CN114187516A - Ground penetrating radar underground cavity target identification method based on BP neural network - Google Patents

Ground penetrating radar underground cavity target identification method based on BP neural network Download PDF

Info

Publication number
CN114187516A
CN114187516A CN202111524442.2A CN202111524442A CN114187516A CN 114187516 A CN114187516 A CN 114187516A CN 202111524442 A CN202111524442 A CN 202111524442A CN 114187516 A CN114187516 A CN 114187516A
Authority
CN
China
Prior art keywords
target
scan
signal
horizontal direction
ground penetrating
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111524442.2A
Other languages
Chinese (zh)
Inventor
薛伟
王泽熙
陈可晖
麻少鹏
李婷
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.)
China University of Geosciences
Original Assignee
China University of Geosciences
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 China University of Geosciences filed Critical China University of Geosciences
Priority to CN202111524442.2A priority Critical patent/CN114187516A/en
Publication of CN114187516A publication Critical patent/CN114187516A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
    • G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Radar Systems Or Details Thereof (AREA)

Abstract

The invention provides a ground penetrating radar underground cavity target identification method based on a BP neural network, which comprises the following steps: performing clutter suppression on the B-Scan image of the ground penetrating radar by using a robust principal component analysis method; extracting time domain, frequency domain and wavelet domain characteristics of the A-Scan signal; integrally identifying the A-Scan signal by using a BP neural network to obtain a target horizontal direction area; segmenting the A-Scan signals in the target horizontal direction region, and extracting the characteristics of the time domain, the frequency domain and the wavelet domain of each segment of signals; identifying each section of signal by adopting a BP neural network to obtain a target vertical direction area in each A-Scan signal; and determining the target position according to the target horizontal direction area and the target vertical direction area. The invention has the beneficial effects that: the method can improve clutter suppression effect and effectively improve target identification efficiency while ensuring target identification accuracy, and is suitable for detecting and positioning underground cavity targets of roads.

Description

Ground penetrating radar underground cavity target identification method based on BP neural network
Technical Field
The invention relates to the field of radar signal processing and target identification, in particular to a ground penetrating radar underground cavity target identification method based on a BP neural network.
Background
Along with the continuous development of economy in China, the urbanization level is gradually improved, the urban traffic pressure is higher and higher, and the detection and repair of underground road diseases are key problems in road maintenance. The underground cavity is the most common road disease, easily causes the ground to cave in, not only causes serious influence to urban traffic, can cause harm to people's life and property safety moreover. Therefore, the detection and the identification of the underground cavity have important significance for guaranteeing the safety of urban roads.
The ground penetrating radar has the advantages of high detection speed, continuous detection process, high resolution, convenience in operation, low detection cost, wide detection range, visual profile and the like, and is increasingly applied to detection of underground cavities of roads. The traditional ground penetrating radar section interpretation work mainly depends on the experience of workers, mistakes are easy to make in the interpretation process, the efficiency is low, and in order to improve the detection accuracy and efficiency, the automatic identification technology of a target needs to be researched.
At present, a ground penetrating radar target identification method mainly comprises a machine learning method (a support vector machine and a neural network) and a deep learning method (a convolutional neural network). The machine learning method needs to extract the features and then identify the features, so that the network structure is simple and the running speed is high; the deep learning method directly utilizes a deep network to extract features, the network structure is complex, a large amount of marked samples are required for training, and the recognition speed is low. Therefore, the machine learning method is more advantageous in view of recognition efficiency. In the machine learning method, from the viewpoint of feature extraction, two categories based on a-Scan signal (mono signal) and B-Scan image are mainly classified. The feature extraction method based on the A-Scan signal is used for extracting features from one-dimensional data of the ground penetrating radar, and the calculation is relatively simple; the feature extraction method based on the B-Scan image mainly extracts features from a two-dimensional image, and the calculation is relatively complex. Therefore, the feature extraction method based on the A-Scan signal has smaller calculation amount.
In the identification of the hole target of the ground penetrating radar, the identification method based on the A-Scan signal feature extraction can be divided into two types: the method has the advantages that the overall characteristics of an A-Scan signal are extracted, a classifier is utilized to identify whether the signal contains a hole target, the identification efficiency is high, but only the horizontal position of the target can be identified, and the method is used for preliminarily judging the hole; one is to extract the characteristics of A-Scan segmented signals, and utilize a classifier to identify whether the segmented signals are cavity target echoes or not. In actual road underground cavity detection, the number of A-Scan signals containing cavity targets is small, and identification of all A-Scan segmented signals is unnecessary, so that how to reduce identification of invalid segmented signals and improve identification efficiency is of great significance in improving the detection performance of the ground penetrating radar cavity.
Disclosure of Invention
In order to solve the problems, the invention provides a ground penetrating radar underground cavity target identification method based on a BP neural network, which adopts a robust principal component analysis method to decompose a B-Scan image of a ground penetrating radar so as to better distinguish clutter and target components and improve clutter suppression effect; extracting multi-dimensional combination characteristics on a time domain, a frequency domain and a wavelet domain, and better reflecting the difference between a target signal and a non-target signal; the BP neural network is used for carrying out integral identification and segmented identification on the A-Scan signal, so that the calculation amount can be reduced, the integral identification and the segmented identification are combined, the identification accuracy is ensured, the identification efficiency can be effectively improved, and the method is suitable for detecting and positioning the underground cavity target of the road. The method for identifying the target of the underground cavity of the ground penetrating radar mainly comprises the following steps:
s1: acquiring a B-Scan image of the ground penetrating radar, and processing the B-Scan image of the ground penetrating radar by using a robust principal component analysis method to suppress clutter signals;
s2: extracting the characteristics of each A-Scan signal in the processed B-Scan image to respectively obtain time domain, frequency domain and wavelet domain characteristics;
s3: integrally identifying each A-Scan signal in the B-Scan image after the clutter suppression by using a BP (Back propagation) neural network to obtain a target horizontal direction area;
s4: segmenting the A-Scan signals in the target horizontal direction region, and extracting the time domain, the frequency domain and the wavelet domain characteristics of each segment of signals;
s5: according to the time domain, frequency domain and wavelet domain characteristics of each segment of signal obtained in the step S4, utilizing a BP neural network to identify segmented signals in the A-Scan to obtain a target vertical direction region in the A-Scan signal;
s6: and determining the target position according to the target horizontal direction area and the target vertical direction area.
Further, the steps of performing clutter signal suppression on the B-Scan image of the ground penetrating radar by using a robust principal component analysis method are as follows:
(1) obtaining B-Scan image X belonging to R of ground penetrating radarM×NWherein N is the track number of A-Scan signals in the B-Scan image, M is the sampling point number of each A-Scan signal, and X is [ X ]1(m),x2(m),…,xN(m)],xi(m)＝(xi(1),xi(2),…,xi(M))TThe ith channel of A-Scan signal is provided, M represents the serial number of a sampling point in each channel of A-Scan signal, and M is more than or equal to 1 and less than or equal to M;
(2) processing each A-Scan signal in the B-Scan image X of the ground penetrating radar to remove the DC offset, wherein the processing process comprises the following steps:
Figure BDA0003409557350000031
wherein the content of the first and second substances,
Figure BDA0003409557350000032
and the mean value of each channel of A-Scan signal data in the B-Scan image X of the ground penetrating radar is shown.
(3) And processing the B-Scan image of the ground penetrating radar after the direct current offset is removed by adopting a robust principal component analysis method to obtain a low-rank matrix G and a sparse matrix S, and selecting the sparse matrix S as a target echo signal to obtain a B-Scan image X1 after clutter suppression.
Further, the time domain, frequency domain and wavelet domain characteristics obtained in step S2 are respectively as follows:
(1) time domain characterization
The time domain features include 3 features: MEANiStandard deviation STDiAnd root mean square value RMSiThe calculation process is as follows:
Figure BDA0003409557350000033
Figure BDA0003409557350000034
Figure BDA0003409557350000035
wherein, x1i(m) the ith channel of A-Scan signal data in the B-Scan image after clutter suppression;
(2) frequency domain features
The frequency domain is characterized by a signal energy spectrum, and the calculation formula is as follows:
Figure BDA0003409557350000036
wherein k1 is a frequency point sequence number;
(3) wavelet domain features
The wavelet domain is characterized by a wavelet packet energy spectrum TiThe wavelet packet energy spectrum is defined as a normalized wavelet packet nodePoint energy:
Figure BDA0003409557350000041
wavelet packet node energy ei,l,jComprises the following steps:
Figure BDA0003409557350000042
wherein i is a track number,
Figure BDA0003409557350000043
is x1i(m) wavelet packet decomposition coefficient, l is the number of layers of decomposition, j is the frequency band, and k2 is the length of the decomposition coefficient.
Further, the step of acquiring the target horizontal direction area is as follows:
(1) respectively selecting a plurality of A-Scan signals containing void targets and non-void targets from the B-Scan image after the clutter suppression, and extracting features to construct a sample set, wherein each sample is a 23-dimensional feature vector;
(2) selecting partial data from the sample set as a training set, training the BP neural network by using the training set to obtain a network model NET1, and identifying the sample set by using NET 1;
(3) for K A-Scan signals identified as containing targets
Figure BDA0003409557350000044
According to its horizontal position ikThen, the target horizontal direction initial region H1 ═ { i ═ is obtainedk,1≤k≤K}；
(4) The fusion and deletion processing is carried out on the target horizontal direction initial region H1, and the specific process is as follows:
1) fusion process
Suppose that
Figure BDA0003409557350000045
And
Figure BDA0003409557350000046
for two adjacent target-containing A-Scan signals, if their track difference is 1 < | ik-ik+1If | ≦ id, all A-Scan signals x1 between the adjacent A-Scan signalsi(m) all of them are judged to contain the target signal, ik≤i≤ik+1Thus obtaining P target horizontal direction areas which are respectively [ ipL ipH]，1<P<K, 1. ltoreq. P. ltoreq.P, where iPLIs the starting track number of the p-th target, ipHThe end track number of the p-th target;
2) delete processing
For the p-th target after the fusion process, its horizontal direction width Wp＝(ipH-ipL) If W ispIf the target is smaller than the threshold value WT, the target is judged to be a false target, and the area corresponding to the target is deleted from the target horizontal direction area, so that the final target horizontal direction area is H2.
Further, the specific steps of segmenting the a-Scan signal in the target horizontal direction region and extracting the time domain, frequency domain and wavelet domain characteristics of each segment of signal are as follows:
(1) all A-Scan signals of the target horizontal direction region are segmented, the length of each segment is L1, and then the ith channel A-Scan signal x1iIn (m), the lk-th segment signal is:
x2i,lk(r)＝x1i((lk-1)*L1+r)，1≤lk≤KL,1≤r≤L1 (8)
wherein, i is track serial number, lk is segment serial number, r is signal point serial number in the segment, the segment number KL ═ floor (M/L1), floor (·) represents rounding down.
(2) The process of extracting the time domain, frequency domain and wavelet domain features of each segment of signals is as follows:
1) time domain characterization
The time domain features include 3 features: MEANi,lkStandard deviation STDi,lkAnd root mean square value RMSi,lkThe calculation formula is as follows:
Figure BDA0003409557350000051
Figure BDA0003409557350000052
Figure BDA0003409557350000053
wherein, x2i,lk(r) the ith channel of lk-segment A-Scan signal data after segmentation processing;
(2) frequency domain features
The frequency domain is characterized by a signal energy spectrum, and the calculation formula is as follows:
Figure BDA0003409557350000054
wherein k3 is a frequency point sequence number;
(3) wavelet domain features
The wavelet domain is characterized by a wavelet packet energy spectrum, which is defined as the normalized wavelet packet node energy:
Figure BDA0003409557350000061
the wavelet packet node energy is:
Figure BDA0003409557350000062
wherein the content of the first and second substances,
Figure BDA0003409557350000063
is x2i,lkThe wavelet packet decomposition coefficient of (r), l is the number of layers of decomposition, j is the frequency band, and k4 is the length of the decomposition coefficient.
Further, the step of acquiring the target vertical direction region in the A-Scan signal is as follows:
(1) respectively selecting a plurality of sections of segmented signals containing a cavity target and a plurality of sections of segmented signals not containing the cavity target from the A-Scan signals in the target horizontal direction region, and extracting features to construct a sample set, wherein each sample is a 19-dimensional feature vector;
(2) selecting partial data from the sample set as a training set, training the BP neural network by using the training set to obtain a network model NET2, and identifying the sample set by using NET 2;
(3) for J segmented signals identified as containing targets
Figure BDA0003409557350000064
By the track number i and its segment number in the track
Figure BDA0003409557350000065
Obtaining the coordinates of the target vertical direction area in the ith channel A-Scan signal as
Figure BDA0003409557350000066
Furthermore, the segmented signals which are identified as containing the target are marked in the corresponding vertical direction and horizontal direction areas, and the position of the target is obtained.
The technical scheme provided by the invention has the beneficial effects that:
1. the B-Scan image of the ground penetrating radar is decomposed by using a robust principal component analysis method, the sparse characteristic of a target component in the image and the low-rank characteristic of a clutter component can be fully utilized, and the separation of the target and the clutter is better realized;
2. the A-Scan signal time domain statistical characteristics, the frequency domain energy spectrum and the wavelet domain wavelet packet energy spectrum are selected as characteristics, complementarity among different types of characteristics can be fully utilized, and a target and a non-target can be better distinguished;
3. firstly, integrally identifying the A-Scan signal by using a BP neural network, and quickly identifying a target horizontal direction area; and then, the A-Scan signals in the target horizontal direction area are identified in a segmented manner, so that the target position can be accurately identified, the calculation of a large number of non-target areas is avoided, and the identification efficiency is improved.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
fig. 1 is a flowchart of a ground penetrating radar underground cavity target identification method based on a BP neural network in the embodiment of the present invention.
FIG. 2 is a schematic diagram of a raw B-Scan image of a ground penetrating radar in an embodiment of the invention.
FIG. 3 is a schematic diagram of a B-Scan image after noise suppression by using a robust principal component analysis method according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of the identified target horizontal initial region in the embodiment of the present invention.
Fig. 5 is a schematic diagram of the target horizontal direction area after the fusion processing and the deletion processing in the embodiment of the present invention.
FIG. 6 is a schematic illustration of identified target locations in an embodiment of the present invention.
Detailed Description
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
The embodiment of the invention provides a ground penetrating radar underground cavity target identification method based on a BP neural network.
Referring to fig. 1, fig. 1 is a flowchart of a method for identifying an underground cavity target of a ground penetrating radar based on a BP neural network in an embodiment of the present invention, which specifically includes the following steps:
1. processing the B-Scan image of the ground penetrating radar by using a robust principal component analysis method to suppress clutter signals; the method comprises the following specific steps:
(1) obtaining B-Scan image X belonging to R of ground penetrating radarM×NWhere N is the number of channels, M is the number of sampling points per channel, and X is [ X ]1(m),x2(m),…,xN(m)],xi(m)＝(xi(1),xi(2),…,xi(M))TThe ith channel of A-Scan signal is provided, M represents the serial number of a sampling point in each channel of A-Scan signal data, and M is more than or equal to 1 and less than or equal to M;
(2) processing each A-Scan signal in the B-Scan image X of the ground penetrating radar to remove the DC offset, wherein the processing process comprises the following steps:
Figure BDA0003409557350000071
wherein the content of the first and second substances,
Figure BDA0003409557350000072
and the mean value of each channel of A-Scan signal data in the B-Scan image X of the ground penetrating radar is shown.
(3) And processing the B-Scan image of the ground penetrating radar after the direct current offset is removed by adopting a robust principal component analysis method to obtain a low-rank matrix G and a sparse matrix S, and selecting the sparse matrix S as a target echo signal to obtain a B-Scan image X1 after clutter suppression.
2. And (3) extracting the characteristics of each A-Scan signal in the processed B-Scan image to respectively obtain the characteristics of a time domain, a frequency domain and a wavelet domain as shown in the following:
(1) time domain characterization
The time domain features include 3 features: MEANiStandard deviation STDiAnd root mean square value RMSiThe calculation process is as follows:
Figure BDA0003409557350000081
Figure BDA0003409557350000082
Figure BDA0003409557350000083
(2) frequency domain features
The frequency domain is characterized by a signal energy spectrum, and the calculation formula is as follows:
Figure BDA0003409557350000084
where k1 is the frequency point number, the energy spectrum of the first 12 odd frequency points (k1 is 1, 3.., 23) is selected as the frequency domain feature in this embodiment.
(3) Wavelet domain features
The wavelet domain is characterized by a wavelet packet energy spectrum TiDefining the wavelet packet energy spectrum as normalized wavelet packet node energy:
Figure BDA0003409557350000085
wavelet packet node energy ei,l,jComprises the following steps:
Figure BDA0003409557350000086
wherein i is a track number,
Figure BDA0003409557350000091
is x1i(m) wavelet packet decomposition coefficient, l is the number of layers of decomposition, j is the frequency band, and k2 is the length of the decomposition coefficient.
In this embodiment, the number of decomposition layers l is selected to be 3, thereby obtaining a wavelet domain feature dimension of 8.
Combining the above three main characteristics, x1 is obtainediThe feature vector dimension of (m) is 23.
3. Integrally identifying the A-Scan signals in the B-Scan image after the clutter suppression by using a BP (Back propagation) neural network to obtain a target horizontal direction area; the steps of acquiring the target horizontal direction area are as follows:
(1) selecting a plurality of A-Scan signals containing void targets and non-void targets from the B-Scan image after the clutter suppression, extracting features and constructing a sample set, wherein each sample is a 23-dimensional feature vector and comprises the following components: 3 time domains, 12 frequency domains and 8 wavelet domains;
(2) selecting partial data from the sample set as a training set, training the BP neural network by using the training set to obtain a network model NET1, and identifying the sample set by using NET 1;
(3) for K A-Scan signals identified as containing targets
Figure BDA0003409557350000092
According to its horizontal position ikThen, the target horizontal direction initial region H1 ═ { i ═ is obtainedk,1≤k≤K}；
(4) The fusion and deletion processing is carried out on the target horizontal direction initial region H1, and the specific process is as follows:
1) fusion process
Suppose that
Figure BDA0003409557350000093
And
Figure BDA0003409557350000094
for two adjacent target-containing A-Scan signals, if their track difference is 1 < | ik-ik+1If | ≦ id, all A-Scan signals x1 between the adjacent A-Scan signalsi(m) all of them are judged to contain the target signal, ik≤i≤ik+1Thus obtaining P target horizontal direction areas which are respectively [ ipL ipH]，1<P<K, 1. ltoreq. P. ltoreq.P, where iPLIs the starting track number of the p-th target, ipHThe end track number of the p-th target;
2) delete processing
For the p-th target after the fusion process, its horizontal direction width Wp＝(ipH-ipL) If W ispIf the target is smaller than the threshold value WT, the target is judged to be a false target, and the area corresponding to the target is deleted from the target horizontal direction area, so that the final target horizontal direction area is H2.
4. Segmenting the A-Scan signal in the target horizontal direction region, and extracting the characteristics of the time domain, the frequency domain and the wavelet domain of each segment of signal, which comprises the following steps:
(1) all A-Scan signals of the target horizontal direction region are segmented, the length of each segment is L1, and then the ith channel A-Scan signal x1iIn (m), the lk-th segment signal is:
x2i,lk(r)＝x1i((lk-1)*L1+r)，1≤lk≤KL,1≤r≤L1 (8)
wherein, i is a track serial number, lk is a segment serial number, r is a serial number of a signal point in a segment, a segment number KL is floor (M/L1), and floor (·) represents downward rounding;
(2) the process of extracting the time domain, frequency domain and wavelet domain features of each segment of signals is as follows:
1) time domain characterization
The time domain features include 3 features: MEANi,lkStandard deviation STDi,lkAnd root mean square value RMSi,lkThe calculation formula is as follows:
Figure BDA0003409557350000101
Figure BDA0003409557350000102
Figure BDA0003409557350000103
wherein, x2i,lk(r) the ith channel of lk-segment A-Scan signal data after segmentation processing;
(2) frequency domain features
The frequency domain is characterized by a signal energy spectrum, and the calculation formula is as follows:
Figure BDA0003409557350000104
wherein k3 is a frequency point sequence number, and in this embodiment, the energy spectrum of the first 8 frequency points (k3 is 1, 2.., 8) is selected as the frequency domain feature;
(3) wavelet domain features
The wavelet domain is characterized by a wavelet packet energy spectrum Ti,lkThe wavelet packet energy spectrum is defined as the normalized wavelet packet node energy:
Figure BDA0003409557350000105
the wavelet packet node energy is:
Figure BDA0003409557350000111
wherein the content of the first and second substances,
Figure BDA0003409557350000112
is x2i,lkThe wavelet packet decomposition coefficient of (r), l is the number of layers of decomposition, j is the frequency band, and k4 is the length of the decomposition coefficient.
In this embodiment, the number of decomposition layers l is selected to be 3, thereby obtaining a wavelet domain feature dimension of 8. And synthesizing the three types of features to obtain a feature vector dimension of the segmented A-Scan signal of 19.
5. According to the time domain, frequency domain and wavelet domain characteristics of each segment of signal obtained in the step 4, a BP neural network is utilized to identify segmented signals in the A-Scan to obtain a target vertical direction region in the A-Scan signal, and the specific steps are as follows:
(1) selecting a plurality of sections of segmented signals containing a hole target and a plurality of sections of segmented signals without the hole target from the A-Scan signals in the target horizontal direction region, extracting features and constructing a sample set, wherein each sample is a 19-dimensional feature vector and comprises the following steps: 3 time domains, 8 frequency domains and 8 wavelet domains;
(2) selecting partial data from the sample set as a training set, training the BP neural network by using the training set to obtain a network model NET2, and identifying the sample set by using NET 2;
(3) for J segmented signals identified as containing targets
Figure BDA0003409557350000113
By the track number i and its segment number in the track
Figure BDA0003409557350000114
Obtaining the coordinates of the target vertical direction area in the ith channel A-Scan signal as
Figure BDA0003409557350000115
6. And marking the segmented signals which are identified as containing the target in the corresponding vertical direction and horizontal direction areas according to the horizontal direction area and the vertical direction area of the target to obtain the position of the target.
With reference to fig. 1, a ground penetrating radar original B-Scan image shown in fig. 2 is taken as an example to illustrate the target identification method of the embodiment of the present invention, which includes the following steps:
1. inputting an original B-Scan image of the ground penetrating radar, as shown in FIG. 2, comprising 551A-Scan signals, wherein each data channel has 512 sampling points;
2. utilizing a robust principal component analysis method to suppress clutter in the original B-Scan image to obtain an image after clutter suppression, wherein the image is shown in FIG. 3;
3. extracting time domain, frequency domain and wavelet domain characteristics for each A-Scan signal, and constructing a 23-dimensional characteristic vector;
4. the constructed BP neural network model NET1 is 3 layers and has a structure of 23-10-2, a sample set is constructed by utilizing a plurality of characteristic vectors containing target and non-target A-Scan signals to train NET1, the trained network is utilized to test the sample set to obtain a target horizontal direction initial region H1, and as shown in FIG. 4, the recognition accuracy is 82.6%;
5. performing fusion processing and deletion processing on the identified target horizontal direction initial region H1 to obtain a final target horizontal direction region H2, as shown in fig. 5, the identification accuracy is 97.5%;
6. segmenting the A-Scan signal in the target horizontal direction region H2, extracting the time domain, frequency domain and wavelet domain characteristics of the segmented signal, and constructing a 19-dimensional characteristic vector;
7. the constructed BP neural network model NET2 is 3 layers and has a structure of 19-10-2, a sample set is constructed by utilizing a plurality of characteristic vectors containing targets and not containing target A-Scan segmented signals to train NET2, and the trained network is utilized to test the sample set to obtain a target vertical direction area V1 in each A-Scan signal;
8. for the segmented signals identified as including the target, marking is performed in the corresponding vertical direction and horizontal direction areas to obtain the position of the target, as shown in fig. 6, the identification accuracy is 95.5%.
Compared with the prior art, the invention has the beneficial effects that:
1. the B-Scan image of the ground penetrating radar is decomposed by using a robust principal component analysis method, the sparse characteristic of a target component in the image and the low-rank characteristic of a clutter component can be fully utilized, and the separation of the target and the clutter is better realized;
2. the A-Scan signal time domain statistical characteristics, the frequency domain energy spectrum and the wavelet domain wavelet packet energy spectrum are selected as characteristics, complementarity among different types of characteristics can be fully utilized, and a target and a non-target can be better distinguished;
3. firstly, integrally identifying the A-Scan signal by using a BP neural network, and quickly identifying a target horizontal direction area; and then, the A-Scan signals in the target horizontal direction area are identified in a segmented manner, so that the target position can be accurately identified, the calculation of a large number of non-target areas is avoided, and the identification efficiency is improved.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the invention, and any modifications, equivalents, improvements and the like that fall within the spirit and principle of the present invention are intended to be included therein.

Claims (7)

1. A ground penetrating radar underground cavity target identification method based on a BP neural network is characterized in that: the method comprises the following steps:
s1: acquiring a B-Scan image of the ground penetrating radar, and processing the B-Scan image of the ground penetrating radar by using a robust principal component analysis method to suppress clutter signals;
s2: extracting the characteristics of each A-Scan signal in the processed B-Scan image to respectively obtain time domain, frequency domain and wavelet domain characteristics;
s3: integrally identifying each A-Scan signal in the B-Scan image after the clutter suppression by using a BP (Back propagation) neural network to obtain a target horizontal direction area;
s4: segmenting the A-Scan signals in the target horizontal direction region, and extracting the time domain, the frequency domain and the wavelet domain characteristics of each segment of signals;
s5: according to the time domain, frequency domain and wavelet domain characteristics of each segment of signal obtained in the step S4, utilizing a BP neural network to identify segmented signals in the A-Scan to obtain a target vertical direction region in the A-Scan signal;
s6: and determining the target position according to the target horizontal direction area and the target vertical direction area.
2. The method for identifying the ground penetrating radar underground cavity target based on the BP neural network as claimed in claim 1, wherein: in step S1, the steps of performing clutter signal suppression on the ground penetrating radar B-Scan image by using the robust principal component analysis method are as follows:
(1) obtaining B-Scan image X belonging to R of ground penetrating radarM×NWherein N is the track number of A-Scan signals in the B-Scan image, M is the sampling point number of each A-Scan signal, and X is [ X ]1(m),x2(m),…,xN(m)],xi(m)＝(xi(1),xi(2),…,xi(M))TThe ith channel of A-Scan signal is provided, M represents the serial number of a sampling point in each channel of A-Scan signal, and M is more than or equal to 1 and less than or equal to M;
(2) processing each A-Scan signal in the B-Scan image X of the ground penetrating radar to remove the DC offset, wherein the processing process comprises the following steps:
Figure FDA0003409557340000011
wherein the content of the first and second substances,
Figure FDA0003409557340000012
representing the mean value of each channel of A-Scan signal data in the B-Scan image X of the ground penetrating radar;
(3) and processing the B-Scan image of the ground penetrating radar after the direct current offset is removed by adopting a robust principal component analysis method to obtain a low-rank matrix G and a sparse matrix S, and selecting the sparse matrix S as a target echo signal to obtain a B-Scan image X1 after clutter suppression.
3. The method for identifying the ground penetrating radar underground cavity target based on the BP neural network as claimed in claim 1, wherein: in step S2, the obtained time domain, frequency domain and wavelet domain features are respectively as follows:
(1) time domain characterization
The time domain features include 3 features: MEANiStandard deviation STDiAnd root mean square value RMSiThe calculation process is as follows:
Figure FDA0003409557340000021
Figure FDA0003409557340000022
Figure FDA0003409557340000023
wherein, x1i(m) the ith channel of A-Scan signal data in the B-Scan image after clutter suppression;
(2) frequency domain features
The frequency domain being characterised by a signal energy spectrum Pi(k1) The calculation formula is as follows:
Figure FDA0003409557340000024
wherein k1 is a frequency point sequence number;
(3) wavelet domain features
The wavelet domain is characterized by a wavelet packet energy spectrum TiThe wavelet packet energy spectrum is defined as the normalized wavelet packet node energy:
Figure FDA0003409557340000025
wavelet packet node energy ei,l,jComprises the following steps:
Figure FDA0003409557340000026
wherein i is a track number,
Figure FDA0003409557340000031
is x1i(m) wavelet packet decomposition coefficient, l is the number of layers of decomposition, j is the frequency band, and k2 is the length of the decomposition coefficient.
4. The method for identifying the ground penetrating radar underground cavity target based on the BP neural network as claimed in claim 1, wherein: in step S3, the step of acquiring the target horizontal direction area is as follows:
(1) respectively selecting a plurality of A-Scan signals containing void targets and non-void targets from the B-Scan image after the clutter suppression, and extracting features to construct a sample set, wherein each sample is a 23-dimensional feature vector;
(2) selecting partial data from the sample set as a training set, training the BP neural network by using the training set to obtain a network model NET1, and identifying the sample set by using NET 1;
(3) for K A-Scan signals identified as containing targets
Figure FDA0003409557340000032
According to its horizontal position ikThen, the target horizontal direction initial region H1 ═ { i ═ is obtainedk,1≤k≤K}；
(4) The fusion and deletion processing is carried out on the target horizontal direction initial region H1, and the specific process is as follows:
1) fusion process
Suppose that
Figure FDA0003409557340000033
And
Figure FDA0003409557340000034
for two adjacent target-containing A-Scan signals, if their track difference is 1 < | ik-ik+1If | ≦ id, all A-Scan signals x1 between the adjacent A-Scan signalsi(m) all of them are judged to contain the target signal, ik≤i≤ik+1Thus obtaining P target horizontal direction areas which are respectively [ ipL ipH]，1<P<K, 1. ltoreq. P. ltoreq.P, where iPLIs the starting track number of the p-th target, ipHThe end track number of the p-th target;
2) delete processing
For the p-th target after the fusion process, its horizontal direction width Wp＝(ipH-ipL) If W ispIf the target is smaller than the threshold value WT, the target is judged to be a false target, and the area corresponding to the target is deleted from the target horizontal direction area, so that the final target horizontal direction area is H2.
5. The method for identifying the ground penetrating radar underground cavity target based on the BP neural network as claimed in claim 1, wherein: in step S4, the specific steps of segmenting the a-Scan signal in the target horizontal direction region and extracting the time domain, frequency domain and wavelet domain features of each segment of signal are as follows:
(1) all A-Scan signals of the target horizontal direction region are segmented, the length of each segment is L1, and then the ith channel A-Scan signal x1iIn (m), the lk-th segment signal is:
x2i,lk(r)＝x1i((lk-1)*L1+r)，1≤lk≤KL,1≤r≤L1 (8)
wherein, i is a track serial number, lk is a segment serial number, r is a serial number of a signal point in a segment, a segment number KL is floor (M/L1), and floor (·) represents downward rounding;
(2) the process of extracting the time domain, frequency domain and wavelet domain features of each segment of signals is as follows:
1) time domain characterization
The time domain features include 3 features: MEANi,lkStandard deviation STDi,lkAnd root mean square value RMSi,lkThe calculation formula is as follows:
Figure FDA0003409557340000041
Figure FDA0003409557340000042
Figure FDA0003409557340000043
wherein, x2i,lk(r) the ith channel of lk-segment A-Scan signal data after segmentation processing;
(2) frequency domain features
The frequency domain is characterized by a signal energy spectrum, and the calculation formula is as follows:
Figure FDA0003409557340000044
wherein k3 is a frequency point sequence number;
(3) wavelet domain features
The wavelet domain is characterized by a wavelet packet energy spectrum, which is defined as the normalized wavelet packet node energy:
Figure FDA0003409557340000045
the wavelet packet node energy is:
Figure FDA0003409557340000046
wherein the content of the first and second substances,
Figure FDA0003409557340000047
is x2i,lkThe wavelet packet decomposition coefficient of (r), l is the number of layers of decomposition, j is the frequency band, and k4 is the length of the decomposition coefficient.
6. The method for identifying the ground penetrating radar underground cavity target based on the BP neural network as claimed in claim 1, wherein: in step S5, the step of acquiring the target vertical direction region in the a-Scan signal is as follows:
(1) respectively selecting a plurality of sections of segmented signals containing a cavity target and a plurality of sections of segmented signals not containing the cavity target from the A-Scan signals in the target horizontal direction region, and extracting features to construct a sample set, wherein each sample is a 19-dimensional feature vector;
(2) selecting partial data from the sample set as a training set, training the BP neural network by using the training set to obtain a network model NET2, and identifying the sample set by using NET 2;
(3) for J segmented signals identified as containing targets
Figure FDA0003409557340000051
By the track number i and its segment number in the track
Figure FDA0003409557340000052
Obtaining the coordinates of the target vertical direction area in the ith channel A-Scan signal as
Figure FDA0003409557340000053
7. The method for identifying the ground penetrating radar underground cavity target based on the BP neural network as claimed in claim 1, wherein: in step S6, the segment signals identified as including the target are marked in their corresponding vertical and horizontal areas to obtain the position of the target.
CN202111524442.2A 2021-12-14 2021-12-14 Ground penetrating radar underground cavity target identification method based on BP neural network Pending CN114187516A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111524442.2A CN114187516A (en) 2021-12-14 2021-12-14 Ground penetrating radar underground cavity target identification method based on BP neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111524442.2A CN114187516A (en) 2021-12-14 2021-12-14 Ground penetrating radar underground cavity target identification method based on BP neural network

Publications (1)

Publication Number Publication Date
CN114187516A true CN114187516A (en) 2022-03-15

Family

ID=80543636

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111524442.2A Pending CN114187516A (en) 2021-12-14 2021-12-14 Ground penetrating radar underground cavity target identification method based on BP neural network

Country Status (1)

Country Link
CN (1) CN114187516A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115113168A (en) * 2022-08-25 2022-09-27 南京宇安防务科技有限公司 Radar clutter suppression method based on neural network
CN115220036A (en) * 2022-09-21 2022-10-21 江苏筑升土木工程科技有限公司 Intelligent road cavity online detection system and detection method
CN117079145A (en) * 2023-10-17 2023-11-17 深圳市城市交通规划设计研究中心股份有限公司 Comprehensive road condition evaluation method, electronic equipment and storage medium

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115113168A (en) * 2022-08-25 2022-09-27 南京宇安防务科技有限公司 Radar clutter suppression method based on neural network
CN115220036A (en) * 2022-09-21 2022-10-21 江苏筑升土木工程科技有限公司 Intelligent road cavity online detection system and detection method
CN117079145A (en) * 2023-10-17 2023-11-17 深圳市城市交通规划设计研究中心股份有限公司 Comprehensive road condition evaluation method, electronic equipment and storage medium
CN117079145B (en) * 2023-10-17 2024-03-26 深圳市城市交通规划设计研究中心股份有限公司 Comprehensive road condition evaluation method, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN114187516A (en) Ground penetrating radar underground cavity target identification method based on BP neural network
CN112462346B (en) Ground penetrating radar subgrade disease target detection method based on convolutional neural network
CN111476088B (en) Asphalt pavement water damage identification model construction method, identification method and system
CN102096821B (en) Number plate identification method under strong interference environment on basis of complex network theory
CN107346420A (en) Text detection localization method under a kind of natural scene based on deep learning
CN101329402B (en) Multi-dimension SAR image edge detection method based on improved Wedgelet
CN110717464B (en) Intelligent railway roadbed disease identification method based on radar data
CN102830404B (en) Method for identifying laser imaging radar ground target based on range profile
CN110308444B (en) Road horizon intelligent identification and interference source elimination method
CN111428631B (en) Visual identification and sorting method for unmanned aerial vehicle flight control signals
CN107450054B (en) A kind of adaptive Coherent Noise in GPR Record denoising method
CN108053412B (en) Method for determining optimal segmentation result of remote sensing image by fusing multiple quantitative evaluation indexes
CN108830140B (en) Volcanic lithology identification method based on electric imaging logging fractal dimension
CN108171119B (en) SAR image change detection method based on residual error network
CN111160199A (en) Highway disaster information detection method based on high-resolution remote sensing image
CN111208479B (en) Method for reducing false alarm probability in deep network detection
CN111025285B (en) Asphalt pavement water damage detection method based on map gray scale self-adaptive selection
CN112818876B (en) Electromagnetic signal extraction and processing method based on deep convolutional neural network
CN112529046A (en) Method for identifying internal disease damage of three-dimensional ground penetrating radar asphalt pavement
CN110426745B (en) Millimeter wave image foreign matter detection method based on block mixing Gaussian low-rank matrix decomposition
CN111091049A (en) Road surface obstacle detection method based on reverse feature matching
CN108319927B (en) Method for automatically identifying diseases
CN110689538B (en) Tunnel lining crack image detection method
CN114355339A (en) Method and system for identifying radar spectrum of road surface void disease
CN108776968B (en) SAR image change detection method based on deep forest

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