CN115236756A - Data acquisition and processing system is patrolled and examined to dykes and dams structure dangerous case hidden danger - Google Patents
Data acquisition and processing system is patrolled and examined to dykes and dams structure dangerous case hidden danger Download PDFInfo
- Publication number
- CN115236756A CN115236756A CN202210993899.6A CN202210993899A CN115236756A CN 115236756 A CN115236756 A CN 115236756A CN 202210993899 A CN202210993899 A CN 202210993899A CN 115236756 A CN115236756 A CN 115236756A
- Authority
- CN
- China
- Prior art keywords
- data
- dam
- hidden danger
- image
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
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
- G01V3/38—Processing data, e.g. for analysis, for interpretation, for correction
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
-
- 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
- G01V3/08—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation operating with magnetic or electric fields produced or modified by objects or geological structures or by detecting devices
-
- 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
- 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
- G01V3/40—Electric or magnetic prospecting or detecting; Measuring magnetic field characteristics of the earth, e.g. declination, deviation specially adapted for measuring magnetic field characteristics of the earth
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
- Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
- Y02A90/30—Assessment of water resources
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Remote Sensing (AREA)
- Geophysics (AREA)
- General Life Sciences & Earth Sciences (AREA)
- Geology (AREA)
- Environmental & Geological Engineering (AREA)
- Theoretical Computer Science (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Electromagnetism (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention relates to a dam structure dangerous case and hidden danger inspection data acquisition and processing system, which utilizes a laser scanner to scan each slope of a dam and carries out three-dimensional measurement on the whole terrain of the dam; respectively acquiring shallow electromagnetic wave amplitude value data and deep resistivity data of a dam and resistivity data of the whole dam structure by adopting geological radar equipment, transient electromagnetic detection equipment and geomagnetism resistance detection equipment, and fusing and reconstructing the acquired data to obtain a dam shallow electromagnetic wave amplitude value distribution model diagram, a dam deep resistivity distribution model diagram and a resistivity distribution model diagram of the whole dam; and inputting the obtained model diagram, carrying out hidden danger identification on the image by using the trained DD-CNN model, and then obtaining a final hidden danger position distribution diagram through image processing. The method of the invention combines the vehicle-mounted platform and multi-source data to realize the comprehensive detection and diagnosis of various dangerous case hidden dangers on the surface of the dam, the shallow layer dam body, the deep part dam foundation and the dam structure in different depth directions.
Description
Technical Field
The invention belongs to the field of engineering safety monitoring, and particularly relates to a dam structure dangerous case and hidden danger inspection data acquisition and processing system.
Background
The potential safety hazard can be found and eliminated in time in the dam routing inspection, and the method is very important for the safety maintenance of dam engineering. For detecting hidden troubles of the dam, a geophysical prospecting detection method is commonly adopted at present to research the characteristic change of the physical aspect of the dam and judge the buried depth, scale and form of the hidden troubles of the dam. With the development of the technology, methods such as ground penetrating radar and resistivity method are used for detecting hidden danger of the dam, and the position of a hidden danger point in the dam is deduced by detecting a local abnormal area of a two-dimensional profile of data. But the interior distance of the dam is long, the distribution range of hidden dangers is wide, and the types of the hidden dangers are complex and changeable. In order to improve the flood control capability of the dam, the accurate positioning and judgment of various hidden dangers, particularly the detection and identification efficiency of the hidden dangers of the dam in the flood season and the flood season, need to be improved urgently, and the time is consumed for emergency rescue.
The dam structure is complex, the hidden danger types are complex and changeable, the hidden danger is difficult to accurately distinguish by only depending on a single method, and the distribution of the hidden danger abnormal bodies is difficult to completely determine by only depending on the detection result at a certain moment. The interior distance of the dam is long, the terrain is complex, manual inspection and detection of fixed sections are adopted, the efficiency is low, and various methods cannot be used for simultaneously and rapidly inspecting the continuity of the whole dam section. Especially, the hidden danger of the dam in the flood season is dynamically developed, the hidden danger cannot be comprehensively detected by one-time detection, and multiple times of detection and confirmation are required. Especially, under the continuous high water level of flood season, all face the safety risk of breaking a levee at any time, the mode validity of traditional artifical inspection is low, wastes time and energy. Therefore, there is a need for rapid and continuous detection of multiple detection methods by means of a driving platform.
The continuous comprehensive detection of various methods has huge real-time transmission data volume, and different methods have different data information and need different software for interpretation. How to select the required data, utilize the vehicle platform to obtain the required data information steadily, and process in order to obtain the required hidden danger data, it is the problem that needs to be solved to detect the hidden danger of dykes and dams at present.
Disclosure of Invention
The invention aims to overcome the problems in the prior art and provides a data acquisition and processing system for routing inspection of dangerous case and hidden danger of a dam structure.
In order to achieve the technical purpose, the invention adopts the following technical scheme:
a dam structure dangerous case and hidden danger inspection data acquisition and processing system comprises a dam inspection vehicle-mounted platform, wherein the vehicle-mounted platform can move along a dam, and detection equipment and a data processing device are arranged on the vehicle-mounted platform;
wherein the detection equipment comprises a laser scanner, geological radar equipment, magnetotelluric detection equipment and transient electromagnetic detection equipment;
the laser scanner scans each slope of the dam to perform three-dimensional measurement of the whole terrain of the dam;
the geological radar equipment, the transient electromagnetic detection equipment and the geomagnetic resistance detection equipment respectively acquire shallow resistivity data and deep resistivity data of the dam and resistivity data of the whole dam structure;
the data processing device processes the data collected by the detection equipment and outputs a hidden danger analysis result, and the hidden danger analysis result comprises the following steps:
establishing a dam model by using three-dimensional measurement data obtained by laser scanning;
respectively importing the shallow electromagnetic wave amplitude value data, the deep resistivity data and the whole dam resistivity data by taking the dam model as a substrate, and reconstructing the obtained dam shallow electromagnetic wave amplitude distribution model diagram, the dam deep resistivity distribution model diagram and the resistivity distribution model diagram of the whole dam;
and respectively taking the shallow electromagnetic wave amplitude distribution model diagram, the deep electrical resistivity distribution model diagram and the whole dam electrical resistivity data distribution model diagram as input, carrying out hidden danger judgment on the images by using a trained DD-CNN model, merging the shallow electromagnetic wave amplitude hidden danger identification result diagram and the deep electrical resistivity hidden danger identification result diagram by using image processing to obtain a first output, and then taking an intersection of the first output and the dam electrical resistivity hidden danger identification result diagram to obtain a final hidden danger position distribution diagram.
In a preferred embodiment, the laser scanner is fixed on the top of the vehicle-mounted platform;
the transient electromagnetic detection equipment is in a dragging type, is fixed on the vehicle-mounted platform through a telescopic rod, and is in contact with a dam to acquire data during working;
the earth magneto-resistance detection equipment and the geological radar equipment are fixed at the tail part of the vehicle-mounted platform.
In a preferred embodiment, the geological radar apparatus employs a multi-channel operation sequence, and the multi-channel antennas with different central frequencies are arranged side by side to simultaneously acquire data.
As a preferred embodiment, the antenna of the geological radar apparatus is a Vivaldi antenna;
the Vivaldi antenna is characterized in that the two slot lines are symmetrical, each slot line is composed of two sections of folding lines and a curve connected with the concave part between the two sections of folding lines, an included angle between the first section of folding line close to the feed structure and the symmetry axis is an acute angle, and the second section of folding line is parallel to the symmetry axis. The geological radar preferably adopts a multi-channel working time sequence, so that the mutual radiation interference among radar antennas of all channels is avoided, the double-fold line type conical groove Vivaldi antenna is designed, the radiation of side surface waves is inhibited, the interference is reduced, the energy radiation of the antenna is concentrated, the detection depth is increased, the stable gain performance is realized in a working frequency band, the beam characteristic is excellent, and the beam width is 180 degrees.
As a preferred embodiment, the Vivaldi antenna is encapsulated with an epoxy insulating board and the slot is bonded with epoxy. Due to the fact that the weather and the terrain are complicated in the rainy season, rainwater is prevented from entering the interior of the antenna and being damaged by collision with the antenna, the antenna external packaging device is made of a compact epoxy insulation board, fixing of bolts is omitted, high-strength epoxy resin is used for bonding of a gap structure, the stability, the reliability, the waterproof function and the damp-proof function of the antenna packaging structure are guaranteed, and measurement under the harsh environment is achieved. The multi-channel geological radar is mainly used for detecting hidden dangers of a plurality of sections of the dike body.
As a preferred embodiment, the transient electromagnetic detection device is a towed small loop transient electromagnetic detection device, and an integrated transient electromagnetic structure with cross-loop decoupling is adopted.
As a preferred embodiment, the transient electromagnetic detection device is packaged by a carbon fiber material, and the inter-insulation structure of the small loop structure is a thallium-barium-calcium-copper-oxygen ceramic material. The carbon fiber packaging material can ensure the firmness, shock resistance and portability of the coil. The thallium-barium-calcium-copper-oxygen ceramic material is selected to manufacture the isolation structure, so that the mutual interference of magnetic fields between small wire return structures can be weakened, and the isolation of strong and weak electricity can be realized. The parallel transient electromagnetic structure is mainly used for detecting a plurality of profile hidden dangers of an embankment base.
As a preferred embodiment, the earth magneto-resistance detection device realizes active and passive resistivity detection in two modes, namely passively measuring an earth electromagnetic field or actively measuring the electromagnetic field by configuring a high-power transmitter to transmit a specific signal source; in the continuous detection process of the vehicle-mounted platform, an active mode is adopted for continuously acquiring resistivity data of the whole dam structure; when the local area is detected in a fine mode, a passive mode can be adopted for continuously acquiring the resistivity data of the whole dam structure.
As a preferred embodiment, the DD-CNN model is trained in the following manner:
establishing a model training database which comprises an engineering actual measurement data pair and a laboratory simulation data pair, wherein the data pair refers to a group of data formed by actual distribution data of hidden dangers and corresponding resistivity and electromagnetic wave amplitude distribution data;
dividing data in a model training database into training set data and verification set data, adding labels to hidden danger parts in hidden danger actual distribution data, inputting the training set data into a DD-CNN model for model training, performing precision evaluation on a model training result by using the verification set data, and using an obtained model with optimal precision for data processing of data acquired by a vehicle-mounted platform. Because the engineering actual measurement data volume is limited, laboratory simulation data can be used for increasing the data volume and improving the model training precision.
As a preferred embodiment, the DD-CNN model consists of an input layer, 2 convolutional layers, 8 Fire modules, 3 maximum pooling layers, 1 average pooling layer, and an output layer;
8 stacked Fire modules and 3 maximum pooling layers are arranged between the 2 convolutional layers;
the average pooling layer connects the 2 nd convolutional layer and the output layer.
In a preferred embodiment, the activation function used by the 2 convolutional layers and the 8 Fire modules is Relu; the output layer is a Softmax function; the loss function for image training is MSE.
As a preferred embodiment, the model processing procedure is:
preprocessing a resistivity and electromagnetic wave amplitude distribution data image into an image with preset pixels, then sending the image into a model in batches for forward calculation, entering a first convolution layer, extracting 64-channel characteristics, then resampling by adopting a first maximum pooling layer, and reducing the image by half; then after 256 channel characteristics are extracted through 3 stacked Fire modules, resampling is carried out by adopting a second largest pooling layer, and the size of the image is reduced by half again; after 512-channel characteristics of the output image are extracted through 4 stacked Fire modules, resampling is carried out by adopting a third largest pooling layer, and the size of the image is reduced by half again; entering a Fire module and a second convolution layer, and outputting an image with 2-channel characteristics through 1 multiplied by 1/1 convolution kernel operation; then, outputting a hidden danger characteristic image through an average pooling layer;
comparing the hidden danger characteristic image output by the network model with the label of the real hidden danger of the corresponding data pair, and calculating the loss by using a loss function; and performing matching iterative computation on the output hidden danger image and the image in the database to minimize a loss function and realize the identification of the hidden danger in the image.
The invention has the following beneficial effects:
1. aiming at the difficulty of 'inaccurate detection and incomplete detection' of dam diseases by the current geophysical prospecting method; various detection devices are carried on the designed vehicle-mounted platform, and detection advantages of various methods are complementary and do not interfere with each other. A laser radar, a ground penetrating radar, a Magnetotelluric (MTR) and transient electromagnetic comprehensive data acquisition system is designed, and a DD-CNN (Depthwise discrete capacitive Neural Network) model is used for data processing, so that comprehensive detection and diagnosis of various dangerous case hidden dangers on the surface of the dam, the shallow dam body, the deep dam foundation and the dam structure in different depth directions are realized.
2. The method has the advantages that the hidden dangers such as various cavities, ant holes, piping channels and the like of the dam structure can be rapidly detected, the detection speed on the dam is not less than 10 kilometers per hour, and the inspection efficiency and accuracy of the dam in the flood season are improved.
3. Through comprehensive complementary detection of various detection methods, the size of the hidden danger can be distinguished to be not more than 1/10 of the detection depth of the dike, the detection depth is not less than 10 meters, and deformation with the slope surface range not more than 1 meter multiplied by 2 meters can be identified; under the detailed inspection mode (when carrying out encryption point detection), the size of the distinguishable hidden danger is not more than 1/20 of the detection depth of the dike, the detection speed is not less than 1 hidden danger point/hour, the detection depth is not shallower than 30 meters, and the detection precision and the inspection efficiency of the hidden danger of the dam structure are improved.
4. Through information transmission and artificial intelligence interpretation of the big data platform, automatic analysis, identification and early warning of hidden dangers are achieved, and the dangerous case reporting function of the dam management system is improved.
Drawings
Fig. 1 is a diagram of an apparatus composition of a dam inspection system, in which: 2-1 is a vehicle-mounted platform; 2-2 is a dragging type transient electromagnetic detection device; 2-3 is a laser scanner; 2-4 is an earth magneto-resistance detection device; 2-5 are multi-channel geological radar equipment; 2-6 are data processing terminals; 2-7 are communication devices.
Fig. 2 is a schematic diagram of a double-folded tapered slot Vivaldi antenna and a radar antenna array, wherein the left diagram is a schematic diagram of a Vivaldi antenna structure, and the right diagram is a radar antenna array formed by packaging and arranging antennas of the left diagram in parallel.
Fig. 3 is a schematic diagram of a side-by-side transient electromagnetic structure.
FIG. 4 is a diagram of a deep learning DD-CNN model network composition for hidden danger identification.
FIG. 5 shows the Fire module in the DD-CNN model.
Fig. 6 is a flow chart of hidden danger intelligent identification.
Fig. 7 is a flow chart of intelligent hidden danger distinguishing and early warning.
Fig. 8 is an exemplary diagram of a hidden danger intelligent identification result in a detection result.
Fig. 9 is a diagram illustrating a result of a hidden trouble defect.
Fig. 10 is a schematic diagram of the final determination of the hidden trouble.
Detailed Description
Example 1
The system for acquiring and processing the inspection data of the dangerous case and hidden danger of the dam structure shown in figures 1-3 is used for reconstructing a vehicle-mounted platform 2-1, placing communication equipment 2-7 and a data processing terminal 2-6 (a computer) in a vehicle body, and carrying various detection equipment outside the vehicle body.
The towed transient electromagnetic detection device 2-2 is placed on a fixed platform at the front end of the vehicle-mounted platform 2-1, and the towed transient electromagnetic detection device 2-2 is connected and fixed with a dam through the front section of a plastic telescopic rod to be in contact with the dam for data acquisition.
A laser scanner 2-3 is arranged on the top of the vehicle-mounted platform 2-1, so that laser three-dimensional measurement of dam landforms and panoramic display of real-time landform conditions are realized. As shown in fig. 3, the transient electromagnetic detection device is a towed small loop transient electromagnetic detection device, and adopts an integrated transient electromagnetic structure with cross-loop decoupling. The transient electromagnetic detection equipment is packaged by adopting a carbon fiber material, and the spacing structure of the small loop structure is a thallium-barium-calcium-copper-oxygen ceramic material.
The method comprises the steps that a large earth magneto-resistance detection device 2-4 and a multi-channel geological radar device 2-5 are carried at the tail part of a vehicle-mounted platform 2-1 through a hydraulic telescopic platform (a hydraulic tail plate), wherein the geological radar device 2-5 adopts multi-channel antennas with different central frequencies to collect data side by side simultaneously. As shown in fig. 2, the antenna of the geological radar apparatus is a Vivaldi antenna; the Vivaldi antenna has two symmetrical slot lines, each slot line consists of two sections of folding lines and a curve connected with the folding line and recessed inwards, the first section of folding line close to the feed structure forms an acute angle with the symmetry axis, and the second section of folding line is parallel to the symmetry axis. The Vivaldi antenna is encapsulated using an epoxy insulation board and the slot is bonded using epoxy. The earth magneto-resistance detection equipment 2-4 can realize active and passive resistivity detection in two modes, namely passive measurement of an earth electromagnetic field can be realized, and active measurement of the electromagnetic field can also be realized by configuring a high-power transmitter to transmit a specific signal source; in the vehicle-mounted continuous detection process, an active mode is adopted for continuously acquiring the resistivity data of the dam structure; when the local area is detected in a fine mode, a passive mode can be adopted for continuously acquiring the resistivity data of the whole dam structure.
Inside a vehicle body system of the vehicle-mounted platform 2-1, communication equipment 2-7 and a data processing terminal 2-6 are placed, wherein a data transmission system of the communication equipment 2-7 integrates different types of data transmission and storage, including terrain positioning information in laser moving three-dimensional measurement, electromagnetic data of a ground penetrating radar, apparent resistivity of transient electromagnetism and resistivity data of earth magnetoresistance.
In the detection process, various data are stored at regular time through time sequence control, the stored data are directionally transmitted to a data processing terminal 2-6 by utilizing communication equipment 2-7, the various data are interpreted through corresponding software, the interpreted radar electromagnetic wave, transient electromagnetic apparent resistivity and earth magneto-resistance resistivity image data are respectively substituted into an artificial intelligent model to learn training, the identification result of the hidden danger is output, and finally the category and the size information of the hidden danger are given comprehensively. Meanwhile, the interpreted laser mobile measurement three-dimensional terrain image is led into a hidden danger visualization platform to serve as an embankment structure model, then hidden dangers are led into the model, real-time visualization and positioning of the hidden dangers are conducted, and information is pushed to an intelligent management platform to send early warning information.
The embodiment uses the DD-CNN artificial intelligence model to process the image data, improves the prior model, increases the convolution layer number and the change of the final output parameter, and the input parameters are all the pixel characteristic values of the image.
The designed network structure of the artificial intelligence model DD-CNN is shown in FIG. 4, and comprises 2 conventional convolutional layers, 3 maximum pooling layers, 1 average pooling layer and 8 Fire modules.
Where Fire models, as shown in FIG. 5, each Fire module involves two further convolutions: compression and expansion. The first layer of the compressed convolution contains 1 × 1 filters, while the second layer of the extended convolution contains two filters, 1 × 1 and 3 × 3.
The convolution layer and the Fire module adopt ReLU activation function, and the output layer is Softmax function.
Establishing a model training database which comprises engineering actual measurement data pairs and laboratory simulation data pairs, wherein the data pairs refer to a group of data formed by actual distribution data of hidden dangers and corresponding resistivity and electromagnetic wave amplitude distribution data;
dividing data in a model training database into training set data and verification set data, adding labels to hidden danger parts in hidden danger actual distribution data, inputting the training set data into a DD-CNN model for model training, performing precision evaluation on a model training result by using the verification set data, and using an obtained model with optimal precision for data processing of data acquired by a vehicle-mounted platform.
The process of model training is as follows: preprocessing the model image into an image of pixels 224 × 224; then sending the image data as an input layer into a model in batches for forward calculation, wherein the layer 1 processed by the model is a convolutional layer (figure 4), and extracting the channel characteristics of an image 64; then, resampling is carried out by utilizing the maximum pooling layer 2, the image size reduced by half is used as an input layer and enters a Fire module from a 3 rd layer to a 5 th layer for calculation, and the number of characteristic channels of the image is increased to 256; resampling by adopting a maximum pooling layer 6, reducing the image size by half to 27 multiplied by 27 to be used as an input layer, entering a Fire module of 7 layers to 10 layers for calculation, and extracting 512-channel characteristics of the image; the output image is resampled again by adopting the maximum pooling layer 11, and the image size is reduced to 13 multiplied by 13; entering the calculation of a 12 th layer of Fire module and a 13 th layer of convolution layer, and outputting an image with 2-channel characteristics and 13 x 13 size through 1 x 1/1 convolution kernel operation; then, outputting a hidden danger characteristic image with the size of 13 multiplied by 13 finally through an average pooling layer 14;
comparing the hidden danger characteristic image output by the network model with the label of the real hidden danger of the corresponding data pair, and calculating the loss by using a loss function; the loss function of the image is MSE:
wherein, DM pr For predicting images, DM re For actual images, DM pr,i For predicting the ith pixel value in the image, DM re,i The ith pixel value in the real image is shown, and m is the total number of pixels in the image.
And performing matching iterative calculation on the output hidden danger image and the image in the database to minimize a loss function, and identifying the hidden dangers in the dam shallow electromagnetic wave amplitude distribution model image, the dam deep resistivity distribution model image and the dam resistivity distribution model image.
The identification of the hidden danger is shown in fig. 6, inputting a picture sample, and obtaining a label (ground route-GT) of the sample; substituting the sample into a model DD-CNN for calculation, and extracting a characteristic graph F of an L layer of the learning model N (ii) a Filling F N And the size of the sample GT is 256 × 256 pixels;calculating F N Average activation map F of Mai (ii) a For each feature map F i Binaryzation;
calculating GT and F after binarization of each feature map bin(i) Has a Hamming distance HD i Co-production of TS x F N Hamming distance:
HD i ＝|F bin(i) -GT|
calculating the sum of all Hamming distances, and finding the minimum Hamming distance according to a threshold value T; selecting a proper feature map F according to the Hamming distance of the girth i Outputting a hidden danger characteristic diagram and providing a prediction result of hidden danger of the picture sample;
the decision of the hidden danger information is shown in fig. 7, and the detection result acquired and processed in real time is brought into a deep learning model DD-CNN for calculation; if the prediction result has no hidden danger, then, no warning is given; if the prediction result image contains hidden dangers, extracting a hidden danger characteristic diagram, and calculating an average activation value diagram F of the characteristic diagram Ma Passing through a threshold T to F Ma Binarization is applied as follows:
from F Ma The hidden danger area is obtained in the image, information (position and outline) of the hidden danger is extracted, warning information of different levels is sent out according to the hidden danger information and the size of the hidden danger area, and a hidden danger image and information of the hidden danger are output.
Example 2
The embodiment provides a specific application case.
When the vehicle-mounted equipment runs on the dam, the laser scanner on the top of the vehicle carries out three-dimensional measurement of the whole terrain of the dam, the three-dimensional measurement is transmitted to the data processing terminal by the communication device, and a real-time three-dimensional model of the dam is established in the data processing terminal through data processing.
Deep resistivity data acquired by the transient electromagnetic detection equipment, the whole dam resistivity data acquired by the earth magneto-resistance detection equipment and shallow electromagnetic wave amplitude data acquired by the geological radar equipment are respectively transmitted to a data processing terminal through communication equipment for interpretation. And for various types of signal data, the data processing terminal stores, interprets and displays the signal data.
As shown in fig. 6, the result graph of the preprocessing is input, and the features of the image are transferred and extracted through the calculation of the DD-CNN model. And then predicting the category of the hidden danger.
As shown in fig. 8, the hidden danger in the interpretation result can be reliably characterized by its color, and the defect is a dark region.
As shown in fig. 7, hidden dangers in the interpretation result are successfully predicted through the DD-CNN model, if the interpretation result prediction result contains hidden dangers, a feature map of the hidden dangers is extracted from a feature map set of the deep learning model, then binarization processing is performed on the feature map of the hidden dangers to obtain position and size information of the hidden dangers, and finally the feature map of the hidden dangers is output. As shown in fig. 9, a result graph of defect risk.
And (3) calculating by using the DD-CNN model, outputting and obtaining a shallow electromagnetic wave amplitude data hidden danger identification result graph, a deep resistivity data hidden danger identification result graph and a whole dam resistivity data hidden danger identification result graph, and performing merging and intersection operation on hidden danger position information in the graphs to determine final position and size information of the hidden danger as shown in FIG. 10.
The identification of hidden dangers in the interpretation result is realized, and detailed information parameters in the deep learning model DD-CNN network structure are shown in the table 1.
TABLE 1
| Sequence of steps | Types of | Convolution kernel/step | Activating a function | Output size |
| Number 1 | Input layer | Long and long | 224×224×3 | |
| 2 | Convolutional layer 1 | 7×7/2(×64) | Relu | 111×111×64 |
| 3 | Maximum pooling layer 2 | 3×3/2 | 55×55×64 | |
| 4 | Fire3 | Relu | 55×55×128 | |
| 5 | Fire4 | Relu | 55×55×128 | |
| 6 | Fire5 | Relu | 55×55×256 | |
| 7 | Maximum pooling layer 6 | 3×3/2 | 27×27×256 | |
| 8 | Fire7 | Relu | 27×27×256 | |
| 9 | Fire8 | Relu | 27×27×384 | |
| 10 | Fire9 | Relu | 27×27×384 | |
| 11 | Fire10 | Relu | 27×27×512 | |
| 12 | Maximum pooling layer | 3×3/2 | 13×13×512 | |
| 13 | Fire12 | Relu | 13×13×512 | |
| 14 | Convolutional layer 13 | 1×1/1(×2) | Relu | 13×13×2 |
| 15 | Average pooling layer | 13×13/1 | 13×13×1 | |
| 16 | Output layer | Softmax | 13×13×1 |
Claims (10)
1. A dam structure dangerous case and hidden danger inspection data acquisition and processing system comprises a dam inspection vehicle-mounted platform, wherein the vehicle-mounted platform can move along a dam, and detection equipment and a data processing device are arranged on the vehicle-mounted platform;
the device is characterized in that the detection equipment comprises a laser scanner, geological radar equipment, magnetotelluric detection equipment and transient electromagnetic detection equipment;
the laser scanner scans each slope of the dam to perform three-dimensional measurement of the whole terrain of the dam;
the geological radar equipment, the transient electromagnetic detection equipment and the earth magneto-resistance detection equipment respectively acquire shallow electromagnetic wave amplitude value data and deep resistivity data of the dam and resistivity data of the whole dam structure;
the data processing device processes the data collected by the detection equipment and outputs a hidden danger analysis result, and the hidden danger analysis result comprises the following steps:
establishing a dam model by using three-dimensional measurement data obtained by laser scanning;
respectively importing the shallow electromagnetic wave amplitude value data, the deep resistivity data and the resistivity data of the whole dam structure by taking the dam model as a substrate, and reconstructing to obtain a dam shallow electromagnetic wave amplitude distribution model diagram, a dam deep resistivity distribution model diagram and a resistivity distribution model diagram of the whole dam;
and respectively taking the shallow electromagnetic wave amplitude distribution model diagram, the deep resistivity distribution model diagram and the resistivity distribution model diagram of the whole dam as input, carrying out hidden danger identification on the images by using a trained DD-CNN model, merging the shallow electromagnetic wave amplitude hidden danger identification result diagram and the deep resistivity hidden danger identification result diagram of the dam by adopting an image processing technology to obtain a first output, and taking an intersection of the first output and the resistivity hidden danger identification result diagram of the whole dam to obtain a final hidden danger position distribution diagram.
2. The system of claim 1, wherein the laser scanner is affixed to the top of the vehicle platform;
the transient electromagnetic detection equipment is in a dragging type, is fixed on the vehicle-mounted platform through a telescopic rod, and is in contact with a dam to acquire data during working;
the earth magneto-resistance detection equipment and the geological radar equipment are fixed at the tail of the vehicle-mounted platform.
3. The system of claim 1, wherein the geological radar apparatus employs a multi-channel operational sequence with multiple channels of different center frequency antennas simultaneously acquiring data side-by-side.
4. A system according to claim 1 or 3, wherein the antenna of the geological radar apparatus is a Vivaldi antenna;
the Vivaldi antenna is characterized in that the two slot lines are symmetrical, each slot line is composed of two sections of folding lines and a curve connected with the concave part between the two sections of folding lines, an included angle between the first section of folding line close to the feed structure and the symmetry axis is an acute angle, and the second section of folding line is parallel to the symmetry axis.
5. The system of claim 4, wherein the Vivaldi antenna is encapsulated with an epoxy insulating board and the slot is bonded with epoxy.
6. The system according to claim 1, wherein the transient electromagnetic detection device is a towed small loop transient electromagnetic detection device, and adopts a cross-loop decoupling integrated transient electromagnetic structure; the transient electromagnetic detection equipment is packaged by adopting a carbon fiber material, and the spacing structure of the small loop structure is a thallium-barium-calcium-copper-oxygen ceramic material.
7. The system of claim 1, wherein the DD-CNN model is trained by:
establishing a model training database which comprises engineering actual measurement data pairs and laboratory simulation data pairs, wherein the data pairs refer to a group of data formed by actual distribution data of hidden dangers and corresponding resistivity and electromagnetic wave amplitude distribution data;
dividing data in a model training database into training set data and verification set data, adding labels to hidden danger parts in hidden danger actual distribution data, inputting the training set data into a DD-CNN model for model training, performing precision evaluation on a model training result by using the verification set data, and using an obtained model with optimal precision for data processing of data acquired by a vehicle-mounted platform.
8. The system of claim 1, wherein the DD-CNN model consists of an input layer, 2 convolutional layers, 8 Fire modules, 3 max pooling layers, 1 average pooling layer, and an output layer;
8 stacked Fire modules and 3 maximum pooling layers are arranged between the 2 convolutional layers;
the average pooling layer connects the 2 nd convolutional layer and the output layer.
9. The system of claim 8, wherein the 2 convolutional layers, 8 Fire modules use an activation function of Relu; the output layer is a Softmax function; loss function of image training isMSE。
10. The system of claim 8, wherein the model processing is performed by:
preprocessing a resistivity and electromagnetic wave amplitude distribution data image into an image with preset pixels, then sending the image into a model in batches for forward calculation, entering a first convolution layer, extracting 64-channel characteristics, then resampling by adopting a first maximum pooling layer, and reducing the image by half; then after 256 channel characteristics are extracted through 3 stacked Fire modules, resampling is carried out by adopting a second largest pooling layer, and the size of the image is reduced by half again; after 512-channel characteristics of the output image are extracted through 4 stacked Fire modules, resampling is carried out by adopting a third largest pooling layer, and the size of the image is reduced by half again; entering a Fire module and a second convolution layer, and outputting an image with 2-channel characteristics through 1 multiplied by 1/1 convolution kernel operation; then, outputting a hidden danger characteristic image through an average pooling layer;
comparing the hidden danger characteristic image output by the network model with the label of the real hidden danger of the corresponding data pair, and calculating the loss by using a loss function; and performing matching iterative calculation on the output hidden danger image and the image in the database to minimize a loss function, thereby realizing the identification of the hidden danger in the image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210993899.6A CN115236756A (en) | 2022-08-18 | 2022-08-18 | Data acquisition and processing system is patrolled and examined to dykes and dams structure dangerous case hidden danger |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210993899.6A CN115236756A (en) | 2022-08-18 | 2022-08-18 | Data acquisition and processing system is patrolled and examined to dykes and dams structure dangerous case hidden danger |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115236756A true CN115236756A (en) | 2022-10-25 |
Family
ID=83678613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210993899.6A Pending CN115236756A (en) | 2022-08-18 | 2022-08-18 | Data acquisition and processing system is patrolled and examined to dykes and dams structure dangerous case hidden danger |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115236756A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115793093A (en) * | 2023-02-02 | 2023-03-14 | 水利部交通运输部国家能源局南京水利科学研究院 | Empty ground integrated equipment for diagnosing hidden danger of dam |
| CN115828054A (en) * | 2023-02-10 | 2023-03-21 | 成都信息工程大学 | Method for automatically identifying south branch groove by improving Laplace |
| CN117031551A (en) * | 2023-08-10 | 2023-11-10 | 水利部交通运输部国家能源局南京水利科学研究院 | Method and system for tour inspection of intelligent unmanned vehicle traversing station in dam engineering |
| CN117910517A (en) * | 2024-01-25 | 2024-04-19 | 河海大学 | Dyke empty hidden danger identification method and system based on physical information neural network |
-
2022
- 2022-08-18 CN CN202210993899.6A patent/CN115236756A/en active Pending
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115793093A (en) * | 2023-02-02 | 2023-03-14 | 水利部交通运输部国家能源局南京水利科学研究院 | Empty ground integrated equipment for diagnosing hidden danger of dam |
| CN115828054A (en) * | 2023-02-10 | 2023-03-21 | 成都信息工程大学 | Method for automatically identifying south branch groove by improving Laplace |
| CN117031551A (en) * | 2023-08-10 | 2023-11-10 | 水利部交通运输部国家能源局南京水利科学研究院 | Method and system for tour inspection of intelligent unmanned vehicle traversing station in dam engineering |
| CN117031551B (en) * | 2023-08-10 | 2024-01-30 | 水利部交通运输部国家能源局南京水利科学研究院 | Method and system for tour inspection of intelligent unmanned vehicle traversing station in dam engineering |
| CN117910517A (en) * | 2024-01-25 | 2024-04-19 | 河海大学 | Dyke empty hidden danger identification method and system based on physical information neural network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN115236756A (en) | Data acquisition and processing system is patrolled and examined to dykes and dams structure dangerous case hidden danger | |
| Kim et al. | A novel 3D GPR image arrangement for deep learning-based underground object classification | |
| CN102496232B (en) | Transmission facility monitoring method and system | |
| CN114548278A (en) | In-service tunnel lining structure defect identification method and system based on deep learning | |
| CN107179449B (en) | A kind of short-wave receiving antenna 3 D stereo directional diagram aerial mobile measuring system and method | |
| CN114266892A (en) | Pavement disease identification method and system for multi-source data deep learning | |
| CN106960548A (en) | A kind of slope geological monitoring and pre-alarming method and system | |
| CN108711172A (en) | Unmanned plane identification based on fine grit classification and localization method | |
| CN112327372B (en) | Comprehensive detection method for internal hidden danger of long-distance linear engineering canal dike | |
| CN114814956A (en) | Earth-rock dam deep-buried disease positioning method and device based on seepage and infiltration area characteristics | |
| CN104267440A (en) | Common middle point (CMP) detection method used for ground penetrating radar (GPR) | |
| CN206833669U (en) | A kind of slope geological monitoring and warning system | |
| CN113138364A (en) | Landslide safety monitoring device and positioning method based on underwater photography equipment | |
| CN116778329A (en) | Urban road underground shallow disease detection method, device, equipment and medium | |
| CN208334238U (en) | A kind of roadbed filling quality rapid determination system | |
| CN105445439A (en) | Outdoor automatic monitoring system applicable to natural salinized soil formation mechanism | |
| CN208254411U (en) | A kind of side slope comprehensive monitor system | |
| CN115508907A (en) | Vehicle-mounted embankment dangerous case hidden danger detection system and early warning method | |
| CN105717375B (en) | A kind of ocean surface layer electromagnetic wave energy distribution comprehensive test system and method | |
| KR102322701B1 (en) | On-site intelligent underground facility safety management device and method through ai robot module for underground facility measurement | |
| CN215643053U (en) | Landslide safety monitoring device based on underwater submerged camera equipment | |
| CN115442398A (en) | River and lake intelligent boundary pile based on base station-free differential GPS positioning technology and monitoring system | |
| CN115409691A (en) | Bimodal learning slope risk detection method integrating laser ranging and monitoring image | |
| CN210294524U (en) | Three-dimensional measuring device for transformer substation grounding grid | |
| CN111950909A (en) | Three-dimensional laser scanning karst water inrush disaster magnitude evaluation test method and device |
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 |