CN111091071A

CN111091071A - Underground target detection method and system based on ground penetrating radar hyperbolic wave fitting

Info

Publication number: CN111091071A
Application number: CN201911185419.8A
Authority: CN
Inventors: 原达; 王崴; 陈飞凡; 李文生; 王冬雨; 苗翠; 崔嘉傲
Original assignee: Shandong Technology and Business University
Current assignee: Shandong Technology and Business University
Priority date: 2019-11-27
Filing date: 2019-11-27
Publication date: 2020-05-01
Anticipated expiration: 2039-11-27
Also published as: CN111091071B

Abstract

The invention discloses a ground penetrating radar hyperbolic wave fitting-based underground target detection method and system, which comprise the following steps: acquiring a GPR image of the ground penetrating radar; performing superpixel segmentation on the GPR image to obtain a superpixel segmented image comprising a plurality of areas; each region is called a superpixel block; all the pixels of each super pixel block are equal to the pixel of the current super pixel block; extracting interested areas of the superpixel segmentation image, and extracting a plurality of interested areas; carrying out image binarization processing on each region of interest to obtain a binarized image; extracting a bone region from the binary image; extracting hyperbolic waves from the skeletal region; and detecting the underground target according to the waveform, amplitude intensity and time change characteristics of the hyperbolic wave.

Description

Underground target detection method and system based on ground penetrating radar hyperbolic wave fitting

Technical Field

The disclosure relates to the technical field of underground target detection, in particular to an underground target detection method and system based on ground penetrating radar hyperbolic wave fitting.

Background

The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.

Identification and fitting of hyperbolic waves in GPR images is very difficult. A Ground Penetrating Radar (GPR) is used as nondestructive detection equipment based on high-frequency electromagnetic waves, and when the electromagnetic waves propagate in an underground medium, hyperbolic-shaped reflection lines are generated when the electromagnetic waves meet an electrical difference boundary surface of an underground target, and the hyperbolic-shaped reflection lines are called hyperbolic waves. Important information such as the spatial position, the structure, the form and the buried depth of the underground target can be deduced through the characteristics such as the waveform, the amplitude intensity and the time change of the hyperbolic wave. This therefore has an important indicative role in understanding the characteristics of subsurface targets. However, due to system noise and inhomogeneity of the underground medium, the generated images are very complex, so that the extraction and fitting of the hyperbolic wave to the GPR image in real time is necessary.

Among the previous methods of identifying and fitting hyperbolic waves, a mathematical calculation method through non-machine learning and a calculation method through machine learning and a hybrid algorithm can be roughly classified. The non-machine mathematical calculation method is distinguished according to the difference of the characteristics of a hyperbolic wave region and other regions, and the calculation method through machine learning is performed by training through a positive sample containing the hyperbolic wave and a negative sample containing the non-hyperbolic wave, finally reducing the region including the hyperbolic wave, and further selecting a reasonable region.

In the course of implementing the present disclosure, the inventors found that the following technical problems exist in the prior art:

in the prior art, a method based on a generalized hough transform is used, but the method is very time-consuming and has high parameter requirements, because the hough transform needs to be performed in a four-dimensional space, and the precision of the hough transform depends on the discretization of parameters; in the prior art, a least square method is used, but the least square method can be regarded as an extension of generalized Hough transform, so that the method is not greatly improved. Chen and Cohn et al proposed a C3 sequence, scanned horizontally and then judged by the sequence, without taking into account the hyperbolic geometry. Collins L M and Torrione PA et al propose an idea of key point extraction. Fritze et al first used an edge adaptation algorithm for GPR images, but this algorithm has a certain requirement on the purity of the GPR images, and therefore the algorithm value is applicable to less noisy GPR images. In the prior art, the vertex of the hyperbola is detected, but other parameters of the relevant hyperbola are lost. However this is essential for identifying other properties of the object. zhou, chen et al propose a method of opening the column sections downward. The gradient in the y direction is calculated for the GPR image, a series of optimization is carried out, a characteristic column segment which is open downwards is searched, and then judgment and identification are carried out.

In the prior art, a method based on machine learning is used. The method comprises the steps of extracting features of a hyperbolic wave sample, combining a machine learning technology to reduce a region where the hyperbolic wave is located, and applying a fitting method to find out parameters of the hyperbolic wave to judge whether the position containing the waveform in the GPR image has a threat or not, wherein the result depends on the quality and quantity of training data.

Disclosure of Invention

In order to overcome the defects of the prior art, the disclosure provides a method and a system for detecting underground targets based on hyperbolic wave fitting of a ground penetrating radar;

in a first aspect, the present disclosure provides a method for detecting an underground target based on hyperbolic wave fitting of a ground penetrating radar;

the underground target detection method based on the hyperbolic wave fitting of the ground penetrating radar comprises the following steps:

acquiring a GPR image of the ground penetrating radar;

performing superpixel segmentation on the GPR image to obtain a superpixel segmented image comprising a plurality of areas; each region is called a superpixel block; all the pixels of each super pixel block are equal to the pixel of the current super pixel block;

extracting interested areas of the superpixel segmentation image, and extracting a plurality of interested areas; carrying out image binarization processing on each region of interest to obtain a binarized image;

extracting a bone region from the binary image;

extracting hyperbolic waves from the skeletal region;

and detecting the underground target according to the waveform, amplitude intensity and time change characteristics of the hyperbolic wave.

In a second aspect, the present disclosure also provides a system for detecting an underground target based on hyperbolic wave fitting of a ground penetrating radar;

underground target detecting system based on ground penetrating radar hyperbolic wave fitting includes:

an acquisition module configured to: acquiring a GPR image of the ground penetrating radar;

a superpixel splitting module configured to: performing superpixel segmentation on the GPR image to obtain a superpixel segmented image comprising a plurality of areas; each region is called a superpixel block; all the pixels of each super pixel block are equal to the pixel of the current super pixel block;

a region of interest extraction module configured to: extracting interested areas of the superpixel segmentation image, and extracting a plurality of interested areas; carrying out image binarization processing on each region of interest to obtain a binarized image;

a bone region extraction module configured to: extracting a bone region from the binary image;

a hyperbolic wave extraction module configured to: extracting hyperbolic waves from the skeletal region;

a subsurface target detection module configured to: and detecting the underground target according to the waveform, amplitude intensity and time change characteristics of the hyperbolic wave.

In a third aspect, the present disclosure also provides an electronic device comprising a memory and a processor, and computer instructions stored on the memory and executed on the processor, wherein the computer instructions, when executed by the processor, perform the steps of the method of the first aspect.

In a fourth aspect, the present disclosure also provides a computer-readable storage medium for storing computer instructions which, when executed by a processor, perform the steps of the method of the first aspect.

Compared with the prior art, the beneficial effect of this disclosure is:

firstly, performing superpixel segmentation on a GPR image under a real condition, obtaining a significant value of each superpixel block by using a multi-scale neighborhood calculation method, and obtaining a plurality of significant images by setting different thresholds. And finally, respectively binarizing the images, extracting a skeleton region in the binary images, finally selecting a region possibly containing the hyperbolic wave, and finally verifying to obtain an equation of the hyperbolic wave.

The result shows that the algorithm can well process the influence of various intersections and noises on the extraction, and can efficiently, accurately and quickly extract the information in the hyperbolic wave from the GPR image under the real condition. Experiments prove that the method is effective, and the algorithm is more robust than the previous algebraic distance algorithm.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the application and, together with the description, serve to explain the application and are not intended to limit the application.

FIG. 1 is a flow chart of a method of the first embodiment;

FIG. 2(a) is a hyperbolic wave of a first embodiment exhibiting a white dominance of grey-white-black-grey;

FIG. 2(b) is a hyperbolic wave of a first embodiment exhibiting a white dominance of grey-black-white-grey;

FIG. 2(c) is a black dominated hyperbolic wave exhibiting gray-black-white-gray for the first embodiment;

FIG. 3 is a diagram illustrating the encoding method of the 8-neighborhood region in the bone region extraction module according to the first embodiment;

fig. 4 is a diagram showing the corresponding effects when a is 10, 30 and 50, and e is 1.05, 1.5 and 2, respectively, when a hyperbolic template is constructed in the underground target detection module;

FIG. 5 is a section of 4 effective downwardly opening rows;

FIG. 6(a) is a schematic diagram showing the expansion of the effective point without waveform crossing in the first embodiment;

fig. 6(b) is a schematic diagram of the first embodiment of the hyperbolic wave intersecting with other curves in different directions (the left wing and other curves of the hyperbolic wave are all from top right to bottom left);

fig. 6(c) is a schematic diagram of the first embodiment of the hyperbolic wave crossing the other curves in the same direction (the left wing of the hyperbolic wave goes from top right to bottom left, and the other curves go from top left to bottom right);

FIG. 7 is a diagram illustrating matching criteria between valid clusters and templates according to the first embodiment;

FIG. 8(a) is an example of a hyperbolic wave to be identified and fitted for the first embodiment;

FIG. 8(b) shows the location of the preliminarily selected hyperbolic wave of the first embodiment;

fig. 8(c) shows the fitting effect of the hyperbolic wave of the first embodiment.

Detailed Description

It should be noted that the following detailed description is exemplary and is intended to provide further explanation of the disclosure. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs.

It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of example embodiments according to the present application. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.

Description of the terms:

SLIC, i.e. simple linear iterative clustering, is used in the present disclosure as a currently best superpixel segmentation algorithm, having the characteristics of few parameters, fast speed, compact and orderly partition.

The concept of saliency is proposed by itti et al. Since the primate's visual system is able to quickly identify salient regions in a scene in front of the eye, itti et al uses this idea in computer images so that salient regions in the image can be located. This has the advantage that the "region of interest" can be selected instead of analyzing the entire image. Later, the identification of the salient region is widely applied to image retrieval, adaptive content delivery, interested adaptive regions, and based on multiple aspects such as image compression and intelligent image adjustment, Achanta et al propose a local contrast comparison method under a multi-scale neighborhood, which can well segment an image, and thus is suitable for the identification of hyperbolic waves in a GPR image.

The waveform in the GPR image is linear, so the skeleton extraction algorithm can effectively extract the key information of the waveform or other noise regions in the region of interest, the key information simplifies the later judgment and is very effective for further removing the non-hyperbolic wave regions. Finally, all connected regions will become connected lines composed of lines by the skeleton extraction algorithm.

The classical algorithm of the opening and closing operation can make the edge points of the binary image smoother, so that the method is very suitable for the binary image obtained through significance calculation.

The method comprises the steps of firstly, providing a method for detecting underground targets based on hyperbolic wave fitting of a ground penetrating radar;

as shown in fig. 1, the method for detecting a subsurface target based on hyperbolic wave fitting of a ground penetrating radar includes:

s1: acquiring a GPR image of the ground penetrating radar;

s2: performing superpixel segmentation on the GPR image to obtain a superpixel segmented image comprising a plurality of areas; each region is called a superpixel block; all the pixels of each super pixel block are equal to the pixel of the current super pixel block;

s3: extracting interested areas of the superpixel segmentation image, and extracting a plurality of interested areas; carrying out image binarization processing on each region of interest to obtain a binarized image;

s4: extracting a bone region from the binary image;

s5: extracting hyperbolic waves from the skeletal region;

s6: and detecting the underground target according to the waveform, amplitude intensity and time change characteristics of the hyperbolic wave.

As one or more embodiments, the super-pixel segmentation is performed on the GPR image to obtain a super-pixel segmented image comprising a plurality of regions; the method comprises the following specific steps:

performing super-pixel segmentation on the GPR image by using an SILC super-pixel segmentation algorithm to obtain a super-pixel segmentation image comprising a plurality of areas; and clustering discrete adjacent points in the segmented super-pixel images, and finally realizing the effect of replacing the area with the central point.

As one or more embodiments, extracting regions of interest from the superpixel segmented image, and extracting a plurality of regions of interest; the method comprises the following specific steps:

s31: calculating a significant value of each super-pixel block of the super-pixel segmentation image;

wherein τ represents a differenceNeighborhood at scale, λ (τ) represents the weight of the saliency map computed at the neighborhood τ, A represents the number of superpixels within the neighborhood at the neighborhood τ,

representing the gray value of the center of the ith cell,

gray value representing the center of the j-th cell,. psi_iRepresenting the saliency value of the ith super-pixel region.

S32: setting a number of different thresholds;

the threshold values are respectively + -0.5, + -0.65, + -1.3, + -1.5, + -1.8, and + -2.25 times the mean of the corresponding saliency values of all superpixel blocks;

s33: from several different thresholds, several saliency maps are obtained:

for the positive number threshold value, changing the super pixel blocks with the significant values larger than the positive number threshold value into the foreground, and changing the super pixel blocks with the significant values smaller than the positive number threshold value into the background;

for a negative number threshold value, changing a super-pixel block with a significant value smaller than the negative number threshold value into a foreground, and changing a super-pixel block with a significant value larger than the negative number threshold value into a background;

and then a plurality of saliency maps are obtained.

As one or more embodiments, the neighborhood selection mode of the neighborhood τ includes a selection mode of a circular neighborhood and a selection mode of a column neighborhood; wherein the content of the first and second substances,

the selection mode of the circular neighborhood is as follows: given a radius R, the center of each superpixel block is taken as the center of a circle, for each center c_iSelect all center sets { c } within the set of centers in the circular neighborhood τ₁,c₂,c₃...c_j}：

D(c_i,c_j)<R；(8)

Wherein D (c)_i,c_j) Denotes c_iAnd c_jThe distance between the two at a location in the image.

Column neighborhood selected squareThe formula is as follows: selecting the center point to be in the interval [ x ]_min,x_max]Of all superpixel blocks in (1), where x_minRepresenting the left boundary, x, of a super-pixel block_maxRepresenting the right boundary of the superpixel block.

As one or more embodiments, the extracting of the bone region from the binarized image; the method comprises the following specific steps: and using a skeleton extraction algorithm to represent all connected regions in the saliency map by linear bones to acquire a bone image.

As one or more embodiments, the extracting hyperbolic waves from a skeletal region; the method comprises the following specific steps:

defining a matching set, pre-screening hyperbolic wave vertexes in the skeleton image, judging and fitting the hyperbolic wave by using the matching set, and extracting the hyperbolic wave.

Further, the hyperbolic vertex is pre-screened in the bone image; the method comprises the following specific steps: and screening a line segment with a downward opening, and taking the midpoint of a flat area in the rising-flat-falling structure as the top point of the pre-screened hyperbolic wave from the line segment with the downward opening.

Further, the hyperbolic waves are judged and fitted by using a matching set, and the hyperbolic waves are extracted; the method comprises the following specific steps:

for each vertex, among all branches, one that most matches the matching set is found as the waveform for that vertex.

In practical situations, due to the influence of noise or due to the measured distances between multiple objects being too close, the waveforms in the generated GPR image will have many kinds of overlaps, such as an overlap of a hyperbolic wave and a hyperbolic wave, an overlap of a hyperbolic wave and a clutter, an overlap of a clutter and a clutter, and the like. However, these waveforms may remain in the saliency calculation algorithm, presenting various intersections. The various paths formed by these intersections are referred to as branches.

It should be appreciated that the SILC super pixel segmentation algorithm is applied to a single channel GPR image. The advantage of using superpixels to segment the source image is that discrete adjacent points in the image can be clustered, eventually achieving the effect of a central point replacing a region. For example, a 500 × 1000 GPR image contains 500000 ten thousand pixels, and after superpixel segmentation is performed at step size 20, only at most 25 × 50 regions are 1250 regions, and the pixel values of the points in these regions are very similar, so that at most 1250 central points can be used to replace these 1250 regions, which obviously greatly reduces the amount of calculation and greatly eliminates the influence of discrete noise on image classification.

The concept of the SILC algorithm using KMEANS clustering is to initialize K centers for all data first and then plan all data to the nearest center point:

kmeans:minD(data⁽ⁱ⁾-μ_k)(1)

the process is iterated successively to obtain the corresponding super-pixel segmented image. Wherein, the data⁽ⁱ⁾Denotes the ith data, μ_kDenotes the kth cluster center and D denotes the distance between the two in a given algorithm, usually in euclidean terms. When the SILC super-pixel segmentation algorithm is applied to a single-channel GPR image, the data⁽ⁱ⁾Described as a three-dimensional vector: data⁽ⁱ⁾＝[G_i-x_i-y_i]The specific method comprises the following steps:

dividing the GPR image into a plurality of square blocks according to a certain step length S, and taking the gray level of the central point of each block and a vector formed by two corresponding coordinates as the initialization center of the superpixel block. A set of initialization centers is thus obtained: c ═ C₁,C₂,...,C_{(m*n)/stride2}},C_i＝[G_i,x_i,y_i]Wherein G is_iRepresenting the pixel value, x, at the center point_i,y_iRepresenting the coordinates.

Such a regular partitioning method may result in the center point being at an edge or noise point, so to prevent the point at which the center is at from being an unreasonable point such as an edge and noise, the center must be moved to the point of minimum gradient in the 3 × 3 neighborhood:

wherein the content of the first and second substances,

the gradient value at the pixel point (i, j) is represented and can be calculated by using a sobel operator. Consider the gradient as a two-dimensional vector of gradients in both x and y directions:

defining gradient values:

wherein P (x, y) is the corresponding pixel value at pixel point (x, y),

is a matrix formed by 8-neighborhood pixel values at pixel point (x, y).

In the superpixel algorithm, the distance between two points is defined by a three-dimensional vector [ G, x, y]^TCalculated by a weighted Euclidean distance, the calculation method is characterized by taking the value and the position of a pixel point into consideration:

where γ is a balance parameter, and its size means the size of the impact of the distance of the position on the cluster, and S is the step size.

For a general GPR image containing hyperbolic waves, the effect of 10 iterations is enough to meet the requirement of segmentation. However, after the process is finished, many isolated points or pixel blocks with small areas are generated, and the isolated points or the pixel blocks need to be assigned to the maximum connected label.

Finally, a super-pixel picture is obtained, wherein the super-pixel picture comprises a plurality of super-pixel units:

U＝{U₁,U₂,...,U_n},N≤(m*n)/stride²(6)

selecting an interested area:

it is readily apparent from the GPR image that a region containing a waveform is significantly different from other regions in that the region containing the waveform is color-significant in the y-direction (or one of the significant regions, as it may contain several waveforms), and in addition, the region containing the waveform will be significant within its circular neighborhood. In general, hyperbolic waves in a GPR image have two closely adjacent regions of black and white saliency, but the distribution and the position of the black saliency and the white saliency are different in different images. The hyperbolic wave regions in fig. 2(a) and 2(b) appear more distinct and more continuous in white but they are also distinct in that the hyperbolic wave regions in fig. 2(a) appear to appear with the background on the y-axis top-down as a sequence of gray-white-black-gray, whereas the hyperbolic wave regions in fig. 2(b) appear as a sequence of gray-black-white-gray; correspondingly, the image in fig. 2(c) appears with more distinct and continuous black areas and presents a sequence of gray-black-white-gray. Therefore, it is not sufficient to consider only a white salient region or a black salient region among the GPR images, and therefore the method in the present disclosure will be considered for both black and white cases.

In the prior art, people calculate and sum saliency maps with different scales in h/2, h/4 and h/8 neighborhoods of pixel points of a color image. Therefore, the method is suitable for multi-scale neighborhood summation of the GPR image. In this way, hyperbolic wave regions with different textures and colors in the GPR image can be extracted.

Therefore, in this section, a new, fast and efficient method for calculating a multi-scale neighborhood is proposed, which first provides a combination of a column neighborhood and a ring neighborhood for a GPR image subjected to superpixel segmentation, calculates the significance of each superpixel block, and then segments the image by setting different thresholds of positive and negative, and different sizes, where the positive and negative differences represent differences in black and white. The method well solves the problem that voids appear after the GPR image is binarized due to the overlapping of noise and hyperbolic waves in the traditional threshold value calculation method, so that the image can be well segmented, and experiments prove the effectiveness of the algorithm. The region preliminarily selected by the algorithm has the characteristics of tidy edges, few redundant regions, continuous waveform and no holes.

In the previous algorithm, some people, such as Xiren Zhou et al, first optimize the GPR image by using a tool such as Matgpr, and then calculate the gradient in the y direction, and then calculate the mean of the gray values of all the pixel point sets with the gradient not being 0 as a threshold to binarize the image. However, simply relying on the high and low pixel values to achieve binarization of an image can result in too coarse an image segmentation; the method proposed by Qingxu Dou et al, first of all, performs mean filtering denoising on a GPR image, then subtracts the mean value of the line set from each line to eliminate ground reflection, and uses an edge detection to binarize the image, and finally selects edge points. The algorithm does not select the whole image, but selects the edge points. However, it is worth noting that this is done although it is possible to effectively segment the relatively clean GPR image. However, the actual GPR image may contain a lot of noise and therefore many edge points, which may eventually result in an excessive region of interest, which may cause difficulties for further operations. Therefore, we introduce a view of local significant computation, which is to select the region of interest under multi-scale conditions and to threshold the black waveform and the white waveform separately. The problems are effectively avoided.

The preliminary selection of the interested region is obtained by calculating the neighborhood distance of a pixel block formed by dividing the superpixel under multiple scales. The specific method comprises the following steps: for the GPR image which is already subjected to superpixel segmentation, the method is used for each unit U_iDefining the significance value:

where τ represents the neighborhood at different scales, λ (τ) represents the weight of the saliency map computed at the neighborhood τ, A represents the number of superpixels in the neighborhood at the neighborhood τ,

representing the gray value of the center of the ith cell,

the reason why the gray value representing the center of the j-th cell, where the differences of pixel values are used for summation rather than taking the absolute value, is that the black and white waveforms can be separated by the value ψ after the region where the black waveform is located is operated_i>0 and the area psi where the white waveform is located_i<0。

The reason for performing the distance calculation at different scales is to prevent the image calculation at a single scale from being haphazard, resulting in generation of errors.

However, due to the inherent irregularity of superpixels, the choice of neighborhood is very difficult, and the combination of column and circular neighborhood is used in the present disclosure, since the column in which the hyperbolic region is located and the block in the adjacent region are significant as mentioned above.

Given the radius R, for each center ci, select all center sets { c1c2 … cj } within the center set within the circular neighborhood τ:

D(c_i,c_j)<R(8)

where D represents the distance between the two at a location in the image. And for the y direction, all center points are selected to be in the interval [ y_min,y_max]Of (2), wherein y_minRepresenting a super-pixel block c_iLeft boundary of, y_maxRepresenting a super-pixel block c_iThe right border of (a).

Converting into a binary image, and removing holes and noise points by using an opening and closing algorithm:

since the division of the superpixels may cause the remaining regions to contain some holes or burrs, it is necessary to process the binary image using a switching algorithm. The principle of the on-off operation is to use the check image to perform an and operation, wherein the on operation is to perform an and operation and then perform an or operation on the image and the structural element, and the off operation is to perform an or operation and then perform an and operation to complete:

wherein the content of the first and second substances,

extracting a bone region from the binarized image: since the final hyperbolic fit is line-based, the binary image obtained in the above step is still an irregular region. In order to simplify the calculation, the saliency maps obtained by different thresholds can be represented by using a skeleton extraction algorithm, wherein all connected regions are represented by linear skeletons. This can simplify the amount of calculation and can facilitate the fitting algorithm described later. The specific process is shown in fig. 3.

Setting and fitting of a matching set: in this section, a new method of determining and fitting a hyperbola is proposed. Firstly, the definition of a matching set is given; then, a method for pre-screening hyperbolic wave peaks in the skeleton image is provided; finally, a method for judging and fitting the hyperbolic wave by using the set is provided.

Definition of matching set:

for the hyperbolic fitting problem, hough transform is a common method. However, we know that the corresponding equation for hyperbolic wave to appear in the GPR image should be:

if the conventional hough transform method is used, in order to obtain the parameter x₀、t₀In [ x, y, a, b ] is required]^TThe calculation is performed in the four-dimensional space. But if the coordinates of the vertices of the hyperbolic wave are known, the problem seems to be simpler. To match a known template, the vertices are first moved to the origin, which is known from the definition of the hyperbola, which is

When the vertex of (2) is moved to the coordinate vertex, the expression is

Obviously, a hyperbolic equation with known vertex coordinates has only two unknown quantities, a and b.

The setting of the template takes the actual situation into consideration, and the speed can be determined within a certain range because the image of the GPR is obtained by manually pushing the GPR at a constant speed, however, the eccentricity e can well measure the speed v. If t is considered a function of x, the hyperbolic expression moved to the origin is:

as shown in equation (12), the eccentricity e can reflect the "steep" condition of the hyperbola, and according to the mathematical definition, in the x-t image, the derivative of each point in the curve is the velocity v, so in the matching set, e (1.05,2) is set and the step size is 0.05. Corresponding v ∈ (0.81,4.42) to represent the situation in the real world. As shown in fig. 4, the eccentricity e is 1.05, 1.5, and 2 when a is 10, 30, and 50, respectively. Essentially, the data set is actually a set corresponding to a number pair consisting of parameters a and b in a hyperbolic equation, a hyperbola in a GPR image according to real conditions is in GHMS, the value of a is taken as [10,50], and the value of b is obtained according to eccentricity, namely

Thus, we obtained a total of 20 × 41 — 820 hyperbolic wave models.

FIG. 4 shows the results when a is 10, 30, 50 and the eccentricity e is 1.05, 1.5, 2. It can be seen that (1) when the eccentricity e is determined, the larger the value of a is, the more smooth the hyperbola is, (2) the eccentricity can well reflect the advancing speed of the ground penetrating radar during working, and the larger the eccentricity is, the slower the speed is.

It can be seen that when the value of a is determined, the eccentricity e can well reflect the moving speed v, and when the value of the eccentricity e is determined, the value of a can reflect the fullness degree of the hyperbolic wave form, so that the fitting set can well correspond to the real situation.

Pre-screening hyperbolic wave peaks in skeleton images:

since each line segment with downward opening can be easily determined after generating the bone image, since a point-segment matching can be performed at all points in the generated bone image, the matching success template should be divided into 4 kinds, which will be shown in fig. 5: from a-d, 4 line segments with downward opening characteristics are shown, all hyperbolic wave vertexes can be obtained by selecting all segments according with the characteristics in a skeleton image, and the middle point of a flat area in 4 rising-flat-falling structures is taken as the preliminarily screened hyperbolic wave vertex.

Identification and fitting of hyperbolic waves: in practical situations, due to the influence of noise or due to the measured distances between multiple objects being too close, the waveforms in the generated GPR image will have many kinds of overlaps, such as an overlap of a hyperbolic wave and a hyperbolic wave, an overlap of a hyperbolic wave and a clutter, an overlap of a clutter and a clutter, and the like. However, these waveforms are likely to remain in the significance calculation algorithm, and thus the advantage of extracting bone regions is very clear. When a vertex is found, matching operation can be performed on all branch conditions of a connected region corresponding to the vertex, then a threshold value P is set, and a region with the matching degree above the threshold value is regarded as a hyperbolic wave. Here however we need to consider three issues:

(1) since the sizes of the hyperbolic waves are different among different GPR images, the threshold value should be related to the sizes of the hyperbolic waves, in other words, should be a matching ratio of the effective points rather than the number.

(2) A region of communication may contain hyperbolic waves, and therefore several hyperbolic waves must be distinguished;

(3) neighboring points may all have a high degree of matching and therefore cannot be fitted repeatedly.

In the previous method, it is quite difficult if the regions are separated.

A thresholding method is provided, all branch path conditions corresponding to each vertex are subjected to parameter matching, specifically, for each vertex, one of all branches which is most matched with a fitting set is found as a waveform of the vertex, but the number of points of the branch is enough, otherwise, the waveform cannot be represented well. And finding the corresponding template which is most consistent with the sub-path from all the templates, and if the matching degree of the template with the highest matching degree still cannot meet the threshold value, judging that the region is not a hyperbolic wave region.

And (3) carrying out a matching algorithm on each point of the preliminarily screened connected region with the downward opening, namely finding the vertex of the hyperbolic wave, establishing a set α of effective points, and expanding the effective points in pairs in the process of expanding the effective points to the two wings of the effective cluster, so that the characteristic of hyperbolic wave axial symmetry is met.

As shown in FIGS. 6(a) -6 (c) where gray indicates a point that has been added to set α, when going to expand from point L towards N, the final set is (A-M) since point M in the same row in the right wing will not continue to expand, as shown in FIG. 6(b), when expanding to point K, there will be 3 ways of expansion H, M, N but point H and point N will obviously not follow the directional nature of the hyperbola and therefore be excluded, i.e. only downward, leftward or downward to the left, in accordance with the nature of the hyperbola, in FIG. 6(c), when the expansion node is G, two points F and I will be selected, but if F is the expansion node, it will only expand to point I, so in this case, point F will be discarded, and path G-I-K will be selected, but the problem will not seem as simple as if point K is the expansion node, since it can expand to points M and N and they can both expand to the highest branch, respectively, the decision of the highest branch and the highest branch of the path of the points are taken.

Since branches may be generated during the expansion process and corresponding branches may also be generated, each vertex may have multiple valid paths α₁,α₁,…,α_n}. Therefore, only the matching degree operation needs to be performed on the paths with the left and right tails long enough in all the effective paths and the matching set, and therefore:

where Δ y is the difference in y direction for each column in the image when the vertex of the template and the vertex of the line segment opening downward are superimposed, and num is the number of points of the effective path.

In this section, experiments are presented that were performed on data collected under real conditions and some comparisons of the methods presented in this disclosure with other methods.

A. Experiment of real data set

The tested data are from the very complicated situation of cement land, sand beach, even grassland, stone road and the like acquired by GPR of GSSI-2000 model under the real condition, and 80 GPR images acquired from the situation are various and have very weak difference between the hyperbolic wave with great difference in black and white and the background, and even contain a great amount of noise and cross. By manual screening, we obtained 204 of the valid hyperbolic waveforms in total. Of these GPRs, a small fraction of the images are clear and well-waved, but most contain a large amount of hyperbolic crossing events and noise, which are used as samples for experiments.

The method divides the image by adopting the superpixel with the step size of 20, and obtains 12 saliency maps, the threshold values of the saliency maps are respectively +/-0.5, +/-0.65, +/-1.3, +/-1.5, +/-1.8 and +/-2.25 times of the mean value of the corresponding saliency values of all the superpixel blocks, and the template matching is carried out after the preliminary screening is carried out when the tail length reaches 30. A detailed result obtained by our method is shown in fig. 7.

To better describe the accuracy of our method in identifying hyperbolas, we use the euclidean distance in the y-direction as an evaluation index. Before this we have manually labeled the hyperbolic waves in the images in GPR and derived the exact equations. We define the similarity between the resulting hyperbolic equation p1 and the standard hyperbolic equation p2 as

Wherein the distance d is defined as the ratio of the distance between two points in the y-direction and the height of the hyperbolaAverage value. See fig. 7 for a detailed description. If the fitted curve is a solid line and the artificially labeled curve is a dashed line in the figure, as shown in fig. 7, it is assumed that the height of the defined valid hyperbolic wave region is h and the height is equal to x_mNext, if the coordinates of both in the y direction are p, respectively₁:(x_m,y_m) And p_2:(x_m,y_n) I.e. the difference in height distance is Δ h ═ y_m-y_nThen, define:

because the OSCA algorithm and the C3 algorithm can both obtain the hyperbolic-like cluster through graphic calculation, the OSCA algorithm and the C3 algorithm are compared. One experimental result on an actual data set using our method is shown in fig. 7, where 4 valid clusters are given. However, using the OSCA algorithm and the C3 algorithm yields much more valid clusters than this result. To demonstrate the superiority of our method more intuitively and effectively, we carried out experiments dividing 80 images into 4 groups. Under the condition that each picture is guaranteed to contain at least one hyperbolic wave, the result obtained by calculating the average fitting rate is obtained.

In all experiments, there were few or no hyperbolic targets missed. However, although this algorithm contains a few redundant targets, it is known that in practical situations this helps to reduce the risk and avoid accidents. It can be seen therefore that our proposed method is very secure and trusted. B. Comparison of our method with other methods

To better validate the model, we also performed some comparative experiments. We compared our method with the C3 column connection clustering algorithm and the OSCA algorithm. Our method, as well as the C3 algorithm and the OSCA algorithm, can obtain a hyperbola by unsupervised processing of the ground penetrating radar image. It is known that the OSCA algorithm performs matching of the downward opening column segment, whereas the C3 algorithm does not. In addition, our method also performs template matching for judgment, but the OSCA, C3 algorithm does not perform similar judgment. We evaluated the above three methods on the real dataset after corresponding pre-processing. We count the number of valid clusters and CPU time separately to measure the performance of the method. The results for some real datasets are shown in fig. 8(a), 8(b) and 8 (c).

It can be seen that the OSCA processing time is better than that of the C3 algorithm because C3 requires each connected region in the horizontal scan image, and the processing time of both is longer than that of our method because the number of regions to be processed is greatly reduced after superpixel segmentation by our method, and finally the processes of screening and fitting are performed simultaneously, and only dozens of points are compared between each vertex by extraction of the bone region, so it is very efficient.

Another important task is to compare the error rate of fitting between various methods, and we respectively use our method and Hough transform method and RADF method to make comparison, i.e. after our template matching step, the regions judged to be yes are respectively fitted using Hough transform and RADF method. The correct recognition rate and the CPU time are given in graphs I and II.

In the case of preprocessing a cluster of points, a hyperbolic equation is obtained, and thus the corresponding parameters of the object, such as depth and radius, are obtained. With the Hough transform taking the longest time. Because the method only needs to match dozens of pixel points with the template after finding the vertex of the hyperbola, a good time effect can be obtained. While the other two methods have to process the region, require computation of a large number of points and are not robust enough in noise resistance, and therefore are more complex in computation time and low in recognition rate. According to experiments, Hough transformation needs to be calculated under four dimensions, so that time complexity is too high, compared with the RADF method, the method has the advantages that the identification rate is basically the same, the rate is higher than that of the RADF method, and the time complexity is lower, so that the method can better identify and fit the hyperbolic wave in real time.

The method comprises the steps of extracting a GPR image salient region, separating the salient region under different thresholds after extraction, extracting a connected skeleton region, extracting a vertex, and then performing matching processing by using a matching set. The effectiveness of the method is proved by experiments.

In the proposed method, the region of interest is computed from the neighborhood under multiple scales and the final matching algorithm is the two core parts. The geometrical characteristics of the hyperbolic wave are considered by the neighborhood calculation in the previous stage under multiple scales and the final matching algorithm, so that the algorithm has stronger robustness.

The second embodiment of the invention also provides an underground target detection system based on ground penetrating radar hyperbolic wave fitting;

In a third embodiment, the present embodiment further provides an electronic device, which includes a memory, a processor, and computer instructions stored in the memory and executed on the processor, where the computer instructions, when executed by the processor, implement the steps of the method in the first embodiment.

In a fourth embodiment, the present embodiment further provides a computer-readable storage medium for storing computer instructions, and the computer instructions, when executed by a processor, perform the steps of the method in the first embodiment.

The above description is only a preferred embodiment of the present application and is not intended to limit the present application, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, improvement and the like made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. The underground target detection method based on the hyperbolic wave fitting of the ground penetrating radar is characterized by comprising the following steps of:

acquiring a GPR image of the ground penetrating radar;

extracting a bone region from the binary image;

extracting hyperbolic waves from the skeletal region;

2. The method according to claim 1, wherein the GPR image is superpixel segmented to obtain a superpixel segmented image comprising a plurality of regions; the method comprises the following specific steps:

3. The method as claimed in claim 1, wherein the extracting of the interested region is performed on the superpixel segmentation image, and a plurality of interested regions are extracted; the method comprises the following specific steps:

s32: setting a number of different thresholds;

s33: several saliency maps are derived from several different thresholds.

4. The method as claimed in claim 1, wherein said extracting a bone region from the binarized image; the method comprises the following specific steps: and using a skeleton extraction algorithm to represent all connected regions in the saliency map by linear bones to acquire a bone image.

5. The method of claim 1, wherein said extracting hyperbolic waves from a bone region; the method comprises the following specific steps:

6. The method of claim 5, wherein the hyperbolic vertices are pre-screened in the bone image; the method comprises the following specific steps: and screening a line segment with a downward opening, and taking the midpoint of a flat area in the rising-flat-falling structure as the top point of the pre-screened hyperbolic wave from the line segment with the downward opening.

7. The method of claim 5, wherein said fitting hyperbolic wave using a set of matches is a decision fit to extract the hyperbolic wave; the method comprises the following specific steps:

8. Underground target detection system based on ground penetrating radar hyperbolic wave fitting, characterized by includes:

9. An electronic device comprising a memory and a processor and computer instructions stored on the memory and executable on the processor, the computer instructions when executed by the processor performing the steps of the method of any of claims 1 to 7.

10. A computer-readable storage medium storing computer instructions which, when executed by a processor, perform the steps of the method of any one of claims 1 to 7.