacing and the like are determined, and the internal structure of an object is detected by transmitting electromagnetic waves and receiving echoes. The geological radar can obtain a piece of B-scan data reflecting underground information in a co-offset two-dimensional section measurement mode. Only a single piece of B-Scan data is detected and identified, accurate positioning and quantitative judgment of three-dimensional diseases are difficult to perform due to the lack of three-dimensional space width direction information, and the fact that geological radar B scanning acquired by adjacent channels has continuity along the width direction is considered, and underground information can be acquired and analyzed more completely through processing a plurality of two-dimensional B-Scan data in a space.
In terms of interpretation of geological radar data, an expert is required to interpret various subsurface objects in the geological radar data, including environmental complexity and experimental noise. Moreover, analyzing large amounts of geological radar data with the naked eye is an inefficient and time consuming task. Therefore, it is important to develop a rapid and accurate automatic detection technology for underground targets. In recent years, convolutional Neural Network (CNN) based methods have been widely used to learn characteristic representations directly from input data. Researchers have developed CNN-based data interpretation methods for automatically detecting targets from GPR data. The main problem of the conventional CNN-based method for geological radar data interpretation is that a single piece of B-scan data is used for interpretation and analysis only for a certain time state of a region to be detected, and the works of three-dimensional dynamic defect detection, target inversion and the like are not considered by utilizing the spatial characteristic information of a plurality of pieces of B-scan data in different time periods.
Disclosure of Invention
The invention aims to solve the defects in the background art, and provides an intelligent inversion method, an intelligent inversion system and a storage medium for tunnel lining damage change characteristics, which are based on a cascading CNN network structure and utilize spatial characteristic information of a plurality of B-scan data in different time periods to realize the dynamic detection of damage in a region to be detected, and can greatly improve the working efficiency and inversion stability.
The technical scheme adopted by the invention is as follows: an intelligent inversion method for tunnel lining defect change characteristics comprises the following steps:
geological radar data acquisition is carried out on the tunnel lining region to be tested in a plurality of different time periods, and B-scan data of the tunnel lining region to be tested in a plurality of time periods are obtained;
respectively carrying out synthetic aperture focusing algorithm processing on the B-scan data of each time period to correspondingly obtain focused B-scan data, and combining the B-scan data of each time period and the focused B-scan data to form two-dimensional B-scan data pairs corresponding to each time period of the tunnel lining region to be tested;
respectively inputting two-dimensional B-scan data pairs corresponding to each time period into a trained neural network model; the neural network model is formed by cascading a time phase slice inversion module and a phase slice inversion module through a four-dimensional inversion module of time phase and phase;
the neural network model outputs a dielectric constant model corresponding to each time period;
and acquiring the positions and the shapes of defects of the region to be detected in different time periods through dielectric constant models corresponding to the time periods, and obtaining the change characteristics of the defects of the tunnel lining region to be detected.
In the technical scheme, after the B-scan data of the region to be detected is obtained, direct wave removal and noise removal pretreatment operation is carried out on the B-scan data; after the pretreatment, the B-scan data is processed by a synthetic aperture focusing algorithm. By converging and focusing the echo signals, the influence of interference signals can be weakened, the transverse resolution can be improved, diseases in the data image are closer to the real size and position, characteristic response signals in the image are enhanced, and high-quality two-dimensional B-scan pairs required by neural network model training are conveniently constructed, so that the neural network model can better extract and analyze data characteristic information, the accuracy of the neural network model training can be improved, and a more accurate recognition effect can be achieved.
According to the technical scheme, a plurality of parallel geological radar measuring lines are uniformly distributed in a tunnel lining region to be tested, and the geological radar measuring lines cover the tunnel lining region to be tested; and acquiring geological radar survey line data through a geological radar in the same time period to obtain B-scan data of the tunnel lining region to be tested in the time period. By expanding the transverse coverage area, the three-dimensional space characteristic information of the disease can be efficiently obtained on the basis that the disease in the tunnel lining can only be positioned by the traditional single measuring line, and a data basis is provided for space-time change of the space characteristic of the disease. The duration of each time period is the time required by the geological radar data acquisition equipment to acquire all geological radar line data in the tunnel lining area to be detected. An operator can set the tunnel lining region to be tested according to actual requirements to perform geological radar data acquisition frequency (such as weekly, every three days or once daily). And in a set period (such as a period of a plurality of months or a year), selecting a certain consistent time point according to a set frequency, and adopting geological radar data acquisition equipment to start acquiring all geological radar line data in the tunnel lining area to be detected. And the B-scan data acquired each time is used as the B-scan data of a corresponding certain time period.
In the technical scheme, the geological radar acquires data of the geological radar survey line in different time periods; the geological radar carrying positioning system and the ranging system realize the initial calibration of geological radar measuring lines, and the measuring lines, the grid size and the grid size of data acquisition are adjusted so that the phase information repeatedly measured at the same appointed position in a plurality of time periods is kept consistent. The accurate registration of B-scan data of a plurality of different time periods at the same position in space is guaranteed, so that the information difference of geological radar data of a plurality of time periods is conveniently utilized, and the accuracy of disease dynamic change analysis in a tunnel lining model is improved.
In the above technical solution, the process of constructing the neural network model includes:
carrying out multi-position geological radar data acquisition on the known tunnel lining simulation model in different time periods; acquiring B-scan data of different positions of a known tunnel lining model in different time periods;
b-scan data of different positions of the known tunnel lining model in different time periods are subjected to synthetic aperture focusing algorithm processing to obtain focusing B-scan data of the corresponding different positions of the known tunnel lining model in different time periods;
the data of different time periods are defined as time phase sequences, the data of different positions are defined as phase sequences, and a two-dimensional B-scan data contrast time phase sequence and a two-dimensional B-scan data contrast phase sequence of a known tunnel lining simulation model are formed;
respectively constructing training sets of a phase slice inversion module, a phase slice inversion module and a four-dimensional inversion module:
the single sample information of the training set of the phase slice inversion module includes: the two-dimensional B-scan data pair of the previous time phase, the two-dimensional B-scan data pair of the current time phase and the two-dimensional B-scan data pair of the next time phase are used as inputs; taking a dielectric constant model of the current-phase two-dimensional B-scan data pair as an inverted label;
the single sample information of the training set of the phase slice inversion module includes: taking a previous bit-phase two-dimensional B-scan data pair, a current bit-phase two-dimensional B-scan data pair and a next bit-phase two-dimensional B-scan data pair as inputs, and taking a dielectric constant model of the current bit-phase two-dimensional B-scan data pair as an inversion label;
the single sample information of the training set of the four-dimensional inversion module includes: the output of the phase slice inversion module and the current output of the phase slice inversion module are used as inputs, and a dielectric constant model of a current phase two-dimensional B-scan data pair is used as an inversion label.
And training the neural network model by adopting a training set.
According to the invention, a neural network model is constructed and trained through a cascade phase slice inversion module, a phase slice inversion module and a four-dimensional inversion module. Firstly classifying the neural network model in the time dimension and the space dimension respectively, then carrying out disease feature recognition in different dimensions, further correlating the feature information of the disease in the space and the time, and then fully utilizing the highly relevant characteristics of the disease development in the time dimension and the space dimension to improve the accuracy and the stability of the tunnel lining change feature inversion of the neural network model.
According to the technical scheme, the time phase slice inversion module and the phase slice inversion module both adopt a Unet network structure integrated into a focusing imaging processing branch, SAR image data is input into the traditional Unet structure, so that the ground penetrating radar data reach high-layer resolution, characteristic information is extracted more clearly and comprehensively, the Unet network uses channel splicing as a fusion mode of the characteristic diagrams, convolutional input is richer, and inversion results obtained in time dimension and space dimension through the two inversion modules can reflect original characteristic information of lining disease data.
The encoder modules of the phase slice inversion module and the phase slice inversion module adopt three channels of B-scan data, focusing B-scan data and dielectric constant model for input, and in the up-sampling stage, the data obtained by the last down-sampling is up-sampled after passing through a convolution layer; the up-sampling data, the B-scan data with the same size and the same channel number in the down-sampling process and the focusing B-scan data are subjected to channel splicing, then the channel number is changed through a convolution layer to carry out up-sampling again, and the channel splicing, convolution and up-sampling operations are repeated; up-sampling is carried out for 3 times in total; the final up-sampled data is subjected to convolution extraction and fusion of channel characteristic information, so that data dimension reduction is realized, and an output result is obtained;
and a two-channel Unet network is adopted as a four-dimensional inversion module.
In the above technical solution, the synthetic aperture focusing algorithm processing is performed on the B-scan data, and the process of obtaining the focused B-scan data includes: and firstly calculating the double-pass time delay from one point to be imaged in the region image to each channel, then searching the scattering response amplitude corresponding to the time delay point on each channel receiving echo, finally performing time domain coherent superposition on the scattering response amplitude of the point in all channel echoes, obtaining the backward scattering intensity value of the target according to the amplitude result, and performing the above operation on each point in the imaging region to realize the focusing treatment on the region image, so that the target on the imaging section is focused to a small-range strong-energy region, the transverse resolution of B-scan data is improved, and the accurate identification capability of the target is improved.
According to the technical scheme, the trained neural network model is subjected to fine adjustment by adopting a small amount of known real tunnel lining region geological radar data, the model is optimized by the real tunnel data, and the accuracy and stability of the application of the cascade neural network model in actual engineering are improved.
The invention provides an intelligent inversion system for tunnel lining damage change characteristics, which comprises a B-scan data acquisition module, a two-dimensional B-scan data pair generation module, a dielectric constant model acquisition module and a change characteristic acquisition module, wherein the two-dimensional B-scan data pair generation module is used for generating a dielectric constant model;
the B-scan data acquisition module is used for acquiring geological radar data of the tunnel lining region to be tested in a plurality of different time periods, and acquiring B-scan data of the tunnel lining region to be tested in a plurality of time periods;
the two-dimensional B-scan data pair generation module is used for respectively carrying out synthetic aperture focusing algorithm processing on the B-scan data of each time period to correspondingly obtain focusing B-scan data, and combining the B-scan data of each time period and the focusing B-scan data to form two-dimensional B-scan data pairs corresponding to each time period of the tunnel lining region to be tested;
the dielectric constant model acquisition module is used for respectively inputting two-dimensional B-scan data pairs corresponding to each time period into the trained neural network model; the neural network model is formed by cascading a phase slice inversion module and a phase slice inversion module through a phase and phase inversion module; the neural network model outputs a dielectric constant model corresponding to each time period;
the change characteristic acquisition module is used for acquiring the positions and the shapes of defects of the region to be detected in different time periods through dielectric constant models corresponding to the time periods, and obtaining the change characteristics of the defects of the tunnel lining region to be detected.
The invention provides a computer readable storage medium, which stores a computer program, wherein the computer program, when executed by a processor, causes the processor to execute the steps of the intelligent inversion method for the tunnel lining disease change characteristics according to the technical scheme. .
The beneficial effects of the invention are as follows: according to the invention, a tunnel lining defect change characteristic intelligent inversion method is adopted, and a high-resolution image and a 'B-scan and focusing B-scan' data pair for constructing subsequent CNN network learning training are obtained by carrying out synthetic aperture focusing (SAR) algorithm processing on geological radar data of multiple time periods and multiple positions of the tunnel lining; and outputs of the time phase slice inversion module and the phase slice inversion module are cascaded through a two-channel Unet network, so that four-dimensional time phase and phase inversion is realized, disease defect morphological contours at different positions of different time sequences can be obtained rapidly, further, rapid detection of three-dimensional dynamics of lining diseases in different time periods is realized, and detection working efficiency and inversion stability are improved greatly.
According to the neural network model, learning training is carried out based on two-dimensional B-scan data pairs, SAR branches are added into a conventional CNN inversion network architecture, constraint information based on synthetic aperture focusing algorithm data is increased, and recognition accuracy is improved. According to the invention, the neural network model fuses time phase sequence information and phase sequence information, SAR branches are introduced into a neural network structure, SAR data are added for splicing in the process of channel splicing of up-sampled and down-sampled data, SAR data characteristic information is fused, and channel splicing is performed by combining local information and integral information, so that the accuracy of identification is improved.
Drawings
FIG. 1 is a flow chart of an intelligent inversion method for tunnel lining defect change characteristics;
FIG. 2 is a model structure diagram of the inversion of a dielectric constant model of three-dimensional disease change characteristics of the geological radar of the cascade CNN;
FIG. 3 is a diagram of a "B-scan, focused B-scan" data pair set constructed in accordance with the present invention and its corresponding dielectric constant model;
FIG. 4 is a diagram of a model structure of a Unet inversion module incorporated into SAR branches;
FIG. 5 is a result diagram of a cascading Unet network recognition simulation model in the present invention;
FIG. 6 is a graph of the results of the cascading Unet network recognition of a real tunnel lining in the present invention.
Detailed Description
The invention will now be described in further detail with reference to the drawings and specific examples, which are given for clarity of understanding and are not to be construed as limiting the invention.
The invention discloses an intelligent inversion method for tunnel lining defect change characteristics, which comprises the following steps:
geological radar data acquisition is carried out on the tunnel lining region to be tested in a plurality of different time periods, and B-scan data of the tunnel lining region to be tested in a plurality of time periods are obtained;
respectively carrying out synthetic aperture focusing algorithm processing on the B-scan data of each time period to correspondingly obtain focused B-scan data, and combining the B-scan data of each time period and the focused B-scan data to form two-dimensional B-scan data pairs corresponding to each time period of the tunnel lining region to be tested;
respectively inputting two-dimensional B-scan data pairs corresponding to each time period into a trained neural network model; the neural network model is formed by cascading a phase slice inversion module and a phase slice inversion module through a phase and phase inversion module;
the neural network model outputs a dielectric constant model corresponding to each time period;
and acquiring the positions and the shapes of defects of the region to be detected in different time periods through dielectric constant models corresponding to the time periods, and obtaining the change characteristics of the defects of the tunnel lining region to be detected.
1-2, in one embodiment of the invention, a tunnel lining defect change feature intelligent inversion method specifically comprises the following steps:
firstly, collecting geological radar data for multiple times at different positions of a known tunnel lining simulation model in different time periods under the same environmental condition, and performing data preprocessing operations such as direct wave removal, noise removal and the like on the obtained B-scan data.
Secondly, after preprocessing B-scan data of different positions of different time periods of a known tunnel lining simulation model, performing synthetic aperture focusing (SAR) algorithm processing on the B-scan data, and constructing a B-scan and focusing B-scan data pair, namely a two-dimensional B-scan data pair, wherein the effect is shown in a figure 3, the two-dimensional B-scan data pair takes a data format as input of a neural network model, and can be presented in a picture form when the simulation model is visualized; the neural network model adopts a CNN cascade network and adopts a dielectric constant model as a training label. Learning training is carried out on the basis of data pairs in a subsequent CNN network, and constraint information based on a synthetic aperture focusing algorithm is increased by adding SAR branches into a conventional CNN inversion network architecture.
The synthetic aperture focusing algorithm branch refers to that resolution characteristic data under different scales are obtained through convolution, pooling and other operations on 'focusing B-scan' data, in the process of up-sampling of a CNN inversion network structure, the characteristics coded under the same scale are fused into a decoding process, and the identification result index is improved by adding constraint information based on the synthetic aperture focusing data.
The synthetic aperture focusing (SAR) algorithm processing principle is that a double-pass time delay of a point to be imaged in an area image is calculated firstly, then a scattering response amplitude corresponding to the time delay point is searched on each channel receiving echo, finally, the scattering response amplitudes of the point in all channel echoes are subjected to time domain coherent superposition, a backward scattering intensity value of a target is obtained according to an amplitude result, and the area can be focused by performing the operation on each point in the imaging area, so that a high-resolution image is obtained. The core idea of the SAR algorithm is delay and accumulation.
The specific SAR algorithm calculation processing steps are as follows:
1) Calculate data point A to each passDouble-pass delay of tracks, the delay being τ A ,k Representation, then:
where k is the channel index, v is the propagation velocity, related to the dielectric constant of the medium, x k X is the abscissa of the channel A ，z A The abscissa and ordinate of the imaging point a.
2) Calculating the scattering response amplitude x of the data point A in each channel A,k ：
x A,k ＝r k (t)
Wherein t is τ A,k ，r k And (t) represents echo data of the geological radar in each channel.
3) And coherently superposing scattering response amplitude values generated by the data point A in each channel to complete focusing of the point A.
Where EA is the scatter response overlap value and M is the number of channels.
4) And processing all points covered by the B-scan data at different positions of different time periods of the known tunnel lining simulation model according to the steps, and correspondingly forming focused B-scan data.
And thirdly, defining data acquired in different time periods as a time phase sequence, defining data acquired in different positions as a phase sequence, dividing 'B-scan' data pair into two dimensions of the time phase sequence and the phase sequence, and forming a two-dimensional B-scan data pair time phase sequence and a two-dimensional B-scan data pair phase sequence of a known tunnel lining simulation model.
The two-dimensional B-scan data pairs acquired by different phases in the same phase are used for training the phase slice inversion module, and the phase slice inversion module are connected through the four-dimensional 'phase + phase' inversion module to perform neural network model training, as shown in fig. 2.
Specifically, the training set of the neural network model is divided into a training set of a phase slice inversion module, a phase slice inversion module and a four-dimensional inversion module.
The single sample information of the training set of the phase slice inversion module comprises: the two-dimensional B-Scan data pair of the previous time phase, the two-dimensional B-Scan data pair of the current time phase and the two-dimensional B-Scan data pair of the next time phase are used as inputs; the permittivity model of the current phase slice is used as the inversion label.
The single sample information of the training set of the phase slice inversion module comprises: the input is a previous bit phase two-dimensional B-Scan data pair, a current bit phase two-dimensional B-Scan data pair and a next bit phase two-dimensional B-Scan data pair; the permittivity model of the current phase slice is used as an inversion label.
The single sample information of the training set of the four-dimensional inversion module comprises: the output of the phase slice inversion module and the output of the phase slice inversion module are used as the input of the four-dimensional phase and phase inversion module, and the dielectric constant model of the current phase slice of the current phase is used as an inversion label.
And training the neural network model by adopting the training set, and using the trained neural network model for outputting a corresponding dielectric constant model based on the output B-Scan data.
Loss functions are represented by loss in fig. 2; and utilizing the recognition result and the training label in the neural network model training process to make loss, and continuously optimizing the parameters of the neural network model.
The phase slice inversion module and the phase slice inversion module are both improved by a Unet network structure, and the Unet network structure is a network based on the improvement of a full convolution neural network, and two parts of an encoder and a decoder are approximately bilaterally symmetrical and are like capital U, so that the network is named as a Unet network.
The improvement of the neural network model is divided into two aspects, one is that the single-channel input of an encoder module of a Unet network structure is changed into three-channel input, so that the phase sequence information and the phase sequence information are conveniently fused. The improvement of another aspect is that SAR branches are introduced into a Unet network structure, SAR data are added for splicing in the process of channel splicing of the data obtained by up-sampling and the data with the same size in the down-sampling process, SAR data characteristic information is fused, and channel splicing is carried out by combining local information and integral information, so that the accuracy of identification is improved. Specifically, in the up-sampling stage, the up-sampling is performed after the data obtained by the last down-sampling passes through the convolution layer in the first step. And secondly, channel splicing is carried out on the data, simulation data with the same size and the same channel number and SAR imaging data in the downsampling process, then the channel number is changed through a convolution layer to carry out upsampling again, channel splicing, convolution and upsampling operations are repeated, and the second step carries out upsampling for 3 times. And thirdly, extracting and fusing channel characteristic information from the last up-sampled data through convolution to realize data dimension reduction and obtain an output identification result. The outputs of the phase slice inversion module and the phase slice inversion module are cascaded through a two-channel Unet network to realize four-dimensional phase and phase inversion, and the flow is shown in figure 4.
Geological radar data (GPR simulation image shown in fig. 5 and representing B-Scan data after preprocessing) generated by adopting a simulation model, a recognition result (fused SAR branch inversion result shown in fig. 5) of a neural network model for a certain sample after training is completed, and a training label (dielectric constant model shown in fig. 5) corresponding to the training sample. As shown in fig. 5, the tunnel lining simulation model identification results including cracks (upper defect) and voids (lower defect) are shown. And comparing the training label with the recognition result of the neural network model after training, and verifying the effectiveness of the neural network model after training.
Fourthly, performing multi-time-period and multi-position data acquisition on a target detection area (namely a tunnel lining area to be detected) through a positioning system and a ranging system which are constructed by a laser ranging module, a real-time quality control module, a data acquisition and storage module, a target coarse positioning module and the like, obtaining B-scan data of the tunnel lining area to be detected in a plurality of time periods, and performing preprocessing operations such as denoising, direct wave interference estimation, removal and the like.
The initial calibration of the geological radar measuring line is realized through the geological radar carrying positioning system and the ranging system to adjust the collecting parameters such as the measuring line, the grid size and the like of tunnel lining disease data collection, and the consistency of phase information of repeated measurement in a plurality of time phases is ensured. And analyzing and monitoring the dynamic change of the damage in the tunnel lining model by utilizing the information difference of geological radar data acquired at regular intervals.
Based on the characteristics of the direct wave of the geological radar signal, a two-dimensional physical wavelet is used as a basic wavelet to carry out wavelet transformation on the geological radar signal, and proper wavelet scale is selected to estimate the direct wave interference so as to remove the direct wave. Based on the characteristic that the effective components of the reflected waves of adjacent channels have stronger correlation in waveform and energy, the KL transformation can be used for retaining the correlation signals in a certain direction so as to suppress uncorrelated noise and other waves, thereby achieving the purpose of removing noise in geological radar signals. In terms of noise removal, KL transformation is to globally decompose echo signals based on differences in signal correlation in the depth direction, that is, a KL transformation can be used to retain correlation signals in a certain direction so as to suppress uncorrelated noise and other waves, so that a good denoising effect can be obtained easily.
The transmitting antenna directly reaches the receiving antenna through space and the earth surface without reflecting part of energy by an underground target, and the received signal is called direct wave. From the two-dimensional physical wavelet characteristics, when the spatial scale factor becomes large, the two-dimensional physical wavelet function becomes gentle, which is consistent with the characteristics of the direct wave. And carrying out two-dimensional physical wavelet transformation on the two-dimensional geological radar signal, and only selecting a large spatial scale to reconstruct the signal, so that estimation and removal of direct wave interference can be realized.
And fifthly, adopting the same method as in the step two to respectively perform synthetic aperture focusing algorithm processing on the B-scan data of each time period to correspondingly obtain focusing B-scan data, and combining the B-scan data of each time period and the focusing B-scan data to form two-dimensional B-scan data pairs corresponding to each time period of the tunnel lining region to be tested.
Step six, respectively inputting two-dimensional B-scan data pairs corresponding to each time period into a trained neural network model; the neural network model outputs a dielectric constant model corresponding to each time period, as shown in fig. 6, in which dielectric constant models of the same region to be measured obtained in two different time periods and corresponding GPR data are shown.
And seventhly, acquiring the positions and the shapes of defects of the region to be tested in different time periods through dielectric constant models corresponding to the time periods, and obtaining the change characteristics of the defects of the tunnel lining region to be tested.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
What is not described in detail in this specification is prior art known to those skilled in the art.
Claims (9)
1. An intelligent inversion method for tunnel lining defect change characteristics is characterized by comprising the following steps: the method comprises the following steps:
geological radar data acquisition is carried out on the tunnel lining region to be tested in a plurality of different time periods, and B-scan data of the tunnel lining region to be tested in a plurality of time periods are obtained;
respectively carrying out synthetic aperture focusing algorithm processing on the B-scan data of each time period to correspondingly obtain focused B-scan data, and combining the B-scan data of each time period and the focused B-scan data to form two-dimensional B-scan data pairs corresponding to each time period of the tunnel lining region to be tested;
respectively inputting two-dimensional B-scan data pairs corresponding to each time period into a trained neural network model; the neural network model is formed by cascading a phase slice inversion module and a phase slice inversion module through a four-dimensional inversion module;
the neural network model outputs a dielectric constant model corresponding to each time period;
obtaining positions and shapes of defects of the region to be tested in different time periods through dielectric constant models corresponding to the time periods, and obtaining change characteristics of the defects of the tunnel lining region to be tested;
the construction process of the neural network model comprises the following steps:
carrying out multi-position geological radar data acquisition on the known tunnel lining simulation model in different time periods; acquiring B-scan data of different positions of a known tunnel lining model in different time periods;
b-scan data of different positions of the known tunnel lining model in different time periods are subjected to synthetic aperture focusing algorithm processing to obtain focusing B-scan data of the corresponding different positions of the known tunnel lining model in different time periods;
the data of different time periods are defined as time phase sequences, the data of different positions are defined as phase sequences, and a two-dimensional B-scan data contrast time phase sequence and a two-dimensional B-scan data contrast phase sequence of a known tunnel lining simulation model are formed;
respectively constructing training sets of a phase slice inversion module, a phase slice inversion module and a four-dimensional inversion module:
the single sample information of the training set of the phase slice inversion module includes: the two-dimensional B-scan data pair of the previous time phase, the two-dimensional B-scan data pair of the current time phase and the two-dimensional B-scan data pair of the next time phase are used as inputs; taking a dielectric constant model of the current-phase two-dimensional B-scan data pair as an inverted label;
the single sample information of the training set of the phase slice inversion module includes: taking a previous bit-phase two-dimensional B-scan data pair, a current bit-phase two-dimensional B-scan data pair and a next bit-phase two-dimensional B-scan data pair as inputs, and taking a dielectric constant model of the current bit-phase two-dimensional B-scan data pair as an inversion label;
the single sample information of the training set of the four-dimensional inversion module includes: taking the output of the phase slice inversion module and the current output of the phase slice inversion module as inputs, and taking a dielectric constant model of a current phase two-dimensional B-scan data pair as an inversion label;
and training the neural network model by adopting a training set.
2. The intelligent inversion method for tunnel lining defect change characteristics according to claim 1, wherein the method comprises the following steps: after B-scan data of the region to be detected are obtained, direct wave removal and noise removal preprocessing operation is carried out on the B-scan data; after the pretreatment, the B-scan data is processed by a synthetic aperture focusing algorithm.
3. The intelligent inversion method for tunnel lining defect change characteristics according to claim 1, wherein the method comprises the following steps: uniformly distributing a plurality of parallel geological radar measuring lines in a tunnel lining region to be tested, wherein the geological radar measuring lines cover the tunnel lining region to be tested; and acquiring geological radar survey line data through a geological radar in the same time period to obtain B-scan data of the tunnel lining region to be tested in the time period.
4. A tunnel lining defect change characteristic intelligent inversion method according to claim 3, wherein: the geological radar acquires data of the geological radar survey line in different time periods; the geological radar carrying positioning system and the ranging system realize the initial calibration of geological radar measuring lines, and the measuring lines, the grid size and the grid size of data acquisition are adjusted so that the phase information repeatedly measured at the same appointed position in a plurality of time periods is kept consistent.
5. The intelligent inversion method for tunnel lining defect change characteristics according to claim 1, wherein the method comprises the following steps: the phase slice inversion module and the phase slice inversion module both adopt a Unet network structure integrated into a focusing imaging processing branch;
the encoder modules of the phase slice inversion module and the phase slice inversion module adopt three channels of B-scan data, focusing B-scan data and dielectric constant model for input, and in the up-sampling stage, the data obtained by the last down-sampling is up-sampled after passing through a convolution layer; the up-sampling data, the B-scan data with the same size and the same channel number in the down-sampling process and the focusing B-scan data are subjected to channel splicing, then the channel number is changed through a convolution layer to carry out up-sampling again, and the channel splicing, convolution and up-sampling operations are repeated; up-sampling is carried out for 3 times in total; the final up-sampled data is subjected to convolution extraction and fusion of channel characteristic information, so that data dimension reduction is realized, and an output result is obtained;
and a two-channel Unet network is adopted as a four-dimensional inversion module.
6. The intelligent inversion method for tunnel lining defect change characteristics according to claim 1, wherein the method comprises the following steps: the synthetic aperture focusing algorithm processing is carried out on the B-scan data, and the process of obtaining the focused B-scan data comprises the following steps: calculating the double-pass time delay from one point to be imaged in the region image to each channel, searching the scattering response amplitude corresponding to the time delay point on each channel receiving echo, performing time domain coherent superposition on the scattering response amplitude of the point in all channel echoes, obtaining the backward scattering intensity value of the target according to the amplitude result, and performing the operation on each point in the imaging region to realize focusing processing of the region image.
7. The intelligent inversion method for tunnel lining defect change characteristics according to claim 1, wherein the method comprises the following steps: and (3) fine-tuning the parameters of the trained neural network model by adopting a small amount of known real geological radar data of the tunnel lining region for the trained neural network model.
8. An intelligent inversion system for tunnel lining defect change features, which is characterized in that the system is used for realizing the intelligent inversion method steps of the tunnel lining defect change features according to any one of claims 1-7.
9. A computer readable storage medium storing a computer program, characterized in that the computer program, when executed by a processor, causes the processor to perform the tunnel lining lesion change feature intelligent inversion method steps according to any one of claims 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211246860.4A CN115630492B (en) | 2022-10-12 | 2022-10-12 | Tunnel lining defect change characteristic intelligent inversion method, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211246860.4A CN115630492B (en) | 2022-10-12 | 2022-10-12 | Tunnel lining defect change characteristic intelligent inversion method, system and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115630492A CN115630492A (en) | 2023-01-20 |
| CN115630492B true CN115630492B (en) | 2023-12-19 |
Family
ID=84905323
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211246860.4A Active CN115630492B (en) | 2022-10-12 | 2022-10-12 | Tunnel lining defect change characteristic intelligent inversion method, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115630492B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116434057B (en) * | 2023-03-01 | 2024-10-01 | 中公高科养护科技股份有限公司 | Road disease identification method, medium and system |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107132539A (en) * | 2017-05-03 | 2017-09-05 | 中国地质科学院探矿工艺研究所 | Landslide early-stage identification method of time sequence InSAR (interferometric synthetic Aperture Radar) based on small baseline set |
| CN110361453A (en) * | 2019-07-24 | 2019-10-22 | 四川幔壳科技有限公司 | Array dystopy based on non-destructive testing receives ultrasound synthesis aperture focus method |
| CN112213724A (en) * | 2020-09-16 | 2021-01-12 | 深圳航天科技创新研究院 | Backward projection imaging method and system based on ground penetrating radar data and storage medium |
| WO2021068848A1 (en) * | 2019-10-09 | 2021-04-15 | 山东大学 | Tunnel structure disease multi-scale measurement and intelligent diagnosis system and method |
| CN113552571A (en) * | 2020-09-25 | 2021-10-26 | 哈尔滨工业大学（威海） | PSM (pulse position modulation) algorithm-based underwater laser induced acoustic SAFT (safety and safety transform) imaging method |
| CN114353876A (en) * | 2022-01-07 | 2022-04-15 | 兰州大学 | Loess highway slope health monitoring method |
| CN114548278A (en) * | 2022-02-22 | 2022-05-27 | 西安建筑科技大学 | In-service tunnel lining structure defect identification method and system based on deep learning |
| CN114707532A (en) * | 2022-01-11 | 2022-07-05 | 中铁隧道局集团有限公司 | Ground penetrating radar tunnel disease target detection method based on improved Cascade R-CNN |
| WO2022147969A1 (en) * | 2021-01-08 | 2022-07-14 | 中国民航大学 | Airport pavement underground structure disease automatic detection method based on deep learning |
| CN114897804A (en) * | 2022-04-26 | 2022-08-12 | 云南航天工程物探检测股份有限公司 | Ground penetrating radar tunnel lining quality detection method based on self-supervision learning |
| CN114942436A (en) * | 2022-05-31 | 2022-08-26 | 西安建筑科技大学 | Tunnel lining disease identification method and system based on geological radar and deep learning |
| CN115009310A (en) * | 2022-06-24 | 2022-09-06 | 中国铁道科学研究院集团有限公司 | Self-walking railway tunnel evaluation and maintenance platform and method |
-
2022
- 2022-10-12 CN CN202211246860.4A patent/CN115630492B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107132539A (en) * | 2017-05-03 | 2017-09-05 | 中国地质科学院探矿工艺研究所 | Landslide early-stage identification method of time sequence InSAR (interferometric synthetic Aperture Radar) based on small baseline set |
| CN110361453A (en) * | 2019-07-24 | 2019-10-22 | 四川幔壳科技有限公司 | Array dystopy based on non-destructive testing receives ultrasound synthesis aperture focus method |
| WO2021068848A1 (en) * | 2019-10-09 | 2021-04-15 | 山东大学 | Tunnel structure disease multi-scale measurement and intelligent diagnosis system and method |
| CN112213724A (en) * | 2020-09-16 | 2021-01-12 | 深圳航天科技创新研究院 | Backward projection imaging method and system based on ground penetrating radar data and storage medium |
| CN113552571A (en) * | 2020-09-25 | 2021-10-26 | 哈尔滨工业大学（威海） | PSM (pulse position modulation) algorithm-based underwater laser induced acoustic SAFT (safety and safety transform) imaging method |
| WO2022147969A1 (en) * | 2021-01-08 | 2022-07-14 | 中国民航大学 | Airport pavement underground structure disease automatic detection method based on deep learning |
| CN114353876A (en) * | 2022-01-07 | 2022-04-15 | 兰州大学 | Loess highway slope health monitoring method |
| CN114707532A (en) * | 2022-01-11 | 2022-07-05 | 中铁隧道局集团有限公司 | Ground penetrating radar tunnel disease target detection method based on improved Cascade R-CNN |
| CN114548278A (en) * | 2022-02-22 | 2022-05-27 | 西安建筑科技大学 | In-service tunnel lining structure defect identification method and system based on deep learning |
| CN114897804A (en) * | 2022-04-26 | 2022-08-12 | 云南航天工程物探检测股份有限公司 | Ground penetrating radar tunnel lining quality detection method based on self-supervision learning |
| CN114942436A (en) * | 2022-05-31 | 2022-08-26 | 西安建筑科技大学 | Tunnel lining disease identification method and system based on geological radar and deep learning |
| CN115009310A (en) * | 2022-06-24 | 2022-09-06 | 中国铁道科学研究院集团有限公司 | Self-walking railway tunnel evaluation and maintenance platform and method |
Non-Patent Citations (3)
| Title |
|---|
| InSAR技术对采空区铁路隧道病害分析研究;胡永占;;铁道工程学报(02);全文 * |
| 地质雷达检测隧道衬砌常见病害模型实验研究;耿大新;李洪梅;郭俊;梁国卿;;华东交通大学学报(06);全文 * |
| 铁路运营隧道检测技术综述;王石磊 等;交通运输工程学报;第20卷(第05期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115630492A (en) | 2023-01-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Merazi-Meksen et al. | Mathematical morphology for TOFD image analysis and automatic crack detection | |
| CN109685011A (en) | A kind of underground utilities detection recognition method based on deep learning | |
| Wang et al. | GPRI2Net: A deep-neural-network-based ground penetrating radar data inversion and object identification framework for consecutive and long survey lines | |
| Xiang et al. | Robust cascaded frequency filters to recognize rebar in GPR data with complex signal interference | |
| Persico et al. | Microwave tomography for processing of GPR data at Ballachulish | |
| CN115630492B (en) | Tunnel lining defect change characteristic intelligent inversion method, system and storage medium | |
| Aboudourib et al. | A processing framework for tree-root reconstruction using ground-penetrating radar under heterogeneous soil conditions | |
| Zhang et al. | Ground-penetrating radar railroad ballast inspection with an unsupervised algorithm to boost the region of interest detection efficiency | |
| Sun et al. | Estimating parameters of the tree root in heterogeneous soil environments via mask-guided multi-polarimetric integration neural network | |
| Dai et al. | 3DInvNet: A deep learning-based 3D ground-penetrating radar data inversion | |
| Ayala-Cabrera et al. | Towards the visualization of water supply system components with GPR images | |
| Hu et al. | Defects identification and location of underground space for ground penetrating radar based on deep learning | |
| Li et al. | 3D ground penetrating radar road underground target identification algorithm using time-frequency statistical features of data | |
| CN107358162B (en) | Polarization SAR image change detection method based on depth stack network | |
| Huang et al. | Automated detection of through-cracks in pavement using three-instantaneous attributes fusion and Swin Transformer network | |
| Zhang et al. | A deep learning method for rough surface clutter reduction in GPR image | |
| Qian et al. | Deep Learning-Augmented Stand-off Radar Scheme for Rapidly Detecting Tree Defects | |
| Xiong et al. | Automatic defect detection in operational high-speed railway tunnels guided by train-mounted ground penetrating radar data | |
| Meschino et al. | A practical guide on using SPOT-GPR, a freeware tool implementing a SAP-DoA technique | |
| Wang et al. | Seafloor classification based on deep-sea multibeam data—Application to the southwest Indian Ridge at 50.47° E | |
| Yang et al. | An adaptive clutter-immune method for pipeline detection with GPR | |
| CN114063063A (en) | Geological disaster monitoring method based on synthetic aperture radar and point-like sensor | |
| Ge et al. | Wavelet-GAN: A GPR Noise and Clutter Removal Method Based on Small Real Datasets | |
| Savita et al. | GPR preprocessing methods to identify possible cavities in lateritic soil | |
| Li et al. | Road sub-surface defect detection based on gprMax forward simulation-sample generation and Swin Transformer-YOLOX |
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 |