CN113608192A - Ground penetrating radar far field positioning method and device and computer readable storage medium - Google Patents

Ground penetrating radar far field positioning method and device and computer readable storage medium Download PDF

Info

Publication number
CN113608192A
CN113608192A CN202110909174.XA CN202110909174A CN113608192A CN 113608192 A CN113608192 A CN 113608192A CN 202110909174 A CN202110909174 A CN 202110909174A CN 113608192 A CN113608192 A CN 113608192A
Authority
CN
China
Prior art keywords
matrix
receiving
array
model
matrix model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110909174.XA
Other languages
Chinese (zh)
Other versions
CN113608192B (en
Inventor
李建中
林旖睿
刘远
詹瑞典
熊晓明
蔡述庭
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.)
Guangdong University of Technology
Original Assignee
Guangdong University of Technology
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 Guangdong University of Technology filed Critical Guangdong University of Technology
Priority to CN202110909174.XA priority Critical patent/CN113608192B/en
Publication of CN113608192A publication Critical patent/CN113608192A/en
Application granted granted Critical
Publication of CN113608192B publication Critical patent/CN113608192B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/411—Identification of targets based on measurements of radar reflectivity
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/885—Radar or analogous systems specially adapted for specific applications for ground probing
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
    • G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
    • G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
    • G01S7/417—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section involving the use of neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization

Abstract

The application discloses a ground penetrating radar far-field positioning method, a device and a computer readable storage medium, wherein a receiving matrix model based on a uniform circular array manifold matrix converts a pattern space into an array manifold matrix capable of meeting a Van der Menu structure through a pattern space conversion method, then the array manifold matrix is divided into a plurality of mutually overlapped sub-arrays through a forward space smoothing method, a covariance matrix of the receiving matrix model is obtained through calculating a data covariance matrix of each sub-array, and finally a machine learning and deep convolution neural network method is introduced to obtain a hyper-parameter and a spectral peak screening threshold value.

Description

Ground penetrating radar far field positioning method and device and computer readable storage medium
Technical Field
The present disclosure relates to the field of ground penetrating radar technologies, and in particular, to a far-field positioning method and apparatus for a ground penetrating radar, and a computer readable storage medium.
Background
Ground Penetrating radar (gpr) is a geophysical method for detecting the characteristics and distribution rules of substances inside a medium by using antennas to transmit and receive high-frequency electromagnetic waves. The method realizes large-scale measurement by utilizing the transmission and the reception of electromagnetic waves, has the depth ranging from several centimeters to dozens of meters, is a nondestructive detection technology, has the advantages of high detection speed, continuous detection process, high resolution, convenient and flexible operation, low detection cost and the like compared with other conventional underground detection methods, and is very suitable for the rapid nondestructive detection of roads. When the required detection resolution ratio breaks through the rayleigh limit (the estimated angle resolution of the signal source signal is very small due to the limitation of the physical size of each antenna), the hardware performance of the ground penetrating radar cannot meet the requirement, so that a super-resolution signal source positioning algorithm in signal processing needs to be used for analyzing and processing the reflected signal, and the nondestructive detection of the super-performance defect is realized.
In the field of signal processing research, array signal processing is a very important research direction. A plurality of sensors called array elements are placed at different positions in space according to a certain rule to form a certain structure, so that space domain signals are sampled, and further array receiving signals are further processed by combining space information and time information of sampled data. Due to the fact that the sensors are arranged, the array signal processing has the advantages of being high in signal gain, capable of flexibly controlling wave beams, strong in spatial resolution capability and the like. The most common array structures in direction-finding systems are uniform linear arrays and uniform circular arrays. The linear array can carry out one-dimensional angle direction measurement from-90 degrees to 90 degrees, but the direction measurement range cannot cover all directions (360 degrees). And in a scene needing to obtain a two-dimensional angle, the linear array cannot simultaneously measure an azimuth angle and a pitch angle. Compared with a linear array, the uniform circular array has the advantages that the direction-finding range covers all directions, two-dimensional direction-finding values of an azimuth angle and a pitch angle can be obtained simultaneously, the direction-finding precision is uniform, the aperture is small, no image blurring exists and the like, and therefore the direction-finding algorithm based on the uniform circular array has a larger and wider effect in actual engineering.
Although high-resolution far-field signal source positioning algorithms have been applied to ground penetrating radars and exhibit resolution and accuracy which break through the limitations of the hardware itself, the above methods can only be used in far-field scenarios with a defective amount of prior knowledge. In practical application scenarios, the number of defects is unknown, and the use of the super-resolution algorithm often requires the introduction of a number estimation algorithm in advance. However, the introduction of the quantity estimation algorithm makes the positioning algorithm redundant on one hand, reducing real-time performance, and on the other hand, in case of inaccurate quantity estimation, the positioning algorithm will fail completely.
Disclosure of Invention
The application provides a ground penetrating radar far-field positioning method and device and a computer readable storage medium, and solves the technical problems that redundancy exists in the existing quantity estimation algorithm, the real-time performance is reduced, and positioning is prone to complete failure.
In view of the above, a first aspect of the present application provides a method for far-field positioning of a ground penetrating radar, the method including:
step S101, constructing a receiving matrix model Y of an antenna receiving array for receiving road surface defect reflection signals, wherein Y represents the road surface defect reflection signals received by an array element, A represents a direction response vector, S represents a signal source incident signal, N represents array noise, and the antenna receiving array is a uniform circular array manifold matrix;
step S102, converting the uniform circular manifold matrix of the receiving matrix model into an array manifold matrix by a mode space conversion method;
step S103, dividing the array manifold matrix of the receiving matrix model into a plurality of mutually overlapped sub-matrices by utilizing a forward space smoothing method;
step S104, determining a covariance matrix of an array manifold matrix of the receiving matrix model according to the data covariance matrix of each subarray;
s105, acquiring hyper-parameters of the receiving matrix model based on a machine learning method;
step S106, constructing a subspace according to the covariance matrix of the array manifold matrix of the receiving matrix model and the hyperparameter, wherein the subspace is full rank and is orthogonal to the direction response vector;
s107, acquiring a spectral peak screening threshold value of the receiving matrix model based on a deep convolutional neural network method;
and S108, defining a space spectrum according to the subspace, the spectrum peak screening threshold and the direction response vector, and searching the frequency spectrum of the space spectrum, wherein the searched angle information corresponding to the peak is the positioning information of the road surface defect.
Optionally, in step S101, a ═ a (θ)1,Φ1),a(θ2,Φ2),…,a(θk,Φk)]TAnd k is the number of the signal sources, theta is the azimuth angle of the incident signals of the signal sources, and phi is the direction angle of the incident signals of the signal sources.
Optionally, the step S102 specifically includes:
multiplying the transformation matrix B and two sides of the receiving matrix model Y which is AS + N at the same time, so that the uniform circular manifold matrix of the receiving matrix model is converted into an array manifold matrix;
the transformation matrix is specifically:
Figure BDA0003202828650000031
wherein, Jh(beta) is a first Bessel function of h stage, h is a mode serial number of a mode space, and h is approximately equal to beta;
the receive matrix model transforms to:
Figure BDA0003202828650000032
wherein the content of the first and second substances,
Figure BDA0003202828650000033
t generationAnd the time of the road surface defect reflected signal is collected by a meter.
Optionally, the kth sub-array in step S103 is:
Figure BDA0003202828650000034
wherein the content of the first and second substances,
Figure BDA0003202828650000035
optionally, the step S104 specifically includes:
according to the data covariance matrix of each subarray
Figure BDA0003202828650000036
Determining a covariance matrix of an array manifold matrix of the receive matrix model as
Figure BDA0003202828650000037
Where H represents the conjugate transpose of the matrix.
Optionally, the step S105 specifically includes:
constructing a first training data set and a first testing data set;
training a preset machine learning model through the first training data set until a loss function of the preset machine learning model reaches an optimal value to obtain optimal parameters of the preset machine learning model;
and inputting the signal-to-noise ratio of the road surface defect reflection signal of the receiving matrix model into a preset machine learning model of the optimal parameter to obtain the hyper-parameter of the receiving matrix model.
Optionally, step S107 specifically includes:
constructing a second training data set and a second testing data set;
training a preset deep convolutional neural network through the second training data set until a loss function of the preset deep convolutional neural network reaches an optimal value to obtain optimal parameters of the preset deep convolutional neural network;
and inputting the road surface defect reflection signals of the receiving matrix model into a preset depth convolution neural network of the optimal parameters to obtain a spectral peak screening threshold value of the receiving matrix model.
Optionally, the spatial spectrum in step S108 specifically includes:
Figure BDA0003202828650000041
wherein Un is a subspace, and Un ═ I-Rf (Rf)HRf+μI)-1RfHI is the identity matrix, μ is the hyperparameter, and a is the column vector of the directional response vector a.
The second aspect of the present application provides a far-field positioning device for ground penetrating radar, the device comprising:
the first construction unit is used for constructing a receiving matrix model Y of an antenna receiving array for receiving the road surface defect reflection signals, wherein Y represents the road surface defect reflection signals received by an array element, A represents a direction response vector, S represents a signal source incident signal, N represents array noise, and the antenna receiving array is a uniform circular array manifold matrix;
a transformation unit for transforming the uniform circular manifold matrix of the receiving matrix model into an array manifold matrix by a mode space transformation method;
the processing unit is used for dividing the array manifold matrix of the receiving matrix model into a plurality of mutually overlapped sub-matrices by utilizing a forward space smoothing method;
the calculation unit is used for determining a covariance matrix of an array manifold matrix of the receiving matrix model according to the data covariance matrix of each sub-matrix;
a first obtaining unit, configured to obtain a hyper-parameter of the receiving matrix model based on a machine learning method;
a second constructing unit, configured to construct a subspace according to a covariance matrix of the array manifold matrix of the receive matrix model and the hyper-parameter, where the subspace is full rank and is orthogonal to the directional response vector;
the second acquisition unit is used for acquiring a spectral peak screening threshold value of the receiving matrix model based on a deep convolutional neural network method;
and the positioning unit is used for defining a space spectrum according to the subspace, the spectrum peak screening threshold and the direction response vector, carrying out spectrum search on the space spectrum, and obtaining the angle information corresponding to the searched peak as the positioning information of the road surface defect.
A third aspect of the present application provides a computer-readable storage medium for storing program code for executing the method for far-field localization of georadar according to the first aspect.
According to the technical scheme, the embodiment of the application has the following advantages:
the method comprises the steps of converting a mode space into an array manifold matrix capable of meeting a Van der Monte structure through a mode space conversion method based on a receiving matrix model of a uniform circular array manifold matrix, dividing the array manifold matrix into a plurality of mutually overlapped sub-arrays by utilizing a forward space smoothing method, obtaining a covariance matrix of the receiving matrix model by calculating a data covariance matrix of each sub-array, finally obtaining a hyper-parameter and a spectral peak screening threshold value by introducing a machine learning and deep convolution neural network method, and determining positioning information of a received pavement defect reflection signal by utilizing a spatial spectrum.
Drawings
Fig. 1 is a flowchart illustrating a method for far-field positioning of a ground penetrating radar in an embodiment of the present application;
FIG. 2 is a schematic structural diagram of a far-field positioning device of a ground penetrating radar in an embodiment of the present application;
FIG. 3 is a diagram illustrating the operation of the ground penetrating radar in monitoring a road surface according to the embodiment of the present application;
fig. 4 is a schematic diagram of a model of a uniform circular array manifold matrix in an embodiment of the present application.
Detailed Description
The embodiment of the application provides a ground penetrating radar far-field positioning method, a ground penetrating radar far-field positioning device and a computer readable storage medium, and solves the technical problems that redundancy exists in an existing quantity estimation algorithm, real-time performance is reduced, and positioning is prone to complete failure.
In order to make the technical solutions of the present application better understood, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The ground penetrating radar far field positioning method, the ground penetrating radar far field positioning device and the computer readable storage medium are designed, road defects are generated inside the ground penetrating radar far field positioning device and are not easy to directly observe, and the ground penetrating radar is widely applied to the engineering of road nondestructive testing due to the characteristics of being nondestructive and capable of being rapidly collected. And because of the limitation of the hardware condition of the ground penetrating radar, a signal source positioning algorithm in signal processing needs to be used to ensure that the detection precision breaks through the Rayleigh limit. However, in the existing signal source positioning algorithm, a priori condition, namely the number of signal sources, is required, and particularly in an application example, the ground penetrating radar is responsible for monitoring the number of defect points of a road. As mentioned above, the road diseases are generated inside and are not easy to be directly observed, and the ground penetrating radar works to monitor the road surface, detect the defects and position the road, so that the estimation of the number of the road defects is needed before the existing signal source positioning algorithm is applied, the working steps are increased, the practicability is reduced, and the accuracy of the estimation result can directly influence the accuracy of the positioning. The ground penetrating radar far-field positioning technology which is not constrained by the priori knowledge of the defect quantity can solve the problem that the defect quantity cannot be directly obtained in the actual application process because the traditional positioning technology depends on the priori knowledge, namely the defect quantity. The ground penetrating radar far-field positioning algorithm which is not constrained by the priori knowledge of the defect quantity is explained in detail below.
For convenience of understanding, please refer to fig. 1, in which fig. 1 is a flowchart illustrating a method for far-field positioning of a ground penetrating radar according to an embodiment of the present disclosure, and as shown in fig. 1, the method specifically includes:
s101, constructing a receiving matrix model Y of an antenna receiving array for receiving the road surface defect reflection signals, wherein Y represents the road surface defect reflection signals received by an array element, A represents a direction response vector, S represents a signal source incident signal, N represents array noise, and the antenna receiving array is a uniform circular array manifold matrix;
the receiving matrix model Y of the antenna receiving array for receiving the road surface defect reflected signal is constructed AS + N, the relationship between the road surface defect reflected signal and the signal source incident signal is expressed by the receiving matrix model, and a plurality of array elements can simultaneously receive a plurality of signals. A ═ a (θ)1,Φ1),a(θ2,Φ2),…,a(θk,Φk)]TAnd k is the number of the signal sources, theta is the azimuth angle of the incident signals of the signal sources, and phi is the direction angle of the incident signals of the signal sources.
The working diagram of the ground penetrating radar during road surface monitoring is shown in fig. 3, the detection radar sends out detection signals to a road and receives reflection signals from defects, and the application of the signal source positioning algorithm in the ground penetrating radar is to position the defects according to the reflection signals from the road defects received by the antenna receiving array.
For ym(t) N-point sampling, the problem to be dealt with becomes by outputting the signal ym(t) sample ymWhen the (t) is 1,2, …, M (M is the number of array elements) estimates the arrival direction angle of the signal source incoming signal, thereby the array signal can be naturally regarded as the superposition of a plurality of spatial harmonics of noise interference, and the arrival direction estimation problem is linked with the spectrum estimation.
S102, converting the uniform circular manifold matrix of the receiving matrix model into an array manifold matrix by a mode space conversion method;
the uniform circular array manifold matrix structure is an array element space non-vandermonde structure as shown in fig. 4, and needs to be converted into a pattern space by a pattern space conversion method so as to satisfy the vandermonde structure. This step is related to the next step of signal decomposition using forward spatial smoothing.
Specifically, two sides of a transformation matrix B and a receiving matrix model Y, namely AS + N, are multiplied simultaneously, so that a uniform circular array manifold matrix of the receiving matrix model is converted into an array manifold matrix;
the transformation matrix is specifically:
Figure BDA0003202828650000071
wherein, Jh(beta) is a Bessel function of the first class in the h stage, h is the mode serial number of the mode space, and h is approximately equal to beta
The receive matrix model transforms to:
Figure BDA0003202828650000072
wherein the content of the first and second substances,
Figure BDA0003202828650000073
and t represents the moment of collecting the reflection signal of the road surface defect.
After transformation
Figure BDA0003202828650000074
The original uniform circular array manifold matrix is converted into a form satisfying the vandermonde structure, and the virtual uniform matrix has M ═ 2h +1 pattern array element elements, has translational invariance as in the actual uniform circular array manifold matrix, and can be spatially smoothed.
S103, dividing the array manifold matrix of the receiving matrix model into a plurality of mutually overlapped sub-matrices by utilizing a forward space smoothing method;
it should be noted that, the decoherence problem of the coherent information source is solved by using a forward spatial smoothing method, the basic idea is to divide the array manifold matrix into a plurality of mutually overlapped sub-arrays, when the coherent information source enters the original whole linear array, the signal sub-space vector of the array covariance matrix is reduced, but when the signal enters different sub-arrays, the signal sub-space vector of the covariance sum of the sub-arrays may not be reduced, thereby achieving the goal of decoherence.
The array elements of the array manifold matrix are divided into p sub-arrays which are staggered with each other from front to back, the number of the array elements of each sub-array is M, and obviously, M is p + M-1. Taking the first subarray from left to right as a reference subarray, the kth subarray specifically is:
Figure BDA0003202828650000081
wherein the content of the first and second substances,
Figure BDA0003202828650000082
s104, determining a covariance matrix of an array manifold matrix of the receiving matrix model according to the data covariance matrix of each sub-matrix;
after obtaining the sub-arrays, the covariance matrix of the data of each sub-array is obtained
Figure BDA0003202828650000083
Determining a covariance matrix of an array manifold matrix of the receive matrix model as
Figure BDA0003202828650000084
Where H represents the conjugate transpose of the matrix.
In order to recover the covariance matrix of the full rank, the forward smoothing space method is implemented by solving the mean value of the data covariance matrix of each sub-matrix, that is, the covariance matrix of the array manifold matrix of the receiving matrix model is obtained by forward smoothing correction.
S105, acquiring hyper-parameters of the receiving matrix model based on a machine learning method;
specifically, the following are provided:
constructing a first training data set and a first testing data set;
and (4) recording data such as signal-to-noise ratio, positioning angle, reasonable threshold, selected hyper-parameters and the like in an experiment. And repeating the experiment for multiple times, and recording corresponding data to obtain a labeled data set consisting of experimental data. And then dividing the data set into a training set and a testing set by a self-help method. And carrying out data clearness on the separated training set data, and separating the pre-measurement from the label.
Training a preset machine learning model through a first training data set until a loss function of the preset machine learning model reaches an optimal value to obtain optimal parameters of the preset machine learning model;
and selecting a proper training model, wherein the problem is a supervised learning task and is a simple univariate regression problem, and selecting a linear regression model as a preset machine learning model for training. Let the characteristic snr be x and the linear regression model function be y ═ b + w.x, where y is the output, in this case, y is the hyper-parameter we need to know. And the values of w and b are the problems to be solved by training.
The loss function is defined as
Figure BDA0003202828650000091
Wherein the content of the first and second substances,
Figure BDA0003202828650000092
the label value (true value) representing the training data,
Figure BDA0003202828650000093
represents the nth set of experimental data. The mathematical meaning of the loss function is to calculate the sum of squares of the differences between the real values and the predicted values of the plurality of groups of experimental data. Taking 10 sets of experimental data, calculating a loss function, wherein the smaller the loss function is, the better the model is trained.
The loss function is a measure for evaluating the quality of model training, and a gradient descent method is required to screen the best model. According to the meaning of the loss function, finding the optimal model is to find the corresponding w and b values when the loss function is minimum. A conceptual learning rate is introduced here: the step length η of the movement. Starting from w, the step of finding w which minimizes the loss function by using a gradient descent method is as follows:
1. randomly selecting a w 0;
2. calculating the differential, namely the current slope, judging the moving direction according to the slope, moving to the right (increasing w) more than 0 and moving to the left (decreasing w) less than 0;
3. move according to the learning rate.
4. And repeating the step 2 and the step 3 until the lowest point is found.
Two parameters, w and b, are introduced similarly, and what needs to be done is partial differentiation, the process is as follows:
1. randomly selecting w0 and b 0;
2. a partial differential is calculated.
Figure BDA0003202828650000094
3. Move according to the learning rate.
Figure BDA0003202828650000095
4. Repeating steps 2 and 3 until the lowest point is found.
And inputting the signal-to-noise ratio of the road surface defect reflection signal of the receiving matrix model into a preset machine learning model with optimal parameters to obtain the hyper-parameters of the receiving matrix model.
The result of the loss function is updated continuously by the gradient descent, which is smaller and smaller, and the gradient descent can basically find the optimal point in the linear model. And the w and b values corresponding to the optimal points are the optimal model after training. In actual defect detection, only the signal-to-noise ratio is used as a characteristic input model, and the corresponding output hyper-parameter can be obtained.
S106, constructing a subspace according to the covariance matrix of the array manifold matrix of the receiving matrix model and the hyperparameter, wherein the subspace is full rank and is orthogonal to the direction response vector;
note that, a subspace Un of full rank and orthogonal to the directional response vector is formed by using the covariance matrix obtained in step S104.
Properties 1: the space spanned by the eigenvector corresponding to the large eigenvalue of the covariance matrix is the same space spanned by the steering vector of the incident signal.
Properties 2: the signal subspace and noise subspace are orthogonal, and when i ═ N +1Hei＝0。
Combining properties 1 and 2, the steering vector of the easily accessible array is also orthogonal to the noise subspace. The conventional MUSIC algorithm uses these two properties to estimate the signal source. It is easy to find that when using this property for estimating signal sources, the signal subspace and the noise subspace have to be separated, and the separation of the signal subspace and the noise subspace has to depend on the exact number K of signal sources.
After feature decomposition is carried out on the covariance matrix, the feature values obtained by decomposition are sequenced, wherein the feature vectors corresponding to the larger K feature values are expanded into a signal subspace, and the rest are noise subspaces. However, the techniques presented herein require that the dependency on the number of signal sources be avoided. Based on the principle of propagation operators, the existing use of the number of signal sources for analysis and the problem when estimation errors occur,
analysis shows that when the estimated source number K1 is smaller than the actual source number, the rank of the constructed subspace is less than or equal to K1, and the K1 is used as the rank of the constructed subspace<K, and K is equal to the rank of the direction vector a, so the constructed subspace does not satisfy orthogonality with a. When the estimated source number K1 is greater than the actual source number, the subspace construction is again induced (assuming that the constructed subspace is U)n) Not full rank, [ U ] at the next stepn H.Un]-1Problems arise in operation.
Therefore, when constructing the subspace Un orthogonal to the direction vector, it is required to ensure that the rank of the separation subspace is greater than the number K of signal sources and the full rank thereof. Here, becauseThe rank of the covariance matrix must be greater than or equal to the number of signal sources, so that Un is only required to be I-Rf (Rf)HRf+μI)-1RfH(where I is the identity matrix, the rank of the matrix equals rank (Rf), and μ is the hyperparameter).
S107, obtaining a spectral peak screening threshold value of the receiving matrix model based on a deep convolutional neural network method;
specifically, the method comprises the following steps:
constructing a second training data set and a second testing data set;
and data such as signal-to-noise ratio, positioning angle, reasonable threshold value and the like are recorded in an experiment. And repeating the experiment for multiple times, and recording corresponding data to obtain a labeled data set consisting of experimental data. And then dividing the data set into a training set and a testing set by a self-help method. And carrying out data clearness on the separated training set data, and separating the pre-measurement from the label.
Training the preset deep convolutional neural network through a second training data set until a loss function of the preset deep convolutional neural network reaches an optimal value, and obtaining optimal parameters of the preset deep convolutional neural network;
selecting a proper training model, constructing a deep convolutional neural network by using a convolutional network DCN and a fully-connected network FCN, and effectively training the network by selecting a proper training strategy.
The convolutional network DCN can manually adjust the performance of the network by changing the number of convolutional kernels. In the process of training the full-connection network FCN, optimal search is carried out on internal parameters of the network by means of a loss function and a gradient descent method so as to obtain an optimal network.
And inputting the road surface defect reflection signals of the receiving matrix model into a preset depth convolution neural network with optimal parameters to obtain a spectral peak screening threshold of the receiving matrix model.
And S108, defining a space spectrum according to the subspace, the spectrum peak screening threshold and the direction response vector, carrying out spectrum search on the space spectrum, and obtaining the angle information corresponding to the searched wave peak as the positioning information of the road surface defect.
It should be noted that the spatial spectrum specifically includes:
Figure BDA0003202828650000111
wherein Un is a subspace, and Un ═ I-Rf (Rf)HRf+μI)-1RfHI is the identity matrix, μ is the hyperparameter, and a is the column vector of the directional response vector a.
After the space spectrum is defined, searching the frequency spectrum, searching a peak within a numerical range exceeding a spectrum peak screening threshold value, wherein an angle corresponding to the peak is positioning information of the road surface defect.
The embodiment provides a ground penetrating radar far-field positioning method, which is based on a receiving matrix model of a uniform circular array manifold matrix, converts a mode space into an array manifold matrix capable of meeting a Van der Monte structure through a mode space conversion method, divides the array manifold matrix into a plurality of mutually overlapped sub-arrays by utilizing a forward space smoothing method, obtains a covariance matrix of the receiving matrix model by calculating a data covariance matrix of each sub-array, finally obtains a hyper-parameter and a spectral peak screening threshold value by introducing a machine learning and deep convolution neural network method, and determines positioning information of a received pavement defect reflection signal by utilizing a space spectrum.
Referring to fig. 2, fig. 2 is a schematic structural diagram of a far-field positioning device of a ground penetrating radar in an embodiment of the present application, as shown in fig. 2, specifically, the far-field positioning device includes:
the first constructing unit 201 is configured to construct a receiving matrix model Y of an antenna receiving array for receiving a road surface defect reflected signal, where Y represents a road surface defect reflected signal received by an array element, a represents a directional response vector, S represents a signal source incident signal, N represents array noise, and the antenna receiving array is a uniform circular array manifold matrix;
a transforming unit 202, configured to transform the uniform circular manifold matrix of the receiving matrix model into an array manifold matrix by a mode space transformation method;
the processing unit 203 is configured to divide the array manifold matrix of the receiving matrix model into a plurality of mutually overlapped sub-matrices by using a forward spatial smoothing method;
a calculating unit 204, configured to determine a covariance matrix of an array manifold matrix of the receiving matrix model according to the data covariance matrix of each sub-matrix;
a first obtaining unit 205, configured to obtain a hyper-parameter of a receiving matrix model based on a machine learning method;
a second constructing unit 206, configured to construct a subspace according to the covariance matrix of the array manifold matrix of the receive matrix model and the hyperparameter, where the subspace is full rank and is orthogonal to the directional response vector;
a second obtaining unit 207, configured to obtain a spectral peak screening threshold of the receiving matrix model based on a deep convolutional neural network method;
and the positioning unit 208 is configured to define a spatial spectrum according to the subspace, the spectral peak screening threshold, and the directional response vector, perform spectrum search on the spatial spectrum, and obtain angle information corresponding to the searched peak as positioning information of the road surface defect.
The present application further provides a computer-readable storage medium for storing a program code for executing any one of the embodiments of the far-field positioning method for a ground penetrating radar in the foregoing embodiments.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
The terms "first," "second," "third," "fourth," and the like in the description of the application and the above-described figures, if any, are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used is interchangeable under appropriate circumstances such that the embodiments of the application described herein are, for example, capable of operation in sequences other than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed, but may include other steps or elements not expressly listed or inherent to such process, method, article, or apparatus.
It should be understood that in the present application, "at least one" means one or more, "a plurality" means two or more. "and/or" for describing an association relationship of associated objects, indicating that there may be three relationships, e.g., "a and/or B" may indicate: only A, only B and both A and B are present, wherein A and B may be singular or plural. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship. "at least one of the following" or similar expressions refer to any combination of these items, including any combination of single item(s) or plural items. For example, at least one (one) of a, b, or c, may represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", wherein a, b, c may be single or plural.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other manners. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one logical division, and other divisions may be realized in practice, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated unit, if implemented in the form of a software functional unit and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application may be substantially implemented or contributed to by the prior art, or all or part of the technical solution may be embodied in a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
The above embodiments are only used for illustrating the technical solutions of the present application, and not for limiting the same; although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions in the embodiments of the present application.

Claims (10)

1. A far-field positioning method for a ground penetrating radar is characterized by comprising the following steps:
step S101, constructing a receiving matrix model Y of an antenna receiving array for receiving road surface defect reflection signals, wherein Y represents the road surface defect reflection signals received by an array element, A represents a direction response vector, S represents a signal source incident signal, N represents array noise, and the antenna receiving array is a uniform circular array manifold matrix;
step S102, converting the uniform circular manifold matrix of the receiving matrix model into an array manifold matrix by a mode space conversion method;
step S103, dividing the array manifold matrix of the receiving matrix model into a plurality of mutually overlapped sub-matrices by utilizing a forward space smoothing method;
step S104, determining a covariance matrix of an array manifold matrix of the receiving matrix model according to the data covariance matrix of each subarray;
s105, acquiring hyper-parameters of the receiving matrix model based on a machine learning method;
step S106, constructing a subspace according to the covariance matrix of the array manifold matrix of the receiving matrix model and the hyperparameter, wherein the subspace is full rank and is orthogonal to the direction response vector;
s107, acquiring a spectral peak screening threshold value of the receiving matrix model based on a deep convolutional neural network method;
and S108, defining a space spectrum according to the subspace, the spectrum peak screening threshold and the direction response vector, and searching the frequency spectrum of the space spectrum, wherein the searched angle information corresponding to the peak is the positioning information of the road surface defect.
2. The method according to claim 1, wherein in step S101, a ═ a (θ) is determined1,Φ1),a(θ2,Φ2),…,a(θk,Φk)]TAnd k is the number of the signal sources, theta is the azimuth angle of the incident signals of the signal sources, and phi is the direction angle of the incident signals of the signal sources.
3. The method according to claim 1, wherein the step S102 specifically comprises:
multiplying the transformation matrix B and two sides of the receiving matrix model Y which is AS + N at the same time, so that the uniform circular manifold matrix of the receiving matrix model is converted into an array manifold matrix;
the transformation matrix is specifically:
Figure FDA0003202828640000021
wherein, Jh(beta) is a first Bessel function of h stage, h is a mode serial number of a mode space, and h is approximately equal to beta;
the receive matrix model transforms to:
Figure FDA0003202828640000022
wherein the content of the first and second substances,
Figure FDA0003202828640000023
and t represents the moment of collecting the reflection signal of the pavement defect.
4. The method according to claim 3, wherein the kth sub-array in step S103 is:
Figure FDA0003202828640000024
wherein the content of the first and second substances,
Figure FDA0003202828640000025
5. the method according to claim 4, wherein the step S104 specifically comprises:
according to the data covariance matrix of each subarray
Figure FDA0003202828640000026
Determining a covariance matrix of an array manifold matrix of the receive matrix model as
Figure FDA0003202828640000027
Where H represents the conjugate transpose of the matrix.
6. The method according to claim 5, wherein the step S105 specifically comprises:
constructing a first training data set and a first testing data set;
training a preset machine learning model through the first training data set until a loss function of the preset machine learning model reaches an optimal value to obtain optimal parameters of the preset machine learning model;
and inputting the signal-to-noise ratio of the road surface defect reflection signal of the receiving matrix model into a preset machine learning model of the optimal parameter to obtain the hyper-parameter of the receiving matrix model.
7. The method according to claim 6, wherein the step S107 specifically comprises:
constructing a second training data set and a second testing data set;
training a preset deep convolutional neural network through the second training data set until a loss function of the preset deep convolutional neural network reaches an optimal value to obtain optimal parameters of the preset deep convolutional neural network;
and inputting the road surface defect reflection signals of the receiving matrix model into a preset depth convolution neural network of the optimal parameters to obtain a spectral peak screening threshold value of the receiving matrix model.
8. The method according to claim 7, wherein the spatial spectrum in step S108 is specifically:
Figure FDA0003202828640000031
wherein Un is a subspace, and Un ═ I-Rf (Rf)HRf+μI)-1RfHI is the identity matrix, μ is the hyperparameter, and a is the column vector of the directional response vector a.
9. A ground penetrating radar far field positioning device, comprising:
the first construction unit is used for constructing a receiving matrix model Y of an antenna receiving array for receiving the road surface defect reflection signals, wherein Y represents the road surface defect reflection signals received by an array element, A represents a direction response vector, S represents a signal source incident signal, N represents array noise, and the antenna receiving array is a uniform circular array manifold matrix;
a transformation unit for transforming the uniform circular manifold matrix of the receiving matrix model into an array manifold matrix by a mode space transformation method;
the processing unit is used for dividing the array manifold matrix of the receiving matrix model into a plurality of mutually overlapped sub-matrices by utilizing a forward space smoothing method;
the calculation unit is used for determining a covariance matrix of an array manifold matrix of the receiving matrix model according to the data covariance matrix of each sub-matrix;
a first obtaining unit, configured to obtain a hyper-parameter of the receiving matrix model based on a machine learning method;
a second constructing unit, configured to construct a subspace according to a covariance matrix of the array manifold matrix of the receive matrix model and the hyper-parameter, where the subspace is full rank and is orthogonal to the directional response vector;
the second acquisition unit is used for acquiring a spectral peak screening threshold value of the receiving matrix model based on a deep convolutional neural network method;
and the positioning unit is used for defining a space spectrum according to the subspace, the spectrum peak screening threshold and the direction response vector, carrying out spectrum search on the space spectrum, and obtaining the angle information corresponding to the searched peak as the positioning information of the road surface defect.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium is configured to store a program code for performing the method of far-field localization of a georadar according to any of claims 1-8.
CN202110909174.XA 2021-08-09 2021-08-09 Ground penetrating radar far field positioning method and device and computer readable storage medium Active CN113608192B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110909174.XA CN113608192B (en) 2021-08-09 2021-08-09 Ground penetrating radar far field positioning method and device and computer readable storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110909174.XA CN113608192B (en) 2021-08-09 2021-08-09 Ground penetrating radar far field positioning method and device and computer readable storage medium

Publications (2)

Publication Number Publication Date
CN113608192A true CN113608192A (en) 2021-11-05
CN113608192B CN113608192B (en) 2022-02-18

Family

ID=78340024

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110909174.XA Active CN113608192B (en) 2021-08-09 2021-08-09 Ground penetrating radar far field positioning method and device and computer readable storage medium

Country Status (1)

Country Link
CN (1) CN113608192B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116054909A (en) * 2022-12-16 2023-05-02 广东工业大学 Satellite communication anti-interference method based on improved propagation operator

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8861655B1 (en) * 2013-06-11 2014-10-14 King Fahd University Of Petroleum And Minerals Method of performing structure-based bayesian sparse signal reconstruction
US20160071526A1 (en) * 2014-09-09 2016-03-10 Analog Devices, Inc. Acoustic source tracking and selection
CN107255793A (en) * 2017-06-16 2017-10-17 中国电子科技集团公司第二十九研究所 A kind of array df method and device for wideband OFDM signal of communication
CN109541526A (en) * 2018-12-29 2019-03-29 西北工业大学 A kind of ring array direction estimation method using matrixing
CN109975842A (en) * 2019-03-04 2019-07-05 广东工业大学 A kind of blind catching method of Big Dipper satellite signal high-precision based on wavelet transformation
CN111123192A (en) * 2019-11-29 2020-05-08 湖北工业大学 Two-dimensional DOA positioning method based on circular array and virtual extension
CN111767791A (en) * 2020-05-21 2020-10-13 中国地质大学（武汉） Arrival angle estimation method based on anti-regularization deep neural network
CN112684408A (en) * 2020-11-06 2021-04-20 西安交通大学 Near-field signal source positioning method, system and equipment based on sparse Bayesian learning

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8861655B1 (en) * 2013-06-11 2014-10-14 King Fahd University Of Petroleum And Minerals Method of performing structure-based bayesian sparse signal reconstruction
US20160071526A1 (en) * 2014-09-09 2016-03-10 Analog Devices, Inc. Acoustic source tracking and selection
CN107255793A (en) * 2017-06-16 2017-10-17 中国电子科技集团公司第二十九研究所 A kind of array df method and device for wideband OFDM signal of communication
CN109541526A (en) * 2018-12-29 2019-03-29 西北工业大学 A kind of ring array direction estimation method using matrixing
CN109975842A (en) * 2019-03-04 2019-07-05 广东工业大学 A kind of blind catching method of Big Dipper satellite signal high-precision based on wavelet transformation
CN111123192A (en) * 2019-11-29 2020-05-08 湖北工业大学 Two-dimensional DOA positioning method based on circular array and virtual extension
CN111767791A (en) * 2020-05-21 2020-10-13 中国地质大学（武汉） Arrival angle estimation method based on anti-regularization deep neural network
CN112684408A (en) * 2020-11-06 2021-04-20 西安交通大学 Near-field signal source positioning method, system and equipment based on sparse Bayesian learning

Non-Patent Citations (6)

* Cited by examiner, † Cited by third party
Title
FANG DONG等: "Real-Valued Sparse DOA Estimation for MIMO Array System Under Unknown Nonuniform Noise", 《IEEE ACCESS》 *
JIANZHONG LI等: "Simplified High-Order DOA and Range Estimation With Linear Antenna Array", 《IEEE COMMUNICATIONS LETTERS》 *
MENG SUN等: "Estimation of time delay andi nterface roughness by GPR using modified MUSIC", 《SIGNAL PROCESSING》 *
THI NGOC THO NGUYEN等: "Robust Source Counting and DOA Estimation Using Spatial Pseudo-Spectrum and Convolutional Neural Network", 《IEEE/ACM TRANSACTIONS ON AUDIO, SPEECH, AND LANGUAGE PROCESSING》 *
张祺等: "基于模式空间算法的声源二维DOA估计", 《数据采集与处理》 *
张薇等: "基于均匀圆阵Toeplitz矩阵集重构解相干算法", 《吉林大学学报(工学版)》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116054909A (en) * 2022-12-16 2023-05-02 广东工业大学 Satellite communication anti-interference method based on improved propagation operator
CN116054909B (en) * 2022-12-16 2023-08-08 广东工业大学 Satellite communication anti-interference method based on improved propagation operator

Also Published As

Publication number Publication date
CN113608192B (en) 2022-02-18

Similar Documents

Publication Publication Date Title
Sun et al. Advanced signal processing methods for ground-penetrating radar: Applications to civil engineering
US5990834A (en) Radar angle determination with music direction finding
CN110045323B (en) Matrix filling-based co-prime matrix robust adaptive beamforming algorithm
CN107544051A (en) Wave arrival direction estimating method of the nested array based on K R subspaces
CN109597021B (en) Direction-of-arrival estimation method and device
CN113608192B (en) Ground penetrating radar far field positioning method and device and computer readable storage medium
Tian et al. Passive localization of mixed sources jointly using MUSIC and sparse signal reconstruction
Li et al. Sparse representation-based joint angle and Doppler frequency estimation for MIMO radar
Ang et al. Multiband grid-free compressive beamforming
Kujawski et al. Fast grid-free strength mapping of multiple sound sources from microphone array data using a Transformer architecture
RU2491569C2 (en) Method of direction finding with increased resolution ability
JP5253278B2 (en) Multidimensional data identification device, multidimensional data identification method, and signal arrival direction estimation device
WO2010066306A1 (en) Apparatus and method for constructing a sensor array used for direction of arrival (doa) estimation
JP2001337148A (en) Extrapolation device for electromagnetic wave arrival direction
CN112444773A (en) Compressed sensing two-dimensional DOA estimation method based on spatial domain fusion
CN109557503B (en) MIMO (multiple input multiple output) co-prime array DOA (direction of arrival) estimation method based on correlation matrix reconstruction decorrelation
Meschino et al. A practical guide on using SPOT-GPR, a freeware tool implementing a SAP-DoA technique
Rajani et al. Direction of arrival estimation by using artificial neural networks
CN113281698A (en) Cascade-based non-Gaussian source direction finding method in nested array
Li et al. DOA estimation for sparse nested MIMO radar with velocity receive sensor array
CN109061564B (en) Simplified near-field positioning method based on high-order cumulant
Zhao et al. Capon cepstrum weighted l2, 1 minimization for wideband DOA estimation with sonar arrays
JP5724548B2 (en) Direction measuring apparatus, direction measuring method and direction measuring program
Du et al. Three stochastic measurement schemes for direction-of-arrival estimation using compressed sensing method
Huang et al. DOA estimation of quasi-stationary signals with a nested array in unknown noise field

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