CN107358617B - Method for detecting target vertex in ground penetrating radar recording profile - Google Patents

Method for detecting target vertex in ground penetrating radar recording profile Download PDF

Info

Publication number
CN107358617B
CN107358617B CN201710667148.4A CN201710667148A CN107358617B CN 107358617 B CN107358617 B CN 107358617B CN 201710667148 A CN201710667148 A CN 201710667148A CN 107358617 B CN107358617 B CN 107358617B
Authority
CN
China
Prior art keywords
target
matching
buffer area
target vertex
edge
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
Application number
CN201710667148.4A
Other languages
Chinese (zh)
Other versions
CN107358617A (en
Inventor
雷文太
满敏
施荣华
左逸玮
彭楠
梁琼
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN201710667148.4A priority Critical patent/CN107358617B/en
Publication of CN107358617A publication Critical patent/CN107358617A/en
Application granted granted Critical
Publication of CN107358617B publication Critical patent/CN107358617B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/13—Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10032—Satellite or aerial image; Remote sensing
    • G06T2207/10044—Radar image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30181—Earth observation

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a method for detecting target vertexes in a ground penetrating radar recording profile, which comprises the steps of firstly calculating a one-dimensional energy curve of the preprocessed ground penetrating radar recording profile along a measuring line dimension, and determining a preset interval of the number of the target vertexes according to the number of local maximum points of the curve; then, performing edge detection on the preprocessed ground penetrating radar recording section in an automatic threshold matching mode to obtain a binary image and estimating vertexes in the binary image until the number of the target vertexes to be estimated is within a preset interval; then establishing a matching template based on the coordinate positions of the vertexes, performing template matching by using the template and a binary image obtained by edge detection, calculating matching similarity, discarding the vertexes with the matching similarity lower than the threshold value, and obtaining the matched vertexes; and filtering the matched vertexes based on a cluster analysis method, and removing false vertexes to obtain a detection result of the target vertex. Compared with the existing method, the method has higher detection precision when the signal-to-noise ratio is lower.

Description

Method for detecting target vertex in ground penetrating radar recording profile
Technical Field
The invention belongs to the technical field of ground penetrating radar detection and application, and particularly relates to a method for detecting a target vertex by using a ground penetrating radar image.
Background
Ground Penetrating Radar (GPR) is an effective shallow hidden target detection technology, and uses the reflection and scattering of electromagnetic waves generated at discontinuities of medium electromagnetic properties to realize imaging detection of targets in a non-metal coverage area. The amplitude and time delay of the GPR echo contain information such as the target position and electromagnetic scattering properties. The GPR performs spatial scanning along one-dimensional survey lines, transmitting electromagnetic waves to the subsurface region at each aperture point of the survey line, and receiving scattered echoes. One echo data is received at each aperture point, and the echo data received at a plurality of aperture points are arranged in columns to form a GPR recording profile. The processing of GPR recorded profiles to enable detection and parametric inversion of subsurface unknown regions is a non-destructive testing approach. The GPR can be used in media such as rocks, soil, ice, fresh water, public roads and various structures, can detect substances, material changes, gaps, cracks and the like under the ground, and is widely applied to the field of nondestructive detection of road engineering, constructional engineering, archaeology and the like.
In the GPR, electromagnetic scattering is generated when an electromagnetic wave meets a target with different electromagnetic properties from a background medium in the scanning process, and part of scattered waves are received by a GPR receiving antenna. The scattered echoes of the subsurface target form a time delay curve in the GPR recording profile as the antenna moves. The vertexes of the delay curves have a one-to-one correspondence with the real space positions of the targets. From the analysis of graphic images, the spatial distribution of the underground target can be obtained only by determining the vertex position of the delay curve in the GPR recording section. Therefore, the algorithm for detecting the position of the vertex in the GPR recorded profile becomes a key problem for positioning the target in GPR ground detection. A great deal of research has been done abroad on the problem of vertex detection in GPR recording profiles. Documents 1 "Liu, m.wang, and q.cai," The target detection for GPR images based on secure determination, "in proc.3rd int.congr.image Signal Process" (CISP), vol.6.yantai, China, oct.2010, pp.2876-2879 "propose methods for extracting an object in an image by curve fitting and edge detection of a curve in a GPR image. Document 2 "z.w.wang, m.zhou, g.g.slabaugh, j.zhai, and t.fang," Automatic detection of bridge deviation from group boundary radar images, "IEEE trans.automatic.sci.eng., vol.8, No.3, pp.633-640, jul.2011" proposes determining the positions of the vertices of a hyperbola in a GPR image by using a mathematical method of partial differential equations for the curve in the GPR image and template matching, thereby detecting the corrosion of the reinforced concrete deck. Document 3 "paneet Kaur; kristin j.dana; francisco a.romero; nenad Gucunski, "Automated GPRRebar Analysis for robust Bridge Deck Evaluation," IEEE Transactions on cybernetics. vol.46,2016, pp.2265-2276 "proposes a hyperbolic signature based on gradient characteristics of an image and a curve fitting manner, and detects the corrosion condition of the reinforcing steel bars of the Bridge Deck by a method of integrated machine learning classification. Document 4 "laurencefmertens; raffaele Persico; loredana Matera; s basic Lambot, "automatic detection of Reflection Hyperbolas in Complex GPR Images With No A priority knowledge on the Medium," IEEE Transactions on Geoscience and remotesensing. vol.54,2016, pp.580-596 "proposes a target detection method for generating a hyperbolic fitting function based on the edge points in the edge detection result of a GPR image. In the above method, the detection accuracy is high when the signal-to-noise ratio is high, but when the signal-to-noise ratio is low, the edge profile of the target hyperbolic curve in the GPR cross-sectional recorded image changes, and curve characteristics such as smoothness and symmetry are destroyed, which causes a miss detection or an error detection of the target point, and directly affects the detection accuracy of the target. Moreover, the edge detection method requires manual threshold adjustment, and is poor in real-time performance. Therefore, it is necessary to design a vertex detection method with high detection accuracy and capable of realizing automation in view of the problems in the prior art.
Disclosure of Invention
The technical problem to be solved by the present invention is to provide a method for detecting a target vertex in a GPR recording profile based on automatic threshold edge detection and template matching, which can achieve better detection performance and automatic vertex detection in the presence of noise, in view of the shortcomings of the prior art.
The technical scheme of the invention is as follows:
a method for detecting a target vertex in a ground penetrating radar recording profile comprises the following steps:
step 1: preprocessing an original ground penetrating radar recording section, calculating a one-dimensional energy curve of the preprocessed ground penetrating radar recording section along the direction of a measuring line, extracting the number of local maximum values of the energy curve, and setting a preset interval (value range) of the number of target vertexes of the ground penetrating radar recording section according to the number;
step 2: performing automatic threshold matching edge detection on the preprocessed ground penetrating radar recording section to obtain a binary image, and estimating a target vertex possibly existing according to the binary image;
and step 3: establishing a matching template according to the estimated target vertex coordinates which possibly exist, performing template matching by using the matching template and a binary image obtained by edge detection, discarding the vertex with low matching similarity, and obtaining the matched target vertex;
and 4, step 4: and performing cluster analysis on the matched target vertexes, and removing false vertexes to obtain a detection result.
Further, the step 1 comprises the following steps:
step 1.1, recording the original ground penetrating radar recording section as a matrix IN×MThe x-th row and t-th column elements are marked as IN×M[x,t]And x represents the direction dimension of the acquired data and has the value range of [1, N]N represents the number of aperture points sampled in the direction of the survey line; t represents a time dimension with a value range of [1, M]M represents a sampling number point of a one-dimensional scattering echo acquired at a certain aperture point after digital sampling; total sampling duration is Tns, time samplingAt an interval of
Figure BDA0001372125100000031
IN×M[x,t]Pixel value pair I of pixel in x-th row and t-th column on original ground penetrating radar recording sectionN×MPreprocessing, namely removing direct waves and normalizing amplitudes, and recording the result as a matrix I ″N×M；
Step 1.2, according to the formula
Figure BDA0001372125100000032
Calculating the energy of each data, thereby obtaining a one-dimensional energy curve P (x) of the ground penetrating radar recording section; wherein I ″)N×M[x,t]Denotes the matrix I ″)N×MRow xth column element;
step 1.3, searching a local maximum value point on a one-dimensional energy curve P (x), and marking as (x, P (x));
step 1.4, counting the number of all local maximum value points and recording as S; setting the preset interval of the number of target vertexes of the ground penetrating radar recording profile as Smin,Smax]In which S isminIs round (u.S), SmaxFor round (v · S), u and v are empirical parameters, round (·) denotes rounding.
Further, u is 0.8 and v is 1.2.
Further, step 1.1 is for IN×MThe steps of the direct wave removing and amplitude normalization processing are as follows:
defining a matrix ONE with all 1 elements1×MFrom the matrix IN×MThe first row of elements is taken out and is marked as IN×1According to formula (1) for matrix IN×MDirect wave removing treatment is carried out to obtain matrix I'N×M(ii) a Then, the matrix I 'is matched according to the formula (2)'N×MCarrying out amplitude normalization processing to obtain I ″)N×M；
I′N×M＝IN×M-IN×1·ONE1×M(1)
I″N×M＝I′N×M/Max{I′N×M[x,t]} (2)。
Further, the step 2 comprises the following steps:
step 2.1, set initial threshold α ═ α0；
Step 2.2: performing edge detection on the preprocessed ground penetrating radar recording section according to a set threshold value by adopting a Canny edge detection algorithm to obtain a binary image B under the corresponding threshold valueα(x,t)；
Step 2.3: from binary image Bα(x, t) estimating the possible target vertices:
first, for the binary image Bα(x, t) traversing the rows first and then traversing the columns to find a binary image Bα(x, t) all points in which the pixel value is 1; defining a binary image BαAll the points with the pixel value of 1 in (x, t) are edge points of the binary image and are marked as E (x)m,tm) M is 1,.. multidot.m ', wherein M is the label of the edge point, and M' is the total number of the edge points; connecting the continuous edge points into an edge curve;
then, all edge points are processed point by point, and possible vertexes are estimated:
first, aiming at any edge point E (x)m,tm) Calculating the number of edge points which are continuous on the right of the point and have the same gradient, and marking as N '(namely the pixel values of N' continuous pixel points on the right of the edge points are all 1, namely the pixel values are all edge points and have the same gradient; calculating the gradient of the edge point to obtain the image gradient; solving the image gradient can regard the image as a two-dimensional discrete function, and the image gradient is the derivation of the two-dimensional discrete function; image gradient: g (x, t) ═ dx (i, j) + dt (i, j); dx (x, t) ═ Iα(i+1,j)-Iα(i,j)；dy(x,t)＝Iα(i,j+1)-Iα(I, j) wherein Iα(i, j) is a binary image BαThe upper position of (x, t) is the pixel value of the pixel point at (i, j);
based on the edge point E (x)m,tm) Setting sub-image [ xm:(xm+n′),(tm-n′):(tm+n′+N′)]Taking binary image BαX in (x, t)mGo to xm+ n' th row, tm-n' columnsTo the t < th > tmThe area between the + N ' + N ' columns is used as a sub-image, wherein N ' is used for setting the size of the sub-image, and the value is 5 in the method;
if the left side of the sub-image [ x ]m:(xm+n′),(tm-n′):tm]The gradient of the edge point is ascending and the right side of the sub-image [ xm:(xm+n′),(tm+N′):(tm+n′+N)]If the gradient of the edge point is in a descending trend, the target vertex possibly existing in the sub-image is shown, and the coordinates of the target vertex possibly existing in the sub-image are calculated to be (x)m,tm+ t '), wherein t' is
Figure BDA0001372125100000041
Recording the coordinates of the target vertex, otherwise, indicating that the target vertex does not exist in the sub-image;
after the above-mentioned treatment is carried out for each edge point, a binary image B is obtainedαAll possible target vertices in (x, t), denoted as Eα(xk,tk),k＝1,2,…,K，(xk,tk) The K is the coordinate of the kth possible target vertex, and the K is the number of the possible target vertices;
step 2.4: judging whether K is in a preset interval of the number of target vertexes [ S ]min,Smax]If yes, ending step 2, and taking the current estimated Eα(xk,tk) K is 1,2, …, K is processed in the subsequent step 3, otherwise, the threshold value is increased, the threshold value α is set to α + delta α, and the process is repeated in the step 2.2 until α is α1So long as α ═ α1And K is not in the preset interval [ S ]min,Smax]The method has the advantages that the signal-to-noise ratio of the profile image recorded by the ground penetrating radar is extremely small, more false vertexes (too serious noise interference) exist in the image, and the situation is generally avoided; if such extreme conditions occur, the target vertex in the image cannot be detected.
Further, the α0＝0.01，Δα＝0.02，α1＝0.9。
Further, the step 3 comprises the following steps:
step 3.1: setting standard hyperbolic template equation as
Figure BDA0001372125100000051
Wherein, a ═ t0，
Figure BDA0001372125100000052
Wherein t is0Representing the minimum time delay, x, of the ground penetrating radar to the underground target0Represents the lateral detection distance, epsilon, corresponding to the minimum time delayrRepresents the relative permittivity of the target;
step 3.2: initializing k to 1;
step 3.3: based on the target vertex Eα(xk,tk) Establishing a sub-binary image;
based on the possible existing target vertex E found in step 2α(xk,tk) Establishing a sub-binary image B'α(x,t),x∈[xk min,xk max],t∈[tk min,tk max]Wherein x isk min、xk max、tk minAnd tk maxAre respectively xk、xk+x′、tk-b and tk+b，x′＝0.2·T0，T0Is the period of the antenna or antennas,
Figure BDA0001372125100000053
f is the frequency of the ground penetrating radar antenna;
step 3.4: based on the target vertex Eα(xk,tk) Establishing a matching template;
in the case of noise, since the edge curve of the actual data image has a certain deviation from the standard hyperbolic template, the standard hyperbolic template is corrected by defining a parameter Δ b indicating the magnitude of the deviation. As the image noise increases, the corresponding threshold value of the edge detection also increases, so that the deviation of the edge curve after the edge detection compared with the standard hyperbolic template also increases, and the detection threshold value has a line with the parameter Δ bA linear relationship, where Δ b is set to 10 × α, α represents a threshold value in edge detection, and the corrected hyperbola is
Figure BDA0001372125100000054
Will [ tk min,tk max]The internal integer t is substituted into the corrected hyperbola
Figure BDA0001372125100000055
Obtaining corresponding x, satisfying x epsilon [ x ∈k min,xk max]All the solutions are edge points corresponding to the corrected hyperbola; rendering the edge point corresponding to the corrected hyperbola in the sub-binary image B'α(x,t),x∈[xk min,xk max],t∈[tk min,tk max]Obtaining a matching template marked as M in the binary images with the same sizek(x,t),x∈[xk min,xk max],t∈[tk min,tk max]Two hyperbolas in the matching template are respectively corrected by the two hyperbolas
Figure BDA0001372125100000056
And
Figure BDA0001372125100000057
two theoretical hyperbolas are obtained because the corrected hyperbolas have ± Δ b in the pair, and have the same vertex as shown in fig. 6, in the frame, the longer curve in the middle is the edge curve to be detected, and the two curves above and below are the corresponding standard hyperbolas of ± Δ b;
step 3.5: using matching templates Mk(x,t),x∈[xk min,xk max],t∈[tk min,tk max]And sub-binary image B'α(x,t),x∈[xk min,xk max],t∈[tk min,tk max]Carrying out template matching and calculating matching similarity rhokAnd will be rhokMatching similarity threshold rho preset0Comparing; if ρk>ρ0Then, consider E (x)k,tk) For matched vertices, otherwise consider E (x)k,tk) Not matching vertices;
step 3.6: setting K to be K +1, if K is less than or equal to K, repeating the steps 3.3-3.5 to establish a sub-binary image and a matching template for the next vertex, and performing matching similarity calculation and comparison judgment with a matching similarity threshold; if k is>K is finished, all the matched vertexes are recorded as V (x)β,tβ),β＝1,...,B，(xβ,tβ) Is the coordinates of the β th matched target vertex, and B is the total number of matched target vertices.
Further, in the step 3.5, the matching similarity ρ is calculated by the following formulak：
ρk＝((ρ1+ρ3+ρ4+ρ5)·2+ρ2+ρ6)·0.1 (3)
Where ρ is1、ρ2、ρ3、ρ4、ρ5And ρ6Respectively representing template matching constraint parameters dreli、sym、drti、drxi、disiAnd the matching similarity of sur, the calculation formula is as follows:
Figure BDA0001372125100000061
Figure BDA0001372125100000062
the calculation method of each template matching parameter is as follows:
buffer, i.e. sub-binary image B'α(x,t),x∈[xk min,xk max],t∈[tk min,tk max]By a line x ═ x0Is boundary component B'α(x,t),x∈(xk min,x0),t∈[tk min,tk max]And B'α(x,t),x∈(x0,xk max),t∈[tk min,tk max]Left and right partsA left buffer area and a right buffer area;
1) the relative edge point densities of the left and right buffers are calculated according to the following formula:
Figure BDA0001372125100000063
wherein drel1And drel2Respectively representing the relative edge point densities of the left buffer area and the right buffer area; n is a radical ofdata1And Ndata2Respectively representing the number of edge points (namely points with the pixel value of 1 in the buffer area) in the left buffer area and the right buffer area; n is a radical oftheoIs a matching template Mk(x,t),x∈[xk min,xk max],t∈[tk min,tk max]The number of the upper edge points (namely the edge points corresponding to the corrected hyperbola); in the ideal case: dreliShould be close to 100%, but the acceptable range is between 50% and 270%.
2) Calculating the symmetry normalization coefficient of the edge curve according to the relative edge point density of the left buffer area and the right buffer area:
Figure BDA0001372125100000071
ideally, sym is close to 1, and an acceptable range is sym ≧ 0.6, which is acceptable as a hyperbolic curve from a symmetry point of view.
3) Calculating data ranges of the left buffer area and the right buffer area relative to the theoretical hyperbolic curve in the t direction and the x direction;
the data range calculation method of the left buffer area and the right buffer area relative to the theoretical hyperbola in the t direction is as follows:
Figure BDA0001372125100000072
wherein, drt1And drt2The data ranges of the left buffer area and the right buffer area relative to the theoretical hyperbola in the direction t are respectively; trt1And trt2Respectively, to match the matching template toWhen the buffer area is on, the data range (the absolute value of the difference value between the minimum value and the maximum value) of the theoretical hyperbola on the matching template in the t direction in the left buffer area and the right buffer area is matched; adrt1And adrt2Are the data ranges in the t direction of the left and right buffers, respectively.
Similarly, the data range calculation method of the left buffer area and the right buffer area in the x direction relative to the theoretical hyperbola is as follows:
Figure BDA0001372125100000073
wherein, drx1And drx2The data ranges of the left buffer area and the right buffer area relative to the theoretical hyperbolic curve in the x direction are respectively; trx1And trx2Respectively means the data ranges (absolute values of the difference between the minimum value and the maximum value) of theoretical hyperbolas on the matching template in the x direction in the left buffer area and the right buffer area when the matching template is matched to the buffer area; adrx1And adrx2Are the data ranges in the x-direction for the left and right buffers, respectively. Under ideal conditions, drtiAnd drxiThe value of (a) approaches 1; in practice, drtiAnd drxiThe acceptable ranges of (a) are: drti≥0.7，drxi≥0.7。
4) Calculating the data distribution of the left buffer and the right buffer in the direction t relative to the theoretical hyperbola:
Figure BDA0001372125100000074
wherein dis1And dis2Respectively representing the data distribution of the left buffer and the right buffer in the direction t relative to the theoretical hyperbola, Nmatch1And Nmatch2Respectively representing the number of edge points in the left buffer and the right buffer between two theoretical hyperbolas falling in the matching template [ note
Figure BDA0001372125100000075
And
Figure BDA0001372125100000076
respectively representing the edge points on two theoretical hyperbolas in the template, and xm1≤xm2For edge points E (x) in the left and right buffersm,tm) If t is satisfiedm1＝tm＝tm2And xm1≤xm≤xm2Then, the edge point E (x) is describedm,tm) Fall between two theoretical hyperbolas in the matching template;
if the relative data range and relative edge distribution density are considered only, it is not enough to determine the hyperbola, such as the pattern generated in case of mutual interference, the data distribution must be considered. The conditions for judgment are as follows: disiBelow 0.7, a hyperbola is considered to be present.
5) The relative number sur of edge points outside the edge curve in the buffer area is calculated as follows.
Figure BDA0001372125100000081
Even if the data distribution conditions satisfy disiLess than 0.7, but possibly a higher number inside the buffer, will affect the determination of the hyperbola. Wherein: n is a radical ofsurNumber of discrete edge points outside the edge curve in the buffer (including both left and right buffers) [ for discrete edge points E (x) in the left bufferd,td) If it is relative to a certain edge point E (x) on the edge curve in the bufferc,tc) Satisfy xd<xcAnd t isd<tc(ii) a For discrete edge points E (x) in the right bufferd,td) If it is relative to a certain edge point E (x) on the edge curve in the bufferc,tc) Satisfy xd>xcAnd t isd<tcThen, these edge points are described to be outside the edge curve; n is a radical ofbfNumber of discrete edge points within the edge curve represented in the buffer (including both left and right buffers) [ for discrete edge points E (x) within the left bufferd,td) If it is relative to a certain edge point E (x) on the edge curve in the bufferc,tc) Satisfy xd>xcAnd t isd>tc(ii) a For discrete edge points E (x) in the right bufferd,td) If it is relative to a certain edge point E (x) on the edge curve in the bufferc,tc) Satisfy xd<xcAnd t isd>tcThen, the edge points are described to be inside the edge curve; ideally, the value of sur is small, i.e., NbfAs the size decreases, the corresponding sur decreases. Therefore, in the case where sur.ltoreq.0.7, it is acceptable.
Setting a template matching constraint parameter rho1～ρ6The template matching similarity when all equal to 1 is taken as a matching similarity threshold value, namely rho0＝1。
Further, the step 4 comprises the following steps:
step 4.1: for the matched target vertex V (x) obtained in step 3β,tβ) B cluster analysis to obtain cluster C, β ═ 1n，n＝1,...,N1Where N is the cluster number, N1For the number of clusters, the target vertices contained in each cluster are recorded as the number Qn，n＝1,...,N1；
Step 4.2: setting an initial value n-1 and gamma-1;
step 4.3: if QnIf 1 is true, cluster C will be formednThe target vertex contained in the cluster is used as the filtered target vertex V in the clusterγ(ii) a If QnIf 1 fails, then cluster C is comparednIf a plurality of target vertexes with the maximum and equal energy values exist in the cluster, taking the corresponding target vertex with the minimum time point in the target vertexes as the filtered target vertex in the cluster; if only one target vertex with the maximum energy value exists in the cluster, taking the target vertex as the filtered target vertex V in the clusterγ；
Step 4.4: judging N as N1Whether it is true or not, if notThen set up
Figure BDA0001372125100000091
Returning to the step 4.3 to filter the vertex in the next cluster; if N is equal to N1If yes, ending; traverse N1Clustering to obtain a vertex estimation sequence V in the original ground penetrating radar recording sectionγ，γ＝1,...,N1As vertex detection results.
Has the advantages that:
the invention provides a method for detecting a target vertex in a GPR recording profile based on automatic threshold edge detection and template matching. And carrying out edge detection on the GPR recording section to obtain a curve edge profile in the image. Estimating the coordinate position of the vertex by utilizing the gradient change relation of the edge points; generating a corresponding sub-image window based on the position of the vertex, and generating a standard hyperbolic matching template by combining hyperbolic fitting; setting the six matching conditions aiming at the edge points to obtain the similarity of template matching, and removing the vertex with lower similarity; and finally, carrying out cluster analysis on the obtained vertex to remove the false vertex so as to obtain the target vertex to be detected. The invention automatically matches the optimal threshold without manual adjustment, thereby improving the flexibility of the algorithm. Under the condition of noise, the edge points of the actual data image have certain deviation compared with the standard hyperbolic template, and therefore, a parameter delta b is defined to represent the magnitude of the deviation; through the control of the delta b parameter, the method has better detection performance under the condition of noise, and can be applied to target detection in the actual GPR recording profile containing noise.
Drawings
FIG. 1 illustrates a process flow diagram of the present invention;
FIG. 2 shows a GPR raw B-Scan image;
FIG. 3 shows an image after raw data has been pre-processed;
FIG. 4 shows a one-dimensional energy curve of an image after pre-processing;
FIG. 5 illustrates predicted vertices after edge detection;
FIG. 6 illustrates a matching template curve generated from the coordinates of the vertices;
FIG. 7 shows the detection results after template matching has been performed to discard vertices with lower similarity;
FIG. 8 shows vertex detection results from removing false vertices using cluster analysis;
FIG. 9 shows the result of vertex detection using conventional methods for the GPR image shown in FIG. 2.
Detailed Description
The invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.
The method mainly and automatically detects the target vertex of the GPR image, as shown in FIG. 1, adopts GPRMax2.0 simulation software to generate an original image, and carries out noise adding processing simulation to generate a noise-containing actual measurement GPR image. The method comprises the steps of processing a noise-containing GPR image by using the algorithm, preprocessing the noise-containing original image, performing edge detection on a preprocessed ground penetrating radar recording section in an automatic threshold matching mode, setting an initial threshold, performing edge detection and vertex estimation, and judging whether the number of vertexes is in a preset interval. If yes, taking the current edge detection result to carry out the next step. And if not, increasing the threshold value until the number of the target vertexes estimated in advance is in a preset interval. And then extracting target vertexes in the binary image obtained by edge detection, establishing a matching template based on the coordinate positions of the target vertexes, performing template matching by using the template and the binary image obtained by edge detection, calculating matching similarity, setting a matching threshold, discarding vertexes with the matching similarity lower than the threshold, and obtaining the matched target vertexes. And filtering the target vertex based on a clustering analysis method, removing false vertices, and finally obtaining a detection result of the target vertex in the ground penetrating radar recording profile.
Example 1:
in this example, a ground penetrating radar original image is generated through gprmax2.0 simulation software, the length of a lateral interval of GPR scanning is 2.5m, the total number of sampling points in the lateral dimension is 920 points, and the sampling interval between two adjacent aperture points is 0.25 mm. The vertical dimension is a time dimension, the total number of sampling points is 1357 points, the total sampling time is 12ns, and the size of the data is 1357 multiplied by 920 of a two-dimensional matrix. The simulated GPR data was normalized to ideal data containing no noise. To simulate a real GPR probing scenario, noise was added such that the signal-to-noise ratio SNR was equal to 10dB and the GPR recording profile after noise addition is shown in fig. 2.
Preprocessing a GPR original image, removing direct waves in the image by using formulas (1) and (2), and normalizing the amplitude to obtain a preprocessed image as shown in FIG. 3;
drawing a one-dimensional energy curve of the original image, as shown in fig. 4; according to the one-dimensional energy curve characteristics of the original image, vertexes possibly exist at extreme points, the number of the extreme points is calculated to be 23, and the value range [18,28] of the number of the vertexes possibly existing is set;
then, the edge detection is carried out on the preprocessed image to find out the optimal edge detection threshold value, a Canny edge detection function is adopted, and an initial threshold value α is set00.01, α are incremented in steps of 0.02 to end at 0.9. binary image B for edge detection is acquired according to each thresholdα(x, t), and estimating the possible target vertex according to the binary image. For the binary image Bα(x, t) traversing the rows first and then traversing the columns to find a binary image Bα(x, t) all points in which the pixel value is 1; defining a binary image BαAll the points with the pixel value of 1 in (x, t) are edge points of the binary image and are marked as E (x)m,tm) M is 1,.. multidot.m ', wherein M is the label of the edge point, and M' is the total number of the edge points;
then, all edge points are processed point by point, and possible vertexes are estimated: for any one edge point E (x)m,tm) Calculating the number of edge points with the same gradient and the same continuous right edge as N, and then calculating the number of edge points based on the edge point E (x)m,tm) Setting sub-image [ x ]m:(xm+n′),(tm-n′):(tm+n′+N′)]Where n 'denotes the number of pixel points, which is used to set the size of the sub-image, and in this method the value is n' 5. If the left side of the sub-image [ x ]m:(xm+n′),(tm-n′):tm]Has a gradient of upper edge pointsIncreasing trend, and the right side of the sub-image [ xm:(xm+n′),(tm+N′):(tm+n′+N)]The gradient of the edge point of (a) is a descending trend; it can be stated that there may be a vertex in this sub-image, when the vertex coordinate is calculated as (x)m,tm+ t '), wherein t' is
Figure BDA0001372125100000111
And recording the vertex coordinates; finding out the estimation result of all the vertexes in the image according to the gradient relation of the edge points in the sub-image and recording the result as Eα(xk,tk) K is 1,2, …, K being the number of vertices. Marking all the estimated vertices in the edge-detected binary image as shown in fig. 5, according to the above-described processing procedure, 23 estimated vertices, i.e., K ═ 23, are found in the present example.
Setting a standard hyperbolic template equation as
Figure BDA0001372125100000112
Wherein a ═ x0,
Figure BDA0001372125100000113
Finding possible vertex E according to the aboveα(xk,tk) K1, 2, …, K, then traverse each vertex, and when K1, then establish [ x ] based on vertex (226,150)k:(xk+x·),(tk-b):(tk+b)]The size of the subimage is Bα(x,t),x∈[145,155],t∈[226,232]In the case of noise, Δ b is defined as 10 · α, and the standard hyperbolic template is corrected to obtain a hyperbolic equation
Figure BDA0001372125100000114
Drawing the edge points corresponding to the corrected hyperbola in the set matching template to establish the matched template Mk(x,t),x∈[145,155],t∈[226,232]As shown in fig. 6.
The following template matching process, step 7, is to determine the matching constraint parameters according to the set 6 matching constraint parametersThe first parameter. Separately, drel is obtained from equation (4)12.75 and drel25.37 according to equation (5), drel1And drel2Obtaining a symmetry normalization coefficient sym which is 0.51; drt is derived from equation (6)1When the value is equal to 0.71, drt is obtained by the same method2＝0.71、dry1＝0.92、dry22.53; next calculate disi: obtaining dis according to equation (7)1＝0.50、dis20.37; the sur is-4.54 obtained according to the formula (8), and the template matching similarity ρ is obtained by combining the above results through the formula (3)k0.6, less than the similarity threshold. Therefore, it is considered (226,150) not to be the target vertex. Repeating the judging mode, and obtaining all matched vertexes by performing the judging process on all vertexes and recording the vertexes as V (x)β,tβ) β, B, and marks the locations of these vertices on the original image, as shown in fig. 7.
And filtering the target vertex after matching:
performing cluster analysis on the target vertexes obtained in the above steps to obtain a cluster CnN is 1, 3, and the number of vertices included in each cluster is Q1＝3,Q2＝2,Q3Setting an initial value n to 1 and γ to 1; for cluster CnIf Q isnIf 1 is true, then Vγ＝VnIn which V isγFor the filtered target vertex, VnIs a cluster CnThe only one vertex included in (a); if Q isnIf 1 is not true, the vertex with the largest energy value is extracted from the cluster, and the vertex number is marked as [ m ]1,...,mp]In this case, if there is only one vertex with the largest energy value in the cluster, that is, if p ═ 1 is true, it is directly considered that
Figure BDA0001372125100000121
For the filtered target vertex V in the clusterγ(ii) a Otherwise, if p ≠ 1, which indicates that a plurality of vertexes with the maximum and equal energy values exist in the cluster, extracting the vertex with the minimum corresponding time point as the filtered target vertex V in the clusterγI.e. by
Figure BDA0001372125100000122
If N is equal to N1If not, setting
Figure BDA0001372125100000123
The above processing procedure is carried out on the next vertex, and N is traversed1Clusters of clusters, resulting in a sequence of vertex estimates V in the GPR recorded profileγI.e., vertex detection results, as shown in fig. 8.
The conventional method is used to perform vertex extraction on the GPR recording section shown in fig. 2, and as a result, 7 target vertices are detected in total by manually adjusting the threshold as shown in fig. 9. There is actually only one target vertex at label 1, whereas the conventional method detects 2 target points, including 1 false vertex. In the figure, there are no target vertices at the positions marked 2, 3 and 4, and the target vertices are detected at the position 3, so that false detection occurs. The invention can automatically match the threshold value, improves the flexibility of the algorithm, has higher detection precision under the condition of low signal-to-noise ratio, and has no conditions of false detection and missing detection.

Claims (10)

1. A method for detecting a target vertex in a ground penetrating radar recording profile is characterized by comprising the following steps:
step 1: preprocessing an original ground penetrating radar recording section, calculating a one-dimensional energy curve of the preprocessed ground penetrating radar recording section along the direction of a measuring line, extracting the number of local maximum values of the energy curve, and setting a preset interval of the number of target vertexes of the ground penetrating radar recording section according to the number;
step 2: performing automatic threshold matching edge detection on the preprocessed ground penetrating radar recording section to obtain a binary image, and performing automatic threshold matching on the binary image Bα(x, t) estimating the possible target vertex, and taking the binary image BαAll possible target vertices in (x, t), denoted as Eα(xk,tk),k＝1,2,…,K，(xk,tk) The K is the coordinate of the kth possible target vertex, and the K is the number of the possible target vertices;
and step 3: establishing a matching template according to the estimated target vertex coordinates which possibly exist, performing template matching by using the matching template and a binary image obtained by edge detection, and discarding a target vertex with low matching similarity to obtain a matched target vertex;
and 4, step 4: performing cluster analysis on the matched target vertexes, and removing false vertexes to obtain a detection result;
the step 3 is based on the estimated possible existing target vertex Eα(xk,tk) The establishment of the matching template by the coordinates comprises the following steps:
step 3.1: setting standard hyperbolic template equation as
Figure FDA0002189812000000011
Wherein, a ═ t0，
Figure FDA0002189812000000012
Wherein t is0Representing the minimum time delay, x, of the ground penetrating radar to the underground target0Represents the lateral detection distance, epsilon, corresponding to the minimum time delayrRepresents the relative permittivity of the target;
step 3.2: based on the target vertex Eα(xk,tk) Establishing a sub-binary image;
based on the possible existing target vertex E found in step 2α(xk,tk) Establishing a sub-binary image B'α(x,t),x∈[xkmin,xkmax],t∈[tkmin,tkmax]Wherein x iskmin、xkmax、tkminAnd tkmaxAre respectively xk、xk+x′、tk-b and tk+b，x′＝0.2·T0，T0Is the period of the antenna or antennas,
Figure FDA0002189812000000013
f is the frequency of the ground penetrating radar antenna;
step 3.3: based on the target vertex Eα(xk,tk) EstablishingMatching the templates;
correcting a standard hyperbolic template to obtain a corrected hyperbolic template
Figure FDA0002189812000000014
Where Δ b — 10 × α denotes a threshold value in edge detection;
will [ tkmin,tkmax]The internal integer t is substituted into the corrected hyperbola
Figure FDA0002189812000000015
Obtaining corresponding x, satisfying x epsilon [ x ∈kmin,xkmax]All the solutions are edge points corresponding to the corrected hyperbola;
rendering the edge point corresponding to the corrected hyperbola in the sub-binary image B'α(x,t),x∈[xkmin,xkmax],t∈[tkmin,tkmax]Obtaining a matching template marked as M in the binary images with the same sizek(x,t),x∈[xkmin,xkmax],t∈[tkmin,tkmax]Two hyperbolas in the matching template are respectively corrected by the two hyperbolas
Figure FDA0002189812000000021
And
Figure FDA0002189812000000022
the hyperbolas to which the corresponding edge points are connected are called theoretical hyperbolas.
2. The method for detecting the target vertex in the georadar recording profile according to claim 1, wherein the step 1 comprises the following steps:
step 1.1, recording the original ground penetrating radar recording section as a matrix IN×MThe x-th row and t-th column elements are marked as IN×M[x,t]And x represents the direction dimension of the acquired data and has the value range of [1, N]N represents the number of aperture points sampled in the direction of the survey line; t represents a time dimension with a value range of [1, M]M tableDisplaying sampling number points of one-dimensional scattering echoes acquired at a certain aperture point after digital sampling; i isN×M[x,t]Representing the pixel value of the pixel of the t column of the x row on the original ground penetrating radar recording section; to IN×MPreprocessing, namely removing direct waves and normalizing amplitudes, and recording the result as a matrix I ″N×M；
Step 1.2, according to the formula
Figure FDA0002189812000000023
Calculating the energy of each data, thereby obtaining a one-dimensional energy curve P (x) of the ground penetrating radar recording section; wherein I ″)N×M[x,t]Denotes the matrix I ″)N×MRow xth column element;
step 1.3, searching a local maximum value point on a one-dimensional energy curve P (x), and marking as (x, P (x));
step 1.4, counting the number of all local maximum value points and recording as S; setting the preset interval of the number of target vertexes of the ground penetrating radar recording profile as Smin,Smax]In which S isminIs round (u.S), SmaxFor round (v · S), u and v are empirical parameters, round (·) denotes rounding.
3. The method for detecting the target vertex in the georadar recording profile according to claim 2, wherein u is 0.8 and v is 1.2.
4. The method for detecting the target vertex in the georadar recording profile according to claim 2, wherein the step 1.1 is IN×MThe steps of the direct wave removing and amplitude normalization processing are as follows:
defining a matrix ONE with all 1 elements1×MFrom the matrix IN×MThe first row of elements is taken out and is marked as IN×1According to formula (1) for matrix IN×MDirect wave removing treatment is carried out to obtain matrix I'N×M(ii) a Then, the matrix I 'is matched according to the formula (2)'N×MCarrying out amplitude normalization processing to obtain I ″)N×M；
I′N×M＝IN×M-IN×1·ONE1×M(1)
I″N×M＝I′N×M/Max{I′N×M[x,t]} (2)。
5. The method for detecting the target vertex in the georadar recording profile according to claim 1, wherein the step 2 comprises the following steps:
step 2.1, set initial threshold α ═ α0；
Step 2.2: performing edge detection on the preprocessed ground penetrating radar recording section according to a set threshold value by adopting a Canny edge detection algorithm to obtain a binary image B under the corresponding threshold valueα(x,t)；
Step 2.3: from binary image Bα(x, t) estimating the possible target vertices:
first, for the binary image Bα(x, t) traversing the rows first and then traversing the columns to find a binary image Bα(x, t) all points in which the pixel value is 1; defining a binary image BαAll the points with the pixel value of 1 in (x, t) are edge points of the binary image and are marked as E (x)m,tm) M is 1,.. multidot.m ', wherein M is the label of the edge point, and M' is the total number of the edge points; connecting the continuous edge points into an edge curve;
then, all edge points are processed point by point, and possible vertexes are estimated:
first, aiming at any edge point E (x)m,tm) Calculating the number of edge points which are continuous at the right side of the point and have the same gradient, and recording the number as N';
based on the edge point E (x)m,tm) Setting sub-image [ xm:(xm+n′),(tm-n′):(tm+n′+N′)]Taking binary image BαX in (x, t)mGo to xm+ n' th row, tmN' columns to tmThe area between the + N ' + N ' columns is used as the sub-image, where N ' is used to set the sub-image size;
if the left side of the sub-image [ x ]m:(xm+n′),(tm-n′):tm]The gradient of the edge point is ascending and the right side of the sub-image [ xm:(xm+n′),(tm+N′):(tm+n′+N)]If the gradient of the edge point is in a descending trend, the target vertex possibly existing in the sub-image is shown, and the coordinates of the target vertex possibly existing in the sub-image are calculated to be (x)m,tm+ t '), wherein t' is
Figure FDA0002189812000000031
Recording the coordinates of the target vertex, otherwise, indicating that the target vertex does not exist in the sub-image;
after the above-mentioned treatment is carried out for each edge point, a binary image B is obtainedαAll possible target vertices in (x, t), denoted as Eα(xk,tk),k＝1,2,…,K，(xk,tk) The K is the coordinate of the kth possible target vertex, and the K is the number of the possible target vertices;
step 2.4: judging whether K is in a preset interval of the number of target vertexes [ S ]min,Smax]If yes, ending step 2, and taking the current estimated Eα(xk,tk) K is 1,2, …, K is processed in the subsequent step 3, otherwise, the threshold value is increased, the threshold value α is set to α + delta α, and the process is repeated in the step 2.2 until α is α1Until now.
6. The method of claim 5, wherein α is used to detect the target vertex in the ground penetrating radar recording profile0＝0.01，Δα＝0.02，α1＝0.9。
7. The method for detecting the target vertex in the ground penetrating radar recording section as claimed in claim 5, wherein in the step 3, the binary image B is processedαEach possible target vertex E in (x, t)α(xk,tk) K is 1,2, …, K, and the corresponding sub-binary image B is created according to step 3.2 to step 3.3 respectively′α(x,t),x∈[xkmin,xkmax],t∈[tkmin,tkmax]And matching the template Mk(x,t),x∈[xkmin,xkmax],t∈[tkmin,tkmax]And using the established matching template to match with the sub-binary image to calculate matching similarity rhokAnd will be rhokMatching similarity threshold rho preset0Comparing; if ρk＞ρ0Then, consider E (x)k,tk) For matched vertices, otherwise consider E (x)k,tk) Not matching vertices;
all the matched vertices obtained are recorded as V (x)β,tβ),β＝1,...,B，(xβ,tβ) Is the coordinates of the β th matched target vertex, and B is the total number of matched target vertices.
8. The method for detecting the target vertex in the ground penetrating radar recording profile according to claim 7, wherein the matching similarity p is calculated by the following formulak：
ρk＝((ρ1+ρ3+ρ4+ρ5)·2+ρ2+ρ6)·0.1 (3)
Where ρ is1、ρ2、ρ3、ρ4、ρ5And ρ6Respectively representing template matching constraint parameters dreli、sym、drti、drxi、disiAnd the matching similarity of sur, the calculation formula is as follows:
Figure FDA0002189812000000041
Figure FDA0002189812000000042
the calculation method of each template matching parameter is as follows:
buffer, i.e. sub-binary image B'α(x,t),x∈[xkmin,xkmax],t∈[tkmin,tkmax]By a line x ═ x0Is boundary component B'α(x,t),x∈(xkmin,x0),t∈[tkmin,tkmax]And B'α(x,t),x∈(x0,xkmax),t∈[tkmin,tkmax]The left and right parts are left and right buffer zones;
1) the relative edge point densities of the left and right buffers are calculated according to the following formula:
Figure FDA0002189812000000051
wherein drel1And drel2Respectively representing the relative edge point densities of the left buffer area and the right buffer area; n is a radical ofdata1And Ndata2Respectively representing the number of edge points in the left buffer area and the right buffer area; n is a radical oftheoIs a matching template Mk(x,t),x∈[xkmin,xkmax],t∈[tkmin,tkmax]The number of edge points above;
2) calculating the symmetry normalization coefficient of the edge curve according to the relative edge point density of the left buffer area and the right buffer area:
Figure FDA0002189812000000052
3) calculating data ranges of the left buffer area and the right buffer area relative to the theoretical hyperbolic curve in the t direction and the x direction;
the data range calculation formula of the left buffer area and the right buffer area relative to the theoretical hyperbola in the t direction is as follows:
Figure FDA0002189812000000053
wherein, drt1And drt2The data ranges of the left buffer area and the right buffer area relative to the theoretical hyperbolic curve in the t direction are respectively; trt1And trt2Respectively, matching the matching templates to the bufferThen, matching the data ranges of the theoretical hyperbolas on the template in the t direction in the left buffer area and the right buffer area; adrt1And adrt2Are respectively the data ranges of the left buffer area and the right buffer area in the t direction;
similarly, the data range calculation method of the left buffer area and the right buffer area in the x direction relative to the theoretical hyperbola is as follows:
Figure FDA0002189812000000054
wherein, drx1And drx2The data ranges of the left buffer area and the right buffer area relative to the theoretical hyperbolic curve in the x direction are respectively; trx1And trx2Respectively means the data ranges of theoretical hyperbolas on the matching template in the x direction in the left buffer area and the right buffer area when the matching template is matched to the buffer area; adrx1And adrx2Are respectively the data ranges of the left buffer area and the right buffer area in the x direction;
4) calculating the data distribution of the left buffer and the right buffer in the direction t relative to the theoretical hyperbola:
Figure FDA0002189812000000055
wherein dis1And dis2Respectively representing the data distribution of the left buffer and the right buffer in the direction t relative to the theoretical hyperbola, Nmatch1And Nmatch2Respectively representing the number of the edge points in the left buffer area and the right buffer area between two theoretical hyperbolas in the matching template;
5) the relative number sur of edge points outside the edge curve in the buffer area is calculated as follows:
Figure FDA0002189812000000061
wherein N issurRepresenting the number of discrete edge points outside the edge curve in the buffer; n is a radical ofbfTo representThe number of discrete edge points within the edge curve in the buffer.
9. The method for detecting the target vertex in the ground penetrating radar recording profile according to claim 8, wherein a template matching constraint parameter p is set1～ρ6The template matching similarity when all equal to 1 is taken as a matching similarity threshold value, namely rho0＝1。
10. The method for detecting the target vertex in the ground penetrating radar recording section according to any one of claims 1 to 9, wherein the step 4 comprises the following steps:
step 4.1: for the matched target vertex V (x) obtained in step 3β,tβ) B cluster analysis to obtain cluster C, β ═ 1n，n＝1,...,N1Where N is the cluster number, N1For the number of clusters, the target vertices contained in each cluster are recorded as the number Qn，n＝1,...,N1；
Step 4.2: setting an initial value n-1 and gamma-1;
step 4.3: if QnIf 1 is true, cluster C will be formednThe target vertex contained in the cluster is used as the filtered target vertex V in the clusterγ(ii) a If QnIf 1 fails, then cluster C is comparednIf a plurality of target vertexes with the maximum and equal energy values exist in the cluster, taking the corresponding target vertex with the minimum time point in the target vertexes as the filtered target vertex in the cluster; if only one target vertex with the maximum energy value exists in the cluster, taking the target vertex as the filtered target vertex V in the clusterγ；
Step 4.4: judging N as N1If it is not true, setting
Figure FDA0002189812000000062
Returning to the step 4.3 to filter the target vertex in the next cluster; if N is equal to N1If it is true, then it ends(ii) a Traverse N1Clustering to obtain a target vertex estimation sequence V in the original ground penetrating radar recording sectionγ，γ＝1,...,N1As vertex detection results.
CN201710667148.4A 2017-08-07 2017-08-07 Method for detecting target vertex in ground penetrating radar recording profile Active CN107358617B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201710667148.4A CN107358617B (en) 2017-08-07 2017-08-07 Method for detecting target vertex in ground penetrating radar recording profile

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201710667148.4A CN107358617B (en) 2017-08-07 2017-08-07 Method for detecting target vertex in ground penetrating radar recording profile

Publications (2)

Publication Number Publication Date
CN107358617A CN107358617A (en) 2017-11-17
CN107358617B true CN107358617B (en) 2020-03-20

Family

ID=60287358

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201710667148.4A Active CN107358617B (en) 2017-08-07 2017-08-07 Method for detecting target vertex in ground penetrating radar recording profile

Country Status (1)

Country Link
CN (1) CN107358617B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108387894B (en) * 2018-04-13 2021-07-27 中南大学 Processing method of through-wall radar echo data
CN109145424B (en) * 2018-08-10 2023-09-26 中国铁道科学研究院集团有限公司铁道建筑研究所 Bridge data identification method and system for ground penetrating radar data
CN110133639B (en) * 2019-04-08 2022-12-02 长安大学 Dowel bar construction quality detection method
CN112099032B (en) * 2020-09-07 2024-07-05 自然资源部国土卫星遥感应用中心 Ice crack morphology analysis method and device based on laser altimetry satellite data
CN112859006B (en) * 2021-01-11 2023-08-29 成都圭目机器人有限公司 Method for detecting metal bending cylindrical structure in multichannel ground penetrating radar data
CN113567946B (en) * 2021-07-20 2023-08-29 同济大学 Method for detecting real target and virtual image of millimeter wave radar
CN113791452B (en) * 2021-09-14 2023-05-23 中铁二十局集团第五工程有限公司 Shield tunnel boulder waveform line imaging method
CN114413841B (en) * 2022-01-21 2024-02-27 中国科学院西北生态环境资源研究院 Frozen soil zoning drawing method and device, electronic equipment and storage medium

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105182328A (en) * 2015-09-09 2015-12-23 河南工业大学 Ground penetrating radar underground target positioning method
CN105353373A (en) * 2015-12-16 2016-02-24 武汉大学 Hough transformation based ground penetrating radar target extraction method and device
CN106446919A (en) * 2016-11-04 2017-02-22 深圳市航天华拓科技有限公司 Quick detection method for ground penetrating radar hyperbolic curve target

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8629799B2 (en) * 2011-03-30 2014-01-14 Sandia Research Corporation Surface penetrating radar system and target zone investigation methodology

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105182328A (en) * 2015-09-09 2015-12-23 河南工业大学 Ground penetrating radar underground target positioning method
CN105353373A (en) * 2015-12-16 2016-02-24 武汉大学 Hough transformation based ground penetrating radar target extraction method and device
CN106446919A (en) * 2016-11-04 2017-02-22 深圳市航天华拓科技有限公司 Quick detection method for ground penetrating radar hyperbolic curve target

Non-Patent Citations (7)

* Cited by examiner, † Cited by third party
Title
Automated Detection of Reflection Hyperbolas in Complex GPR Images With No A Priori Knowledge on the Medium;Laurence Mertens 等;《IEEE Transactions on Geoscience and Remote Sensing》;20160131;第54卷(第1期);第580-596页 *
Automatic Analysis of GPR Images: A Pattern-Recognition Approach;Edoardo Pasolli 等;《IEEE Transactions on Geoscience and Remote Sensing》;20090327;第47卷(第7期);第2206-2217页 *
The target detection for GPR images based on curve fitting;Yayu Liu 等;《2010 3rd International Congress on Image and Signal Processing》;20101018;第2876-2879页 *
地下管线探测回波中目标双曲线的快速拟合方法;夏锋 等;《计算机系统应用》;20130515;第22卷(第5期);第85-89页 *
探地雷达图像目标检测算法研究;陈天斌 等;《计算机测量与控制》;20150725;第23卷(第7期);第2450-2452页 *
探地雷达目标回波信号双曲线提取算法研究;张旭 等;《计算机测量与控制》;20161025;第24卷(第10期);第247-250页 *
浅地层探地雷达目标探测和定位新方法;胡进峰 等;《仪器仪表学报》;20060430;第27卷(第4期);第371-375页 *

Also Published As

Publication number Publication date
CN107358617A (en) 2017-11-17

Similar Documents

Publication Publication Date Title
CN107358617B (en) Method for detecting target vertex in ground penetrating radar recording profile
CN105182328B (en) A kind of GPR buried target detection method based on two-dimensional empirical mode decomposition
CN109410230B (en) Improved Canny image edge detection method capable of resisting noise
CN105005042B (en) A kind of GPR buried target localization method
CN108805904B (en) Moving ship detection and tracking method based on satellite sequence image
CN101539629B (en) Remote sensing image change detection method based on multi-feature evidence integration and structure similarity
CN111025286B (en) Ground penetrating radar map self-adaptive selection method for water damage detection
CN110097537B (en) Meat quality quantitative analysis and evaluation method based on three-dimensional texture features
Rahnemoonfar et al. Automatic ice surface and bottom boundaries estimation in radar imagery based on level-set approach
CN111798411B (en) Concrete inner steel bar intelligent positioning method based on ground penetrating radar and deep learning
CN115564771A (en) Concrete crack identification method based on building foundation column
CN113359101B (en) Underground target detection method, system and computer storage medium
CN110110675A (en) A kind of wavelet field of combination of edge information divides shape infrared cirrus detection method
CN111160477B (en) Image template matching method based on feature point detection
Ma et al. Automatic detection of steel rebar in bridge decks from ground penetrating radar data
CN108710888A (en) A kind of Coherent Noise in GPR Record method for registering
Liu et al. The target detection for GPR images based on curve fitting
Reddy et al. Comparative analysis of common edge detection algorithms using pre-processing technique
CN116993740B (en) Concrete structure surface defect detection method based on image data
CN110716239B (en) Fine evaluation method for lithology of well logging gravel rock mass based on electrical imaging
CN107301647B (en) The Qinling Mountains golden monkey method for detecting image edge for improving Prewitt is associated based on grey Type B
Dyana et al. 3D Segmentation of ground penetrating radar data for landmine detection
CN111611858B (en) Multi-angle discrimination-based automatic detection method and device for tilting track surface
CN106355576A (en) SAR image registration method based on MRF image segmentation algorithm
CN109061593B (en) Target extraction method based on ground monitoring radar spectrum data

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