CN108896990A - A kind of building masonry wall imaging method and device using coupled mode dictionary learning - Google Patents

A kind of building masonry wall imaging method and device using coupled mode dictionary learning Download PDF

Info

Publication number
CN108896990A
CN108896990A CN201810444171.1A CN201810444171A CN108896990A CN 108896990 A CN108896990 A CN 108896990A CN 201810444171 A CN201810444171 A CN 201810444171A CN 108896990 A CN108896990 A CN 108896990A
Authority
CN
China
Prior art keywords
dictionary
wall
imaging
signal
vector
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
CN201810444171.1A
Other languages
Chinese (zh)
Other versions
CN108896990B (en
Inventor
晋良念
冯飞
谢辉玉
纪元法
孙希延
刘庆华
谢跃雷
蒋俊正
欧阳缮
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.)
Guilin University of Electronic Technology
Original Assignee
Guilin University of Electronic 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 Guilin University of Electronic Technology filed Critical Guilin University of Electronic Technology
Priority to CN201810444171.1A priority Critical patent/CN108896990B/en
Publication of CN108896990A publication Critical patent/CN108896990A/en
Application granted granted Critical
Publication of CN108896990B publication Critical patent/CN108896990B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

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
    • 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
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • 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

Abstract

The present invention is suitable for building masonry wall imaging field, provides a kind of building masonry wall imaging method and device using coupled mode dictionary learning.The method includes：The echo-signal of the target construction wall of receiving antenna pulse radar system acquisition；Compression sampling, which is carried out, by echo-signal of the random measurement matrix to target construction wall obtains observation vector；It is designed like dictionary according to free-space propagation characteristic, and compressed sensing model is constructed based on observation vector and imaging dictionary；Building masonry wall imaging is obtained by coupled mode Bayes's dictionary learning algorithm by the sparse signal vector reconstruction of image after wall according to compressed sensing model.The present invention integrally improves the reconstruction performance of wall extension target reflection factor in SOI compared to existing method, preferably solving unknown quantity, position, wall parameter and measurement noise leads to dictionary and image scene mismatch and wall astigmatism coke, positional shift occurs, while effectively reducing the required data volume of imaging.

Description

A kind of building masonry wall imaging method and device using coupled mode dictionary learning
Technical field
The invention belongs to building masonry wall imaging field more particularly to a kind of buildings using coupled mode dictionary learning Wall imaging method and device.
Background technique
Building layout imaging radar penetrates building by emitting electromagnetic wave, and then realizes the position to building masonry wall It sets, detected and be imaged towards equal distribution information, rescue in disaster, the fields such as city street fighting, anti-riot anti-terrorism, enemy's situation are scouted With important application value.Traditional through-wall radar imaging algorithm is mainly based on backward projection imaging method, although the party Method is simple and efficient, but its image quality needs intensive spatial sampling and time sampling and huge storing data space ability It is guaranteed.
Under normal circumstances, the synthetic aperture array or real array that detection fabric structure layout uses are parallel to front wall , the imaging for being parallel to all walls of front wall so necessarily becomes the main body of area-of-interest.At a distance from area-of-interest To dimension compare, the quantity for being parallel to the wall of array is relatively fewer, and wall has sparsity in the region of interest, institute Imaging with wall is actually the Problems of Reconstruction of a sparse signal.Although this sparse imaging method can efficiently reduce Data volume needed for imaging simultaneously saves signal bandwidth, however its recovery algorithms has higher requirement to the signal-to-noise ratio of echo-signal, Meanwhile it is contemplated that the linearly continuous structural distribution of wall, then interested target wall echo is actually also It is a structural signal sparse with block, thus there are a kind of relevances each other for sparse vector element.
In addition, requirement must construct accurate and matched dictionary matrix in sparse imaging process, and construct such word The accurate focusing time delay of allusion quotation matrix requirements and known wall parameter.Although the prior art gives many typical focusing time delays Calculation method, but the time delay obtained by these methods can not be equal to true electromagnetic wave propagation time delay completely, in addition, building Layout imaging often can not accurately known wall parameter, it is clear that accurately compensates from wall parameter Estimation or time delay merely The dictionary matrix that angle constructs can not be exactly matched with image scene, directly be easy for occurring using the imaging of this dictionary Wall target is out of focus, and deviates actual position.
Summary of the invention
The purpose of the present invention is to provide a kind of building masonry wall imaging method, devices using coupled mode dictionary learning And computer readable storage medium, it is intended to solve in the imaging of complicated case building layout, because of unknown quantity, position, wall Parameter and measurement noise lead to dictionary and image scene mismatch and the problem of burnt wall astigmatism, positional shift occur.
In a first aspect, the present invention provides a kind of building masonry wall imaging method using coupled mode dictionary learning, institute The method of stating includes：
The echo-signal of the target construction wall of receiving antenna pulse radar system acquisition；
Compression sampling, which is carried out, by echo-signal of the random measurement matrix to target construction wall obtains observation vector；
It is designed like dictionary according to free-space propagation characteristic, and based on observation vector and imaging dictionary building compression sense Perception model；
According to compressed sensing model, the sparse signal of image after wall is sweared by coupled mode Bayes's dictionary learning algorithm Amount reconstruct, to obtain building masonry wall imaging.
Second aspect, the present invention provides a kind of building masonry wall imaging device using coupled mode dictionary learning, institutes Stating device includes：
Receiving module, the echo-signal of the target construction wall for receiving antenna pulse radar system acquisition；
Compression sampling module is adopted for carrying out compression by echo-signal of the random measurement matrix to target construction wall Sample obtains observation vector；
Module is constructed, for being designed like dictionary according to free-space propagation characteristic, and is based on observation vector and imaging Dictionary constructs compressed sensing model；
Reconstructed module, for being schemed after wall by coupled mode Bayes's dictionary learning algorithm according to compressed sensing model The sparse signal vector reconstruction of picture, to obtain building masonry wall imaging.
The third aspect, the present invention provides a kind of computer readable storage medium, the computer readable storage medium is deposited Computer program is contained, is realized when the computer program is executed by processor such as the above-mentioned dictionary learning using coupled mode The step of building masonry wall imaging method.
In the present invention, due to passing through coupled mode Bayes dictionary learning algorithm for the sparse signal vector of image after wall Reconstruct, presets the prior informations such as noise variance and degree of rarefication without requiring, and the structure sparsity of coupled mode is melted Enter wherein, wall in SOI (Scene of interest, scene of interest) is integrally improved compared to existing method and extends mesh Mark reflection coefficient reconstruction performance, can preferably solve unknown quantity, position, wall parameter and measurement noise cause dictionary with Image scene mismatch and there is the problems such as burnt wall astigmatism, positional shift, while data volume needed for imaging can be effectively reduced.
Detailed description of the invention
Fig. 1 is the stream for the building masonry wall imaging method using coupled mode dictionary learning that the embodiment of the present invention one provides Cheng Tu.
Fig. 2 is the imitative of the building masonry wall imaging method using coupled mode dictionary learning that the embodiment of the present invention one provides True mode schematic diagram.
Fig. 3 be the embodiment of the present invention one provide the building masonry wall imaging method using coupled mode dictionary learning in, Electromagnetic wave propagation model schematic diagram through walls.
Fig. 4 is the imitative of the building masonry wall imaging method using coupled mode dictionary learning that the embodiment of the present invention one provides True scene imaging result schematic diagram.
Fig. 5 is the function of the building masonry wall imaging device provided by Embodiment 2 of the present invention using coupled mode dictionary learning It can module frame chart.
Specific embodiment
In order to which the purpose of the present invention, technical solution and beneficial effect is more clearly understood, below in conjunction with attached drawing and implementation Example, the present invention will be described in further detail.It should be appreciated that specific embodiment described herein is only used to explain this hair It is bright, it is not intended to limit the present invention.
In order to illustrate technical solutions according to the invention, the following is a description of specific embodiments.
Embodiment one：
Referring to Fig. 1, the building masonry wall imaging side using coupled mode dictionary learning that the embodiment of the present invention one provides Method includes the following steps：It is noted that if having substantially the same as a result, of the invention using coupled mode dictionary learning Building masonry wall imaging method is not limited with process sequence shown in FIG. 1.
S101, receiving antenna pulse radar system acquisition target construction wall echo-signal.
In the embodiment of the present invention one, the echo-signal of the target construction wall acquires in the following manner：
The antenna that antenna impulse radar system uses transmitting-receiving to set altogether scans target construction wall in a manner of synthetic aperture The imaging region of body, as shown in Fig. 2, antenna is expressed as the narrow pulse signal of s (t), s (t)=exp (- 2 π using transmitting signal2f2 (t-f)2), wherein f indicates the centre frequency of transmitting signal；
The echo-signal that the antenna set altogether measures Aperture receiving target construction wall at L respectively is received and dispatched, L is measured hole The quantity of diameter, first of received echo-signal in measurement aperture are expressed as rl(t)=sl(t)+el(t), l ∈ 1,2 ... L, wherein sl (t) and el(t) target wall reflecting component and noise component(s) are respectively indicated, wherein target wall reflecting componentWherein, τl；(i,j)Aperture is measured to the double of pixel (i, j) from first for signal Journey propagation delay, ζ[l；(i,j)]For indicator function, if pixel (i, j) measures the front in aperture, ζ at first[l；(i,j)]Value It is 1, other situation ζ[l；(i,j)]Value is 0.
S102, compression sampling acquisition observation is carried out by echo-signal of the random measurement matrix to target construction wall Vector.
In the embodiment of the present invention one, S102 can specifically include following steps：
The echo-signal r of target construction wall that aperture receive is measured L respectivelyl(t) sampling of K point is carried out, Echo-signal r after being sampledl=[rl(0),rl(1),…,rl(K-1)]T；
By constructing gaussian random calculation matrix ΦlTo the echo-signal r after each samplinglCompression sampling is done, is obtained each Measure the vector y in aperturel, yl=Φlrl, wherein ylIt is the column vector of kL × 1, k indicates random matrix from original sample point K Extract k sampled point, ΦlIt is the sampling matrix of k × K；
Total observation vector is calculated according to the vector in each measurement apertureIts dimension is M × 1, M=kL, wherein being expressed as：yl=Φlrl。
S103, dictionary is designed like according to free-space propagation characteristic, and based on observation vector and imaging dictionary building Compressed sensing model.
In the embodiment of the present invention one, S103 can specifically include following steps：
S1031, by the imaging region of target construction wall distance to be divided into N in orientationx×NyA pixel, Distance has N to orientation respectivelyyAnd NxA grid cell, the coordinate of any one grid cell are (i, j), i ∈ 1,2 ... Nx, j ∈ 1,2 ... Ny, poly- between the pixel of related domain information and each grid cell based on echo-signal and each measurement aperture Burnt time delay constructs L imaging dictionary Ψl；
S1032, compressed sensing model is constructed based on total observation vector, L imaging dictionary and L calculation matrix.
Wherein, S1031 is specifically as follows：
N=N is turned to by the imaging region of target construction wall is discretex×NyAfter a mesh point, grid distance unit it is big Small to represent radar resolution, image pixel informations all in this way are included into a dimensional vector, i.e. sparse signal vector σ, and one The dimension size of dimensional vector is N × 1；
rlIt is considered as the echo information superposition that each pixel generates, dictionary Ψ is imaged using the characteristic Design of each pixell, rl=Ψlσ, ΨlThe imaging dictionary of radar system at first of measurement aperture, dimension is K × N, first measurement aperture at As the row k vector of dictionary is expressed as：Wherein, τl；(i,j)Aperture is measured to the round trip propagation delay of pixel (i, j) from first for signal, which can electromagnetism according to Fig.3, Wave propagation model through walls calculates, and specific calculating is as follows：
The dielectric constant and thickness of wall are respectively d and εr, the coordinate of any point P is P (x after wallp,yp), B point is electricity The refraction point of magnetic wave, coordinate are B (x2,y2), the coordinate of antenna is A (xa,ya), θ1WithFolding when respectively electromagnetic wave is through walls Firing angle and incidence angle, from the figure 3, it may be seen that transmitting antenna is to the electrical length between point PR thereinABWith rBPIndicate launch point to refraction point and refraction point to the linear distance of picture point P, and
When picture point and antenna distance are much larger than thickness of wall body, it is approximately considered line segment PQ and PA and is parallel to each other, then at this time RBPIt is expressed as rBP=rAPΔ r, wherein Δ r indicates the length difference of this two sections of line segments, lAPIndicate A point and P point it Between linear distance, value be in summary various, lAPIt is expressed as
τ at this timel；(i,j)It is calculated as τl；(i,j)=lAPWall is imaged in actual building layout imaging applications in/c The distance between antenna is much larger than thickness of wall body, so θ1Approximate representation is θ1≈arctan(|xp-xa|/|yp-ya|)。
S1032 is specifically as follows：
With L gaussian random calculation matrix ΦlOverall measurement matrix Φ is constituted, with L imaging dictionary ΨlConstitute total dictionary Ψ =[Ψ1 Ψ2 … ΨL]T, total observation vectorY, Φ, Ψ and σ's in compressed sensing model Relationship is expressed as：Y=Φ Ψ σ=D σ, wherein Ψ is multiplied with Φ indicates that row and column is randomly selected from dictionary Ψ constitutes new word Allusion quotation matrix D solves sparse signal vector σ by way of dictionary learning.
S104, according to compressed sensing model, by coupled mode Bayes's dictionary learning algorithm by after wall image it is sparse Signal phasor σ reconstruct, to obtain building masonry wall imaging.
In the embodiment of the present invention one, S104 can specifically include following steps：
Sparse Signal Representation model is：Y=D σ+w, wherein y is total observation vector, and D indicates random from dictionary Ψ It extracts row and column and constitutes new dictionary matrix, σ is imaging sparse signal vector, and w is the additive noise in imaging process, and additivity is made an uproar Sound obedience p (w | γ)~N (0, γ-1I) Gaussian Profile.
In order to need to be introduced into the correlation between element using the structural attributes of the echo-signal of target construction wall Property in, specifically operation be the element σ for making σnPrior distribution depend not only on itself hyper parameter, also to examine Consider its adjacent σ in front and backn+1、σn-1The influence of the hyper parameter of element, so the probability distribution of σ is expressed as：In formula, the coefficient of coup of the λ between adjacent element meets 0≤λ ≤ 1, αnIndicate the hyper parameter (α therein of non-negative sparse control0=αN+1=0), and the distribution of the Gamma known to parameter a and b is made For hyper parameter a prior distribution, i.e.,
Scene is matched to describe the pronunciation dictionary adaptation of imaging process, while limiting the element mistake in dictionary in an iterative process Greatly, each atom in dictionary D is enabled to obey Gaussian Profile independent of each other, i.e.,In formula, β is indicated The variance for the dictionary atom known.
According to mean field theory, the Posterior probability distribution of parameter can be solved and be converted into seeking single parameter APPROXIMATE DISTRIBUTION Solution, that is, by creep quantity set { D, σ, alpha, gammawPosterior probability density function p (D, σ, α, γw| it y) is decomposed into a series of mutual The product form of independent approximation Posterior distrbutionp function, is expressed as
First to qσ(σ) is solved, and the probability density function of logarithm is expressed asHerein<·>Indicate mathematic expectaion, the hereafter expression in formula Symbol is in this way, the Posterior probability distribution due to latent variable can be expressed as following form：Abbreviation can obtain after ignoring the variable unrelated with σ
In formula,<Λ>=diag (<α1>+ρ1(<α0>+<α2>),…,<αN>+ρN(<αN-1>+<αN+1>)), wherein<αn>Table Show by qαThe desired value that (α) is obtained,<γw>,<D>With<DTD>Respectively indicate byAnd qd(D) desired value obtained.From upper Q known to formulaσ(σ) obey mean value be μ=<γw>Σ<D>TY, variance be Σ=(<γw><DTD>+<Λ>)-1Gauss point Cloth, i.e. qσ(σ)=N (σ | μ, Σ).As can be seen that be asked<σ>=μ, it with<D>,<DTD>、<αn>, n=1,2 ..., N and< γw>It is related, therefore calculate<σ>These parameters must be solved in advance.
Next, solving qd(D) Posterior distrbutionp, ignoresIn The amount unrelated with D, abbreviation can obtain
In formula, A=(<γw><σσT>+β-1I)-1, C=<γw>y<σ>T, dmAnd cmRespectively represent the of matrix D and Matrix C M row.In addition, in A<σσT>=<σ><σ>T+Σ.The every a line of dictionary matrix D is all independent from each other, and its every row is all obeyed Mean value is cmA, covariance matrix are the Gaussian Profile of C, i.e.,Thus,<D>=CA,<DTD> =<D>T<D>+M<A>。
Then, q is similarly solvedαThe distribution of (α), the lnq of available logarithmic formαThe approximate Posterior distrbutionp of (α)：
Further, it can obtain
In formula, αnObey parameterWithGamma probability distribution, and αn-1And αn+1Take It is 0.5 ρ from parametern+ 1,Gamma distribution.In view of having coupled relation between the element in α, Therefore each is being updated<αn>When, it is also necessary to it updates<αn-1>With<αn+1>.Wherein,Hereμ (n) therein Indicate the nth elements of μ, Σ (n, n) indicates n-th of diagonal element of Σ.
Finally, can similarly obtainApproximate Posterior distrbutionp meet：WhereinAndInFurther abbreviation isDue toObey Gamma Distribution, it can thus be concluded that γwUpdated value be
Calculate vectorAfterwards, according to each grid of correspondence with NxN is carried out for the periodySecondary assignment, obtains building masonry wall Imaging results.
The citing for the building masonry wall imaging method using coupled mode dictionary learning that the embodiment of the present invention one provides is such as Under：
The four sides wall that length and width dimensions are 3m × 2m is made of uniform dielectric material, wherein including the small of a 1.5m × 1m Room, wall thickness, relative dielectric constant and conductivity are 0.2m, 6.4 and 0.01S/m respectively.The antenna set altogether using transmitting-receiving At wall 1.5m, along the detection of front wall synthetic aperture is parallel to, one echo signal reception point is set every 0.1m, is shared 24 are received and dispatched the antenna element synthesizing linear array set altogether, are evenly distributed at horizontal axis 0.8-3.1m, longitudinal axis 0.1m.In emulation, The central instant and pulse width of the narrow Gaussian pulse signal of transmitting are 1ns, the grid cell of GPRMAX, time step and are adopted Sample time window is 0.01m, 23ps and 30ns respectively.
200 sampling point methods building calculation matrix Φ of each receiving antenna are randomly choosed, according to free-space propagation mould Type establishes initial Ψ0, and then initial dictionary D is obtained, 10 are all set as in order not to excessively bring prior information parameter a, b, c, d into-6, Such parameter setting makes preparatory Gamma distribution be intended to be uniformly distributed, while β value is set as the larger value 108So that each Atom all obeys spindleless roller, in addition γwInitial value be set as 10-3。
Simulation result is as shown in Figure 4.Black wire indicates wall actual profile in Fig. 4, and grey threadiness is expressed as the wall of picture Body profile, from fig. 4, it can be seen that making full use of the correlation of wall reflection coefficient during signal reconstruction, and while iteration The true dictionary of scene is gradually approached constantly in a manner of dictionary learning, so imaging wall front and rear surfaces are all relatively more coherent, Wall contour edge is shown obviously, and wall position is more accurate, and clutter has obtained preferable inhibition.
Embodiment two：
Referring to Fig. 5, dress is imaged in the building masonry wall provided by Embodiment 2 of the present invention using coupled mode dictionary learning Set including：
Receiving module 11, the echo letter of the target construction wall for receiving antenna pulse radar system acquisition Number；
Compression sampling module 12, for being compressed by echo-signal of the random measurement matrix to target construction wall Sampling obtains observation vector；
Construct module 13, for being designed like dictionary according to free-space propagation characteristic, and based on observation vector and at As dictionary constructs compressed sensing model；
Reconstructed module 14 is used for according to compressed sensing model, will be after wall by coupled mode Bayes's dictionary learning algorithm The sparse signal vector reconstruction of image, to obtain building masonry wall imaging.
Building masonry wall imaging device and the present invention provided by Embodiment 2 of the present invention using coupled mode dictionary learning What embodiment one provided belongs to same design using the building masonry wall imaging method of coupled mode dictionary learning, implements Process is detailed in specification full text, and details are not described herein again.
Embodiment three：
The embodiment of the present invention three provides a kind of computer readable storage medium, the computer-readable recording medium storage There is computer program, is realized when the computer program is executed by processor as what the embodiment of the present invention one provided utilizes coupled mode The step of building masonry wall imaging method of formula dictionary learning.
In the present invention, due to passing through coupled mode Bayes dictionary learning algorithm for the sparse signal vector of image after wall Reconstruct, presets the prior informations such as noise variance and degree of rarefication without requiring, and the structure sparsity of coupled mode is melted Enter wherein, the reconstruction performance of wall extension target reflection factor in SOI is integrally improved compared to existing method, can preferably be solved Certainly unknown quantity, position, wall parameter and measurement noise leads to dictionary and image scene mismatch and wall astigmatism coke, position occurs The problems such as setting offset, while the required data volume of imaging can be effectively reduced.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of above-described embodiment is can It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage Medium may include：Read-only memory (ROM, Read Only Memory), random access memory (RAM, Random Access Memory), disk or CD etc..
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.

Claims (10)

1. a kind of building masonry wall imaging method using coupled mode dictionary learning, which is characterized in that the method includes：
The echo-signal of the target construction wall of receiving antenna pulse radar system acquisition；
Compression sampling, which is carried out, by echo-signal of the random measurement matrix to target construction wall obtains observation vector；
It is designed like dictionary according to free-space propagation characteristic, and compressed sensing mould is constructed based on observation vector and imaging dictionary Type；
According to compressed sensing model, by coupled mode Bayes's dictionary learning algorithm by the sparse signal vector weight of image after wall Structure, to obtain building masonry wall imaging.
2. the method as described in claim 1, which is characterized in that the echo-signal of the target construction wall is by following What mode acquired：
The antenna that antenna impulse radar system uses transmitting-receiving to set altogether scans target construction wall in a manner of synthetic aperture Imaging region, antenna are expressed as the narrow pulse signal of s (t), s (t)=exp (- 2 π using transmitting signal2f2(t-f)2), wherein f Indicate the centre frequency of transmitting signal；
The echo-signal that the antenna set altogether measures Aperture receiving target construction wall at L respectively is received and dispatched, L is measurement aperture Quantity, first of received echo-signal in measurement aperture are expressed as rl(t)=sl(t)+el(t), l ∈ 1,2 ... L, wherein sl(t) And el(t) target wall reflecting component and noise component(s) are respectively indicated, wherein target wall reflecting componentWherein, τl；(i,j)Aperture is measured to the double of pixel (i, j) from first for signal Journey propagation delay, ζ[l；(i,j)]For indicator function, if pixel (i, j) measures the front in aperture, ζ at first[l；(i,j)]Value It is 1, other situation ζ[l；(i,j)]Value is 0.
3. method according to claim 2, which is characterized in that it is described by random measurement matrix to target construction wall Echo-signal carries out compression sampling acquisition observation vector and specifically includes：
The echo-signal r of target construction wall that aperture receive is measured L respectivelyl(t) sampling for carrying out K point, is adopted Echo-signal r after samplel=[rl(0),rl(1),…,rl(K-1)]T；
By constructing gaussian random calculation matrix ΦlTo the echo-signal r after each samplinglCompression sampling is done, each measurement is obtained The vector y in aperturel, yl=Φlrl, wherein ylIt is the column vector of kL × 1, k indicates that random matrix extracts k from original sample point K A sampled point, ΦlIt is the sampling matrix of k × K；
Total observation vector is calculated according to the vector in each measurement apertureIts dimension is M × 1, M= KL, wherein being expressed as：yl=Φlrl。
4. method as claimed in claim 3, which is characterized in that it is described to be designed like dictionary according to free-space propagation characteristic, And it is specifically included based on observation vector and imaging dictionary building compressed sensing model：
By the imaging region of target construction wall distance to be divided into N in orientationx×NyA pixel is believed based on echo Number related domain information and each grid cell pixel and each measurement aperture between focusing time delay construct L imaging dictionary Ψl；
Compressed sensing model is constructed based on total observation vector, L imaging dictionary and L calculation matrix.
5. method as claimed in claim 4, which is characterized in that the imaging region by target construction wall distance to With N is divided into orientationx×NyA pixel, the pixel of related domain information and each grid cell based on echo-signal and each The focusing time delay measured between aperture constructs L imaging dictionary ΨlSpecially：
N=N is turned to by the imaging region of target construction wall is discretex×NyA mesh point, distance have N to orientation respectivelyy And NxA grid cell, the coordinate of any one grid cell are (i, j), i ∈ 1,2 ... Nx, j ∈ 1,2 ... Ny, grid distance The size of unit represents radar resolution, and all image pixel informations are included into a dimensional vector, i.e. sparse signal vector σ, And one the dimension size of dimensional vector be N × 1；
rlIt is considered as the echo information superposition that each pixel generates, dictionary Ψ is imaged using the characteristic Design of each pixell, rl= Ψlσ, ΨlIt is the imaging dictionary of radar system at first of measurement aperture, dimension is K × N, the imaging word in first of measurement aperture The row k vector of allusion quotation is expressed as：Wherein, τl；(i,j) Round trip propagation delay of the aperture to pixel (i, j), the value propagation model meter through walls according to electromagnetic wave are measured from first for signal It calculates.
6. method as claimed in claim 5, which is characterized in that specifically calculated according to electromagnetic wave propagation model through walls as follows：
The dielectric constant and thickness of wall are respectively d and εr, the coordinate of any point P is P (x after wallp,yp), B point is electromagnetic wave Refraction point, coordinate be B (x2,y2), the coordinate of antenna is A (xa,ya), θ1WithRefraction angle when respectively electromagnetic wave is through walls With incidence angle, transmitting antenna is to the electrical length between point PR thereinABWith rBPIndicate that launch point arrives Refraction point and refraction point to picture point P linear distance, and
When picture point and antenna distance are much larger than thickness of wall body, it is approximately considered line segment PQ and PA and is parallel to each other, then r at this timeBP It is expressed as rBP=rAPΔ r, wherein Δ r indicates the length difference of this two sections of line segments, lAPIndicate linear distance between A point and P point, Value isIn summary various, lAPIt is expressed as
τ at this timel；(i,j)It is calculated as τl；(i,j)=lAPWall and antenna is imaged in actual building layout imaging applications in/c The distance between be much larger than thickness of wall body, so θ1Approximate representation is θ1≈arctan(|xp-xa|/|yp-ya|)。
7. method as claimed in claim 6, which is characterized in that described based on total observation vector, L imaging dictionary and L A calculation matrix constructs compressed sensing model：
With L gaussian random calculation matrix ΦlOverall measurement matrix Φ is constituted, with L imaging dictionary ΨlConstitute total dictionary Ψ= [Ψ1 Ψ2 … ΨL]T, total observation vectorThe pass of y, Φ, Ψ and σ in compressed sensing model System is expressed as：Y=Φ Ψ σ=D σ, wherein Ψ is multiplied with Φ indicates that row and column is randomly selected from dictionary Ψ constitutes new dictionary Matrix D solves sparse signal vector σ by way of dictionary learning.
8. the method for claim 7, which is characterized in that it is described according to compressed sensing model, pass through coupled mode pattra leaves This dictionary learning algorithm specifically includes the sparse signal vector reconstruction of image after wall to obtain building masonry wall imaging：
Sparse Signal Representation model is：Y=D σ+w, wherein y is total observation vector, and D expression is randomly selected from dictionary Ψ Row and column constitutes new dictionary matrix, and σ is imaging sparse signal vector, and w is the additive noise in imaging process, additive noise clothes From p (w | γ)~N (0, γ-1I) Gaussian Profile；
The probability distribution of σ is expressed as：In formula, λ is adjacent element Between the coefficient of coup, meet 0≤λ≤1, αnIndicate the hyper parameter of non-negative sparse control, α0=αN+1=0, with parameter a and b The Gamma distribution known is as hyper parameter α prior distribution, i.e.,
Each atom in dictionary D is enabled to obey Gaussian Profile independent of each other, i.e.,In formula, β is indicated The variance of known dictionary atom；
According to mean field theory, the solution being converted into single parameter APPROXIMATE DISTRIBUTION is solved to the Posterior probability distribution of parameter, It is exactly by creep quantity set { D, σ, α, γwPosterior probability density function p (D, σ, α, γw| it y) is decomposed into a series of mutually indepedent Approximate Posterior distrbutionp function product form, be expressed as
First to qσ(σ) is solved, and the probability density function of logarithm is expressed asHerein<·>Mathematic expectaion is indicated, since the posteriority of latent variable is general Rate distribution is expressed as following form：Ignore the variable unrelated with σ Abbreviation obtains afterwards
In formula,<Λ>=diag (<α1>+ρ1(<α0>+<α2>),…,<αN>+ρN(<αN-1>+<αN+1>)), wherein<αn>Indicate by qαThe desired value that (α) is obtained,<γw>,<D>With<DTD>Respectively indicate byAnd qd(D) desired value obtained；It can from above formula Know qσ(σ) obey mean value be μ=<γw>Σ<D>TY, variance be Σ=(<γw><DTD>+<Λ>)-1Gaussian Profile, i.e., qσ(σ)=N (σ | μ, Σ)；
Next, solving qd(D) Posterior distrbutionp, ignoresIn with D without The amount of pass, abbreviation obtain
In formula, A=(< γw><σσT〉+β-1I)-1, C=< γw〉y〈σ〉T, dmAnd cmRespectively represent the m row of matrix D and Matrix C, A In < σ σT>=<σ><σ>T+Σ；The every a line of dictionary matrix D is all independent from each other, and it is c that its every row, which all obeys mean value,mA, Covariance matrix is the Gaussian Profile of C, i.e.,Thus it obtains, < D>=CA,<DTD>=<D>T<D>+ M<A>；
Then, q is similarly solvedαThe distribution of (α) obtains the lnq of logarithmic formαThe approximate Posterior distrbutionp of (α)：
Further, it obtains
In formula, αnObey parameterWithGamma probability distribution, and αn-1And αn+1Obey ginseng Number is 0.5 ρn+ 1,Gamma distribution；In view of having coupled relation between the element in α, each is being updated< αn>When, it updates simultaneously<αn-1<With<αn+1>；Wherein, Whereinμ (n) indicates the nth elements of μ, and Σ (n, n) indicates n-th of diagonal element of Σ；
Finally, obtainingApproximate Posterior distrbutionp meet：WhereinAndInFurther abbreviation isDue toObey Gamma Distribution, thus obtains γwUpdated value be
Calculate vectorAfterwards, according to each grid of correspondence with NxN is carried out for the periodySecondary assignment obtains the imaging of building masonry wall As a result.
9. a kind of building masonry wall imaging device using coupled mode dictionary learning, which is characterized in that described device includes：
Receiving module, the echo-signal of the target construction wall for receiving antenna pulse radar system acquisition；
Compression sampling module is obtained for carrying out compression sampling by echo-signal of the random measurement matrix to target construction wall Obtain observation vector；
Module is constructed, for being designed like dictionary according to free-space propagation characteristic, and based on observation vector and imaging dictionary Construct compressed sensing model；
Reconstructed module, for according to compressed sensing model, by coupled mode Bayes's dictionary learning algorithm by image after wall Sparse signal vector reconstruction, to obtain building masonry wall imaging.
10. a kind of computer readable storage medium, the computer-readable recording medium storage has computer program, and feature exists In the computer program realizes utilization coupled mode dictionary as claimed in any one of claims 1 to 8 when being executed by processor The step of building masonry wall imaging method of study.
CN201810444171.1A 2018-05-10 2018-05-10 Building wall imaging method and device by using coupled mode dictionary learning Active CN108896990B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810444171.1A CN108896990B (en) 2018-05-10 2018-05-10 Building wall imaging method and device by using coupled mode dictionary learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810444171.1A CN108896990B (en) 2018-05-10 2018-05-10 Building wall imaging method and device by using coupled mode dictionary learning

Publications (2)

Publication Number Publication Date
CN108896990A true CN108896990A (en) 2018-11-27
CN108896990B CN108896990B (en) 2022-06-03

Family

ID=64343942

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810444171.1A Active CN108896990B (en) 2018-05-10 2018-05-10 Building wall imaging method and device by using coupled mode dictionary learning

Country Status (1)

Country Link
CN (1) CN108896990B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109799499A (en) * 2019-01-28 2019-05-24 沈阳航空航天大学 A kind of through-wall radar wall method for parameter estimation
CN111665500A (en) * 2020-06-12 2020-09-15 沈阳航空航天大学 Pulse through-wall radar imaging method based on single-bit compressed sensing
CN111766575A (en) * 2020-06-08 2020-10-13 桂林电子科技大学 Through-wall radar self-focusing sparse imaging method and computer equipment
CN112198506A (en) * 2020-09-14 2021-01-08 桂林电子科技大学 Method, device and system for learning and imaging ultra-wideband through-wall radar and readable storage medium
CN113219432A (en) * 2021-05-14 2021-08-06 内蒙古工业大学 Moving object detection method based on knowledge assistance and sparse Bayesian learning

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106226765A (en) * 2016-09-12 2016-12-14 桂林电子科技大学 A kind of building layout formation method and system
CN106772365A (en) * 2016-11-25 2017-05-31 南京理工大学 A kind of multipath based on Bayes's compressed sensing utilizes through-wall radar imaging method

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106226765A (en) * 2016-09-12 2016-12-14 桂林电子科技大学 A kind of building layout formation method and system
CN106772365A (en) * 2016-11-25 2017-05-31 南京理工大学 A kind of multipath based on Bayes's compressed sensing utilizes through-wall radar imaging method

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
TIAN JIN等: "Image-Domain Estimation of Wall Parameters for Autofocusing of Through-the-Wall SAR Imagery", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 *
晋良念等: "利用块间耦合稀疏贝叶斯学习的建筑物布局成像方法", 《电子与信息学报》 *

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109799499A (en) * 2019-01-28 2019-05-24 沈阳航空航天大学 A kind of through-wall radar wall method for parameter estimation
CN109799499B (en) * 2019-01-28 2023-04-28 沈阳航空航天大学 Wall parameter estimation method of through-wall radar
CN111766575A (en) * 2020-06-08 2020-10-13 桂林电子科技大学 Through-wall radar self-focusing sparse imaging method and computer equipment
CN111766575B (en) * 2020-06-08 2023-04-21 桂林电子科技大学 Self-focusing sparse imaging method of through-wall radar and computer equipment
CN111665500A (en) * 2020-06-12 2020-09-15 沈阳航空航天大学 Pulse through-wall radar imaging method based on single-bit compressed sensing
CN112198506A (en) * 2020-09-14 2021-01-08 桂林电子科技大学 Method, device and system for learning and imaging ultra-wideband through-wall radar and readable storage medium
CN112198506B (en) * 2020-09-14 2022-11-04 桂林电子科技大学 Method, device and system for learning and imaging ultra-wideband through-wall radar and readable storage medium
CN113219432A (en) * 2021-05-14 2021-08-06 内蒙古工业大学 Moving object detection method based on knowledge assistance and sparse Bayesian learning

Also Published As

Publication number Publication date
CN108896990B (en) 2022-06-03

Similar Documents

Publication Publication Date Title
CN108896990A (en) A kind of building masonry wall imaging method and device using coupled mode dictionary learning
CN106772365B (en) A kind of multipath based on Bayes&#39;s compressed sensing utilizes through-wall radar imaging method
CN103713288B (en) Sparse Bayesian reconstruct linear array SAR formation method is minimized based on iteration
CN109100718B (en) Sparse aperture ISAR self-focusing and transverse calibration method based on Bayesian learning
Lagunas et al. Joint wall mitigation and compressive sensing for indoor image reconstruction
Cuomo et al. Ultrawide-band coherent processing
CN106461775B (en) Subregion for radar system
CN105911544B (en) A kind of spread spectrum through-wall radar imaging method based on compressed sensing technology
CN103698763B (en) Based on the linear array SAR sparse formation method of hard-threshold orthogonal matching pursuit
CN105929386B (en) A kind of wave based on Higher Order Cumulants reaches method of estimation
Ning et al. A highly efficient compressed sensing algorithm for acoustic imaging in low signal-to-noise ratio environments
CN104360345B (en) High-resolution through-wall imaging system and method based on random antenna array and microwave relevance imaging principle
CN110098882A (en) Multiple antennas broadband frequency spectrum detection method based on compressed sensing and entropy
AlBeladi et al. Evaluating compressive sensing algorithms in through-the-wall radar via F1-score
Crocco et al. Uncalibrated 3D room geometry estimation from sound impulse responses
CN109696672B (en) High-resolution through-wall radar imaging method based on spatial structure correlation
Liang et al. Accurate near-field millimeter-wave imaging of concave objects—A case study of dihedral structures under monostatic array configurations
CN115453528A (en) Method and device for realizing segmented observation ISAR high-resolution imaging based on rapid SBL algorithm
Gurbuz et al. Sparse ground-penetrating radar imaging method for off-the-grid target problem
Tang et al. Indoor scene reconstruction for through-the-wall radar imaging using low-rank and sparsity constraints
CN106291480A (en) A kind of multipath mitigation algorithms based on node coupling
Duman et al. Performance analysis of compressive-sensing-based through-the-wall imaging with effect of unknown parameters
Chen et al. A hand gesture recognition method for Mmwave radar based on angle-range joint temporal feature
Leigsnering et al. CS based wall ringing and reverberation mitigation for through-the-wall radar imaging
Andrews et al. High resolution population density imaging of random scatterers with the matched filtered scattered field variance

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
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20181127

Assignee: Guilin Dazhi Technology Co.,Ltd.

Assignor: GUILIN University OF ELECTRONIC TECHNOLOGY

Contract record no.: X2022450000183

Denomination of invention: A method and device for building wall imaging using coupled mode dictionary learning

Granted publication date: 20220603

License type: Common License

Record date: 20221125

Application publication date: 20181127

Assignee: Guilin Feimin Electronic Technology Co.,Ltd.

Assignor: GUILIN University OF ELECTRONIC TECHNOLOGY

Contract record no.: X2022450000182

Denomination of invention: A method and device for building wall imaging using coupled mode dictionary learning

Granted publication date: 20220603

License type: Common License

Record date: 20221125

EE01 Entry into force of recordation of patent licensing contract