CN102253388B - Collaborative detection method for multiple unmanned underwater vehicles on basis of compressed sensing - Google Patents

Abstract

The invention provides a collaborative detection method for multiple unmanned underwater vehicles on the basis of compressed sensing, comprising the following steps of: (1) taking one of multiple UUVs (unmanned underwater vehicles) as a mater UUV on the basis of the space compression sampling of the compressed sensing; taking other UUVs as slave UUVs; carrying out binary system pseudorandom coding on each array element of sonar arrays of all the UUVs; forming a detection matrix V by the coded vectors; (2) reconstructing based on a distributed image combined with an optimal estimation, wherein each slave UUV corresponds to one sub estimator, and the UUV corresponds to a master estimator; causing the master UUV and the slave UUVs to simultaneously and independently operate to carry out optimal estimation operation on a sonar receiving signal to obtain respective detection data; and then carrying out fusion processing in the main estimator to realize the purpose of reconstructing the sonar. With the collaborative detection method, time for obtaining images by each UUV node can be effectively shortened, the problem that multiple UUV nodes can not simultaneously detect same target can be solved, and the rapidity and the accuracy of underwater target detection can be improved.

Description

How unmanned submarine navigation device collaborative detection method based on compressed sensing

Technical field

What the present invention relates to is a kind of Underwater Detection method, is specifically related to a kind of method of working in coordination with detecting underwater object based on many UUV of compressed sensing.

Background technology

Unmanned submarine navigation device (UUV, Unmanned Underwater Vehicle) is that indispensable instrument is used in following ocean development, has vast potential for future development.Because detection angle and the scope of the sonar that single UUV is entrained are limited, therefore can form cooperative detection system by many UUV, the expansion sensing range improves detection efficiency.

In many UUV cooperative detection system, small-sized phased array sonar of every upper outfit of UUV.Traditional sonar detection mode is digital beam formation method, and collection and reconstruct piece image are very consuming time.Although can shorten imaging time by the mode that reduces number of scanning line, this can reduce again the lateral resolution of image.In addition, when active probe, disturb for preventing from intersecting, a plurality of UUV nodes must carry out the detection of target in timesharing, and this has not only increased detection time, and makes moving target at the image that each UUV node obtains phase differential be arranged, and brings difficulty to data fusion.Therefore, must explore new distributed object detection mode, a plurality of UUV can be surveyed simultaneously, and shorten the time that single UUV node obtains target image.

It is theoretical a kind of new acquisition of information to occur in recent years, is called compressed sensing (Compressive Sensing, CS).Its core concept is, if signal itself or be sparse at certain transform domain can be sampled and Accurate Reconstruction to this signal with the sampling rate that is lower than Nyquist so.The sparse property of sonar target image is so that the CS theory can be used for the detection problem of sonar target.If discrete signal x ∈ is R ^NUnder transform-based ψ, be sparse, by selecting and the inconsistent observing matrix Φ of ψ ∈ R ^{M * N}(M＜N), set up and measure y=Φ x can realize the compression sampling to x.Utilize the problem of measured value y reconstruct x to be summed up as a l ₀The optimization problem of-Norm minimum, but because therefore a lot of sub-optimal algorithm appear in its unsolvability.Although these methods are more effective, propose theoretically, do not consider the noise immunity problem in the practical application.

To sum up, with the theoretical target detection that is used for sonar of CS the time, design suitable detection matrix Φ and fast and accurately the distributed reconfiguration algorithm be two key issues utilizing the CS theory.

Summary of the invention

The object of the present invention is to provide a kind of how unmanned submarine navigation device collaborative detection method based on compressed sensing that can improve rapidity and the accuracy of Underwater Target Detection.

The object of the present invention is achieved like this:

Step 1 is based on the space compression sampling of compressed sensing

With many UUV one of them as main UUV, and other UUV are as from UUV; Each array element to the sonar array of all UUV adopts scale-of-two pseudorandomcode, and these coding vectors consist of surveys matrix Φ; The coding pseudo-noise code generator produces, and random code generator is realized by shift register; The register maintenance is synchronous with system time, and its digital units carries out mould 2 and calculates within each clock period, and bit-by-bit generates pseudo-random code; The transmitting-receiving mode of data is each array element while transceiving data of sonar array, if the array number of sonar array is M, then once receives and dispatches process implementation to M the detection in a two-dimensional sector-shaped zone, and wherein, array number is for surveying the line number of matrix Φ;

Step 2 is based on the distributed image reconstruct of associating optimal estimation

Each is from the corresponding sub-estimator of UUV, the corresponding main estimator of main UUV; Main estimator and every simultaneously isolated operation of sub-estimator receive signal to sonar and carry out the optimal estimation computing, obtain detection data separately, carry out afterwards fusion treatment in main estimator, realize sonar image reconstruct; Wherein, compressed sensing signal reconstruction constraint condition in conjunction with the minimum variance criterion, in the form embedding state estimation problem with virtual detection, is proofreaied and correct the state through optimal estimation; Virtual detection trimming process realizes in the mode of iteration.

More than two steps do not isolate, the data that detect in step 1 be used for to survey are upgraded in step 2.

The present invention compared with prior art has following advantage:

1) detection mode.Compare with traditional delay-accumulation mode, once transmitting-receiving process can realize the repeatedly detection to a two-dimensional sector-shaped zone, greatly reduces the time that the phased array sonar obtains piece image, and surveys when can realize many UUV node to same target.

2) image reconstructing method.The collaborative distributed image reconstruction of surveying of many UUV is converted into the associating optimal estimation problem of signal, and the Reconstruction Constraints condition of the compressed sensing mode with virtual detection is embedded in the state estimation problem.Algorithm structure is simple, is fit to distributed treatment and in real time processing.

Description of drawings

Fig. 1 is the principle of work block diagram based on many UUV collaborative detection method of compressed sensing.

Fig. 2 is sonar detection imaging synoptic diagram.

Embodiment

Below in conjunction with accompanying drawing the specific embodiment of the present invention and process are further described.The collaborative detection process of many UUV is divided into space compression sampling and Image Reconstruction two parts, as shown in Figure 1.These two parts do not isolate, and in the process of carrying out, Image Reconstruction has partly utilized the detection data of space compression sampling section to carry out the state renewal.Two-part concrete execution in step is as follows:

1. space compression sampling

Each array element to all UUV sonar arrays adopts the scale-of-two pseudorandomcode, and the coding pseudo-noise code generator produces, and random code generator is realized by shift register.The register maintenance is synchronous with system time, and its digital units carries out mould 2 and calculates within each clock period, and bit-by-bit generates pseudo-random code.Coding vector consists of the required detection matrix Φ of back image reconstruction procedure by rows.

Each array element of sonar sends coded data (data length is N) simultaneously, and receives the detection data that returns simultaneously.The array number of sonar array is M, then once receives and dispatches M detection finishing a sector region.M coding vector consists of the detection matrix Φ ∈ R of sonar by row ^{M * N}(M＜N), then detection data

Can be expressed as

y _k＝Φx _k+n _k (1)

Wherein, x _k∈ R ^NFor treating reconstruct data; n _k∈ R ^MSurvey noise for sonar, be approximately the white Gaussian noise of zero-mean, covariance matrix is R _kThe detection data that sonar array element receives in this step

Below be used for the Image Reconstruction to the detection of a target.

Among the present invention, many UUV detecting target image reconstruction is converted into associating optimal estimation problem, and with the form realization of recursion, every recursion once reconstructs a data curve x of image _k, the number of times that recursion is estimated has determined the size of image.The structure of sonar detection image as shown in Figure 2.The process of Image Reconstruction comprises following concrete steps:

2. information distribution

Being engraved in the globalstate estimation and the estimation error variance that obtain in the main estimator during k is

With

Before carrying out next time recursion estimation, at first these information are carried out following distribution in sub-estimator

${\hat{x}}_k^i={\hat{x}}_k^g---\left(2\right)$

$P_k^i={\mathrm{\β}}_i^{-1}{P}_k^g---\left(3\right)$

Wherein,

With Represent respectively state estimation value and estimation error variance battle array in i the sub-estimator, i=1,2,, p, p are the nodes of many UUV; β _iBe the information distribution factor, satisfy

Among the present invention, adopt as follows based on the multidate information distribution method of estimation error variance

${\mathrm{\β}}_i=\frac{{\mathrm{trP}}_i}{{\mathrm{trP}}_1+\mathrm{tr}{P}_2+\·\·\·+{\mathrm{tr}P}_p+\mathrm{tr}{P}_m}=\frac{\mathrm{tr}{P}_i}{\underset{i=1}{\overset{p,m}{\mathrm{\Σ}}}\mathrm{tr}{P}_i}--\left(4\right)$

In the formula: P _iBe i sub-estimator to the estimation error variance of state, m represents main estimator, the mark of tr representing matrix.

After information distributed in sub-estimator, every independent state estimation of sub-estimator specifically comprised step 3 and step 4.Because implementation is identical in each estimator, is formula of reduction, below ignores superscript symbol i.

3. the time upgrades and surveys and upgrade

At k constantly, by the known state valuation

With estimation error variance battle array P _k, prediction k+1 state constantly

With estimation error variance battle array P _K+1kAs follows

${\hat{x}}_{k+1|k}=A{\hat{x}}_k---\left(5\right)$

P _k+1k＝AP _kA ^T+Q _k (6)

In the formula (5), the prediction of state has been utilized signal model

x _k+1＝Ax _k+w _k (7)

In the formula: A ∈ R ^{N * N}Be state-transition matrix, its value is determined by beam synthesizing technology; w _kBe the Gaussian sequence of zero-mean, covariance matrix is Q _k, the heterogeneous body perturbation in the expression seawater.

When in step 1, obtaining new detection y _K+1Afterwards, utilize it right

And P _K1+|kUpgrade, get k+1 state estimation constantly With estimation error variance P _K+1As follows

${\hat{x}}_{k+1}={\hat{x}}_{l+1k}+K_k(y_{k+1}-\mathrm{\Φ}{\hat{x}}_{k+1k})---\left(8\right)$

P _k+1＝(I-K _kΦ)P _k+1|k (9)

In the formula, I is unit matrix, K _kBe gain matrix, computing formula is as follows

K _k＝P _k+1|kΦ ^T(ΦP _k+1kΦ ^T+R _k) ^-1 (10)

4. virtual detection is proofreaied and correct

Set up virtual detection model

$m_k=H{\hat{x}}_k-\mathrm{\ϵ}---\left(11\right)$

In the formula, ε is arbitrarily small random normal number, and variance is R _εH=[sign (x _k(1)),, sign (x _k(N))], sign (x _k(i)) expression x _kThe sign function of i element.

Below utilize virtual detection model, to the optimal State Estimation that is obtained by formula (8) and (9) in the previous step

With estimation error variance P _K+1Proofread and correct.Trimming process realizes in the mode of iteration.

Definition

, Ω ₀=Pk+1 gets j=1, and 2,, J-1 (J is iterations, and the visual requirement to picture quality of J is set), it is as follows to carry out iteration

${\stackrel{\‾}{H}}_j=[\mathrm{sign}\left({\hat{z}}_j\left(1\right)\right),\·\·\·,\mathrm{sign}\left({\hat{z}}_j\left(n\right)\right)]---\left(11\right)$

$G_j={\mathrm{\Ω}}_j{\stackrel{\‾}{H}}_j^T{({\stackrel{\‾}{H}}_j{\mathrm{\Ω}}_j{\stackrel{\‾}{H}}_j^T+R_{\mathrm{\ϵ}})}^{-1}---\left(12\right)$

${\hat{z}}_{j+1}=(I-G_j{\stackrel{\‾}{H}}_j){\hat{z}}_j---\left(13\right)$

${\mathrm{\Ω}}_{j+1}=(I-G_j{\stackrel{\‾}{H}}_j){\mathrm{\Ω}}_j---\left(14\right)$

After iteration finishes, with state and variance correction as a result assignment give

And P _K+1, namely

${\hat{x}}_{k+1}={\hat{z}}_{j+1}---\left(15\right)$

P _k+1＝Ω _j+1 (16)

5. optimum fusion

Obtain at every sub-estimator

With

After, they are merged in main estimator, obtain the state that the overall situation is estimated

With the estimation error variance battle array As follows

${\hat{x}}_{k+1}^g=P_{k+1}^g\underset{i=1}{\overset{p}{\mathrm{\Σ}}}{\left(P_{k+1}^i\right)}^{-1}{\hat{x}}_{k+1}^i---\left(17\right)$

$P_{k+1}^g={\left(\underset{i=1}{\overset{p}{\mathrm{\Σ}}}{\left(P_{k+1}^i\right)}^{-1}\right)}^{-1}---\left(18\right)$

Globalstate estimation in the formula (17)

For in the target image at the discrete camber line constantly of k+1, all are discrete constantly 0,1 when obtaining,, k,, behind the globalstate estimation of N, can recover target image.Wherein, numerical value of N has determined the size of detecting target image.

Claims (5)

1. how unmanned submarine navigation device collaborative detection method based on compressed sensing is characterized in that comprising based on the space compression sampling of compressed sensing with based on distributed image reconstruct two parts of associating optimal estimation;

Described space compression sampling based on compressed sensing comprises:

With many UUV one of them as main UUV, and other UUV are as from UUV; Each array element to the sonar array of all UUV adopts scale-of-two pseudorandomcode, and these coding vectors consist of surveys matrix Φ; The coding pseudo-noise code generator produces, and random code generator is realized by shift register; The register maintenance is synchronous with system time, and its digital units carries out mould 2 and calculates within each clock period, and bit-by-bit generates pseudo-random code; The transmitting-receiving mode of data is each array element while transceiving data of sonar array;

Described distributed image reconstruct based on the associating optimal estimation comprises:

Each is from the corresponding sub-estimator of UUV, the corresponding main estimator of main UUV; Main estimator and every simultaneously isolated operation of sub-estimator receive signal to sonar and carry out the optimal estimation computing, obtain detection data separately, carry out afterwards fusion treatment in main estimator, realize sonar image reconstruct; Wherein, compressed sensing signal reconstruction constraint condition in conjunction with the minimum variance criterion, in the form embedding state estimation problem with virtual detection, is proofreaied and correct the state through optimal estimation; Virtual detection trimming process realizes in the mode of iteration.

2. the how unmanned submarine navigation device collaborative detection method based on compressed sensing according to claim 1, each array element while transceiving data that it is characterized in that described sonar array refers to: each array element of sonar sends coded data simultaneously, data length is N, and receives the detection data that returns simultaneously; The array number of sonar array is M, then once receives and dispatches M detection finishing a sector region; M coding vector consists of the detection matrix Φ ∈ R of sonar by row ^{M * N}, M＜N, then detection data Be expressed as

y _k＝Φx _k+n _k

Wherein, x _k∈ R ^NFor treating reconstruct data; n _k∈ R ^MSurvey noise for sonar, be approximately the white Gaussian noise of zero-mean, covariance matrix is R _k

3. the how unmanned submarine navigation device collaborative detection method based on compressed sensing according to claim 2 is characterized in that based on the information distributing method in the distributed image restructuring procedure of associating optimal estimation being:

Being engraved in the globalstate estimation and the estimation error variance that obtain in the main estimator during k is

With

Before carrying out next time recursion estimation, at first these information are carried out following distribution in sub-estimator

${\hat{x}}_k^i={\hat{x}}_k^g$

$P_k^i={\mathrm{\β}}_i^{-1}{P}_k^g$

Wherein,

With

Represent respectively state estimation value and estimation error variance battle array in i the sub-estimator, i=1,2,, p, p are the nodes of many UUV; β _iFor the information distribution factor, satisfy β _i〉=0 and

Employing is as follows based on the multidate information distribution method of estimation error variance

${\mathrm{\β}}_i=\frac{{\mathrm{trP}}_i}{{\mathrm{trP}}_1+{\mathrm{trP}}_2+\·+{\mathrm{trP}}_p+{\mathrm{trP}}_m}=\frac{{\mathrm{trP}}_i}{\underset{i=1}{\overset{p,m}{\mathrm{\Σ}}}{\mathrm{trP}}_i}$

In the formula: P _iBe i sub-estimator to the estimation error variance of state, m represents main estimator, the mark of tr representing matrix.

4. the how unmanned submarine navigation device collaborative detection method based on compressed sensing according to claim 3 is characterized in that described main estimator and every simultaneously isolated operation of sub-estimator receive signal to sonar and carry out the optimal estimation computing and comprise:

Time upgrades and surveys and upgrade

At k constantly, by the known state valuation

With estimation error variance battle array P _k, prediction k+1 state constantly

With estimation error variance battle array P _K+1|kAs follows

${\hat{x}}_{k+1|k}=A{\hat{x}}_k$

$P_{k+1|k}={\mathrm{AP}}_k{A}^T+Q_k$

Formula In, the prediction of state has been utilized signal model

x _k+1＝Ax _k+w _k

In the formula: A ∈ R ^{N * N}Be state-transition matrix, its value is determined by beam synthesizing technology; w _kBe the Gaussian sequence of zero-mean, covariance matrix is Q _k, the heterogeneous body perturbation in the expression seawater;

When obtaining new detection y _K+1Afterwards, right

And P _K+1|kUpgrade, get k+1 state estimation constantly

With estimation error variance P _K+1As follows

${\hat{x}}_{k+1}={\hat{x}}_{k+1k}+K_k(y_{k+1}-\mathrm{\Φ}{\hat{x}}_{k+1k})$

P _k+1＝(I-K _kΦ)P _k+1|k

In the formula, I is unit matrix, K _kBe gain matrix, computing formula is as follows

K _k＝P _k+1|kΦ ^T(ΦP _k+1|kΦ ^T+R _k) ^-1；

Virtual detection is proofreaied and correct

Virtual detection model

$m_k=H{\hat{x}}_k-\mathrm{\ϵ}$

In the formula, ε is arbitrarily small random normal number, and variance is R _εH=[sign (x _k(1)),, sign (x _k(N))], sign (x _k(i)) expression x _kThe sign function of i element;

Utilize virtual detection model, to optimal State Estimation With estimation error variance P _K+1Proofread and correct; Trimming process realizes in the mode of iteration;

Definition

Ω ₀=P _K+1, get j=1,2,, J-1, J are iterations, it is as follows to carry out iteration

${\stackrel{\‾}{H}}_j=[\mathrm{sign}\left({\hat{z}}_j\left(1\right)\right),\·,\mathrm{sign}\left({\hat{z}}_j\left(n\right)\right)]$

$G_j={\mathrm{\Ω}}_j{\stackrel{\‾}{H}}_j^T{({\stackrel{\‾}{H}}_j{\mathrm{\Ω}}_j{\stackrel{\‾}{H}}_j^T+R_{\mathrm{\ϵ}})}^{-1}$

${\hat{z}}_{j+1}=(I-G_j{\stackrel{\‾}{H}}_j){\hat{z}}_j$

${\mathrm{\Ω}}_{j+1}=(I-G_j{\stackrel{\‾}{H}}_j){\mathrm{\Ω}}_j$

After iteration finishes, with state and variance correction as a result assignment give

And P _K+1, namely

${\hat{x}}_{k+1}={\hat{z}}_{j+1}$

P _k+1＝Ω _j+1。

5. the how unmanned submarine navigation device collaborative detection method based on compressed sensing according to claim 4 is characterized in that the method for described fusion treatment is:

Obtain at every sub-estimator

With

After, they are merged in main estimator, obtain the state that the overall situation is estimated With the estimation error variance battle array

As follows

${\hat{x}}_{k+1}^g=P_{k+1}^g\underset{i=1}{\overset{p}{\mathrm{\Σ}}}{\left(P_{k+1}^i\right)}^{-1}{\hat{x}}_{k+1}^i$

$P_{k+1}^g={\left(\underset{i=1}{\overset{p}{\mathrm{\Σ}}}{\left(P_{k+1}^i\right)}^{-1}\right)}^{-1}$

Globalstate estimation

For in the target image at the discrete camber line constantly of k+1, all are discrete constantly 0,1 when obtaining,, k,, behind the globalstate estimation of N, can recover target image; Wherein, numerical value of N has determined the size of detecting target image.

Images