A method and system for continuous estimation of sedimentary layer parameters based on Kalman filtering
By processing and model correction of the seabed backscattered echo signal based on Kalman filtering, the problem of large calculation overhead and slow convergence speed in the existing technology is solved, and efficient sedimentary layer parameter estimation is achieved and seabed substratum classification.
Patent Information
- Application Number
- CN202010869759.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-08-26
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2040-08-26
AI Technical Summary
The existing single-beam bottom classification method has high calculation overhead and slow convergence speed, which is easy to fall into local optimality, difficult to tune hyperparameters, and difficult to achieve efficient estimation of subsea sedimentary layer parameters.
Using Kalman filtering method, the seabed backscattered echo signal is preprocessed and the time domain high-frequency model is established. The parameters are dynamically corrected by the traceless Kalman filter to realize the adaptive estimation of the deposition layer parameters.
It significantly improves the algorithm efficiency, shortens the estimation time, improves the parameter prediction accuracy and result credibility, and ensures the accuracy and calculation efficiency of seabed substrate classification.
Smart Images

Figure CN114114221B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the field of hydroacoustic physics, and in particular to a method and system for continuously estimating sediment layer parameters based on Kalman filtering. Background Art
[0002] In-depth understanding of the physical characteristics and spatial distribution of seabed sediments is of great significance to seabed resource exploration, marine environmental monitoring, seabed engineering construction and other activities. Compared with the traditional method of measuring sediment characteristics by fixed-point sampling of the seabed bottom, the acoustic remote sensing method uses the seabed acoustic echo signal to achieve high-density surveys of seabed sediments in a large area. Among them, single-beam bathymetric sonar is widely used in seabed sediment characteristics identification and bottom classification due to its advantages such as low price, easy operation, and wide-angle measurement. The commonly used single-beam bottom classification method is the matching inversion method. This method requires repeated calls to the forward model calculation during the optimization process. It generally has other problems such as high computational overhead, slow convergence, easy to fall into local optimality, and difficulty in hyperparameter tuning. Summary of the invention
[0003] The purpose of the present invention is to overcome the defects of the prior art and propose a method and system for continuous estimation of sedimentary layer parameters based on Kalman filtering.
[0004] The present invention proposes a method for continuous estimation of sedimentary layer parameters based on Kalman filtering, the method comprising:
[0005] The seabed backscatter echo signals collected by the single-beam sonar are preprocessed to obtain the envelopes of the seabed backscatter echo signals corresponding to different sampling moments, and the ensemble average envelopes of several groups of seabed backscatter echo signals are obtained through average processing;
[0006] A high-frequency seabed backscattering model in the time domain is established to obtain the mapping relationship between the envelope of the seabed backscattering echo signal and the sediment layer parameters;
[0007] An unscented Kalman filter for sediment layer parameters is established based on a high-frequency seafloor backscatter model in the time domain;
[0008] The ensemble average envelope of the seabed backscatter echo signal is input into the unscented Kalman filter. According to the error feedback between the predicted and measured data, the parameters of the time-domain high-frequency seabed backscatter model are dynamically corrected to obtain an adaptive estimation of the sediment layer parameters corresponding to the envelope of each group of seabed backscatter echo signals.
[0009] As an improvement of the above method, the seabed backscatter echo signal collected by the single-beam sonar is preprocessed to obtain the envelope of the seabed backscatter echo signal corresponding to different sampling moments, and the ensemble average envelope of several groups of seabed backscatter echo signals is obtained by averaging; specifically, the method includes:
[0010] The seabed backscatter echo signal collected by the single-beam sonar is filtered, and the two-dimensional matrix p[m,n] of the seabed backscatter echo signal envelope is obtained by Hilbert transform, where m is a different signal pulse, m=1,2,3,…,M, M is the number of signal pulses; n is the sampling point, and each signal pulse has N sampling points;
[0011] The minimum threshold method is used to align the envelope of the seabed backscatter echo signal. The alignment subscript of the mth row of the two-dimensional matrix p[m,n] is j m , 1≤j m ≤N, the average aligned sampling point integer j is expressed as:
[0012]
[0013] By the delay d of the mth row m =jj m , and obtain the aligned two-dimensional matrix p[m,(nd m )];
[0014] The aligned seabed backscatter echo signal envelopes are averaged to obtain the ensemble average envelope p of the seabed backscatter echo signal. al [n]:
[0015]
[0016] As an improvement of the above method, the method of establishing a time-domain high-frequency seabed backscattering model to obtain a mapping relationship between the envelope of the seabed backscattering echo signal and the sediment layer parameters specifically includes:
[0017] A high-frequency seabed backscattering model in the time domain is established to obtain the seabed backscattering echo envelope sound pressure p a (t), and then we get p a (t) and parameters w2,γ,σ v ,ρ,v,k ρ The mapping relationship between them; where w2 is the spectrum intensity, γ is the spectrum index, σ v is the seafloor volume scattering coefficient, ρ is the ratio of the saturated volume density of the sediment layer to the mass density of the seawater, v is the sound velocity ratio of the sediment layer to the seawater, k ρ is the attenuation constant;
[0018] According to the sedimentary layer parameter M z With parameters w2,γ,σ v ,ρ,v,k ρ The relationship between the backscattered echo signal envelope and the sediment layer parameter M is obtained. z The mapping relationship.
[0019] As an improvement of the above method, the unscented Kalman filter of sediment layer parameters is established according to the time domain high-frequency seabed backscattering model; specifically includes:
[0020] Establish the system state equation of sediment layer parameters:
[0021] X k =X k-1 +V k
[0022] Among them, X k is the sedimentary layer parameter at time k, X k-1 is the deposition layer parameter at time k-1, V k is the random disturbance in the process at time k, which obeys a Gaussian distribution with a mean of zero and a covariance matrix of Q = q I, where I is the identity matrix;
[0023] Establish the system observation equation of sedimentary layer parameters:
[0024] Y k =h k (X k-1 )+W k
[0025] Among them, Y k is the observed signal at time k, W k is the measurement noise, which follows a Gaussian distribution with a mean of zero and a variance of r. k (·) is the observation function:
[0026]
[0027] in, represents the sedimentary layer parameters to be estimated, n represents the discrete sampling points, is the ensemble average envelope of the seafloor backscatter echo signal at the nth sampling point, p k (n) is the seabed backscatter echo envelope actually measured at the nth sampling point;
[0028] Based on the state space model, an unscented Kalman filter is established to estimate the Kalman filter value at the k-1 moment of the system. Correction is performed to update the sedimentary layer parameters and obtain the estimated value of the Kalman filter at time k
[0029]
[0030] in, is the scaling amount of the prediction error, K k is the Kalman gain, E k is the least square mean error:
[0031]
[0032] As an improvement of the above method, the ensemble average envelope of the seabed backscatter echo signal is input into the unscented Kalman filter, and the parameters of the time-domain high-frequency seabed backscatter model are dynamically corrected according to the error feedback between the prediction and the measured data, so as to obtain an adaptive estimation of the sediment layer parameters corresponding to the envelope of each group of seabed backscatter echo signals; specifically, the method includes:
[0033] The sediment layer parameters are estimated by matching inversion method for the ensemble average envelope of the first group of seafloor backscatter echo signals, and used as the initial state of the unscented Kalman filter;
[0034] The initial state covariance of the unscented Kalman filter is set to the identity matrix I, and the sequence length is set to the number of signal scattering envelopes;
[0035] The ensemble average envelopes of the seabed backscatter echo signals of the other groups except the first group are sequentially input into the unscented Kalman filter model;
[0036] For each group of seabed backscatter echo signal, the average envelope of the ensemble is used to calculate the average particle size M of the seabed sediment using the simulated annealing algorithm. z Search, and according to M z With parameters w2,γ,σ v ,ρ,v,k ρ The functional relationship between M z Each search value of gets a set of parameters w2,γ,σ v ,ρ,v,k ρ The values of M are substituted into the time domain high frequency seabed backscattering model to calculate M z The least square mean square error corresponding to the search value; the least square mean square error is used as the optimized cost function, and the cost function is minimized through multiple iterations to obtain the average particle size M of the seabed sediment corresponding to the ensemble average envelope of the group of seabed backscatter echo signals z , i.e., the sediment layer parameter prediction result; repeat this step until the sediment layer parameter prediction result corresponding to the ensemble average envelope of each group of seabed backscatter echo signals is obtained.
[0037] A continuous estimation system for sediment layer parameters based on Kalman filtering, the system comprises: a preprocessing module, a backscattering model establishment module, an unscented Kalman filter establishment module and a sediment layer parameter estimation module; wherein,
[0038] The preprocessing module is used to preprocess the seabed backscatter echo signal collected by the single-beam sonar to obtain the envelope of the seabed backscatter echo signal corresponding to different sampling moments, and obtain the ensemble average envelope of several groups of seabed backscatter echo signals through average processing;
[0039] The backscattering model building module is used to build a time-domain high-frequency seabed backscattering model to obtain a mapping relationship between the envelope of the seabed backscattering echo signal and the sediment layer parameters;
[0040] The unscented Kalman filter establishment module is used to establish an unscented Kalman filter of sediment layer parameters according to a time-domain high-frequency seabed backscattering model;
[0041] The sediment layer parameter estimation module is used to input the ensemble average envelope of the seabed backscatter echo signal into the unscented Kalman filter, dynamically correct the parameters of the time domain high-frequency seabed backscatter model according to the error feedback between the predicted and measured data, and obtain the adaptive estimation of the sediment layer parameters corresponding to the envelope of each group of seabed backscatter echo signals.
[0042] As an improvement of the above system, the specific implementation process of the preprocessing module is as follows:
[0043] The seabed backscatter echo signal collected by the single-beam sonar is filtered, and the two-dimensional matrix p[m,n] of the seabed backscatter echo signal envelope is obtained by Hilbert transform, where m is a different signal pulse, m=1,2,3,…,M, M is the number of signal pulses; n is the sampling point, and each signal pulse has N sampling points;
[0044] The minimum threshold method is used to align the envelope of the seabed backscatter echo signal. The alignment subscript of the mth row of the two-dimensional matrix p[m,n] is j m , 1≤j m ≤N, the average aligned sampling point integer j is expressed as:
[0045]
[0046] By the delay d of the mth row m =jj m , and obtain the aligned two-dimensional matrix p[m,(nd m )];
[0047] The aligned seabed backscatter echo signal envelopes are averaged to obtain the ensemble average envelope p of the seabed backscatter echo signal. al [n]:
[0048]
[0049] As an improvement of the above system, the specific implementation process of the backscatter model building module is as follows:
[0050] A high-frequency seabed backscattering model in the time domain is established to obtain the seabed backscattering echo envelope sound pressure p a (t), and then we get p a (t) and parameters w2,γ,σ v ,ρ,v,k ρ The mapping relationship between them; where w2 is the spectrum intensity, γ is the spectrum index, σ v is the seafloor volume scattering coefficient, ρ is the ratio of the saturated volume density of the sediment layer to the mass density of the seawater, v is the sound velocity ratio of the sediment layer to the seawater, k ρ is the attenuation constant;
[0051] According to the sedimentary layer parameter M z With parameters w2,γ,σ v ,ρ,v,k ρ The relationship between the backscattered echo signal envelope and the sediment layer parameter M is obtained. z The mapping relationship.
[0052] As an improvement of the above system, the specific implementation process of the unscented Kalman filter establishment module is as follows:
[0053] Establish the system state equation of sediment layer parameters:
[0054] X k =X k-1 +V k
[0055] Among them, X k is the sedimentary layer parameter at time k, X k-1 is the deposition layer parameter at time k-1, V k is the random disturbance in the process at time k, which obeys a Gaussian distribution with a mean of zero and a covariance matrix of Q = q I, where I is the identity matrix;
[0056] Establish the system observation equation of sedimentary layer parameters:
[0057] Y k =h k (X k-1 )+W k
[0058] Among them, Y k is the observed signal at time k, W k is the measurement noise, which follows a Gaussian distribution with a mean of zero and a variance of r. k (·) is the observation function:
[0059]
[0060] in, represents the sedimentary layer parameters to be estimated, n represents the discrete sampling points, is the ensemble average envelope of the seafloor backscatter echo signal at the nth sampling point, p k (n) is the seabed backscatter echo envelope actually measured at the nth sampling point;
[0061] Based on the state space model, an unscented Kalman filter is established to estimate the Kalman filter value at the k-1 moment of the system. Correction is performed to update the sedimentary layer parameters and obtain the estimated value of the Kalman filter at time k
[0062]
[0063] in, is the scaling amount of the prediction error, K k is the Kalman gain, E k is the least square mean error:
[0064]
[0065] As an improvement of the above system, the specific implementation process of the deposition layer parameter estimation module is as follows:
[0066] The sediment layer parameters are estimated by matching inversion method for the ensemble average envelope of the first group of seafloor backscatter echo signals, and used as the initial state of the unscented Kalman filter;
[0067] The initial state covariance of the unscented Kalman filter is set to the identity matrix I, and the sequence length is set to the number of signal scattering envelopes;
[0068] The ensemble average envelopes of the seabed backscatter echo signals of the other groups except the first group are sequentially input into the unscented Kalman filter model;
[0069] For each group of seabed backscatter echo signal, the average envelope of the ensemble is used to calculate the average particle size M of the seabed sediment using the simulated annealing algorithm. z Search, and according to M z With parameters w2,γ,σ v ,ρ,v,k ρ The functional relationship between M z Each search value of gets a set of parameters w2,γ,σ v ,ρ,v,k ρ The values of M are substituted into the time domain high frequency seabed backscattering model to calculate M zThe least square mean square error corresponding to the search value; the least square mean square error is used as the optimized cost function, and the cost function is minimized through multiple iterations to obtain the average particle size M of the seabed sediment corresponding to the ensemble average envelope of the group of seabed backscatter echo signals z , i.e., the sediment layer parameter prediction result; repeat this step until the sediment layer parameter prediction result corresponding to the ensemble average envelope of each group of seabed backscatter echo signals is obtained.
[0070] Compared with the prior art, the advantages of the present invention are:
[0071] 1. The method of the present invention greatly improves the efficiency of the algorithm, and the average time consumed by the estimation is significantly shortened;
[0072] 2. The prediction accuracy of the sediment layer parameters obtained by the method of the present invention is high, and the prediction value and the seabed sediment classification result are highly credible;
[0073] 3. The present invention can adaptively adjust the excitation of the parameter to be estimated according to the error feedback between the predicted and measured data, the least square mean square error is small, and the prediction result has good credibility;
[0074] 4. The present invention is used to determine the average particle size M of the seabed sediment. z The predictions are in good agreement with the results of seabed classification and experimental forecasts;
[0075] 5. The present invention significantly improves the efficiency of the algorithm while ensuring similar accuracy to the traditional method, and the overall calculation time is shortened by about 4 / 5. BRIEF DESCRIPTION OF THE DRAWINGS
[0076] Figure 1 It is a flow chart of a method for continuous estimation of sedimentary layer parameters based on Kalman filtering of the present invention;
[0077] Figure 2 is a schematic diagram of the original backscatter envelope;
[0078] Figure 3 is a schematic diagram of the backscatter envelope after alignment;
[0079] Figure 4(a) is a stem plot of the least square mean square error between the predicted envelope and the measured envelope obtained using the traditional matched inversion method;
[0080] FIG4( b ) is a stem diagram of the least square mean square error between the predicted envelope and the measured envelope obtained by the continuous estimation method of sedimentary layer parameters based on Kalman filtering of the present invention;
[0081] Figure 5(a) is the measured seabed backscatter echo envelope;
[0082] FIG5( b ) is a diagram of the seabed backscattering envelope obtained by forward modeling of the seabed acoustic parameters obtained by the Kalman filter-based sediment layer parameter continuous estimation method of the present invention;
[0083] Figure 6 It is a comparison diagram of the average particle size of the seabed sediment predicted by the traditional method and predicted by the continuous estimation method of sediment layer parameters based on Kalman filtering of the present invention;
[0084] Figure 7 It is a distribution map of seabed sediment types. DETAILED DESCRIPTION
[0085] The present invention proposes a method for continuous estimation of seafloor sediment parameters based on Kalman filtering. The seafloor backscatter echo is collected by single-beam sonar, and the acoustic inversion problem of shallow sea geoacoustic parameters is modeled as a continuous estimation problem through an unscented Kalman filter. A time-domain high-frequency seafloor backscatter model is used to predict the seafloor scattered echo envelope, and the model parameters are dynamically corrected according to error feedback to achieve adaptive estimation of sediment parameters.
[0086] The technical solution of the present invention is described in detail below with reference to the accompanying drawings and embodiments.
[0087] Example 1
[0088] like Figure 1 As shown, Embodiment 1 of the present invention proposes a method for continuous estimation of seabed sediment layer parameters based on Kalman filtering. The method flow can be divided into three main steps: data preprocessing, Kalman filter construction, and continuous estimation of sediment layer parameters.
[0089] (1) Data preprocessing
[0090] The seabed backscatter echo signal collected by the single-beam sonar is filtered, and the seabed backscatter echo signal envelope is obtained as a two-dimensional matrix p[m,n] through Hilbert transformation, where m is a different signal pulse, m=1,2,3,…,M, M is the number of signal pulses; n is the sampling point, and each signal pulse has N sampling points;
[0091] The minimum threshold method is used to align the envelope of the seabed backscatter echo signal. The alignment subscript of the mth row of the two-dimensional matrix p[m,n] is j m , 1≤j m ≤N; the average aligned sampling point integer is expressed as:
[0092]
[0093] Then the delay d of the mth row m =jj m ; Thus, the aligned two-dimensional matrix p[m,(ndm )];
[0094] The aligned seabed backscatter echo signal envelopes are averaged, and the ensemble average p of the seabed backscatter echo signal envelope is a [n] is represented by:
[0095]
[0096] p a [n] is the time domain envelope.
[0097] (2) High-frequency seafloor backscatter modeling
[0098] A high-frequency seabed backscattering model is established. Assuming that the roughness of the seabed interface is isotropic and satisfies Gaussian distribution, its power spectrum can be written as where k b is the wave number, w2 is the roughness, and γ=3.5 is the spectral index of the model in the present invention. The backscatter echo envelope intensity is divided into two parts: seabed surface scattering and seabed volume scattering. The intensity received by the single-beam sonar over time t is expressed as
[0099] I(t)=I i (t)+I v (t)
[0100] In the formula, I i is the scattering envelope of the seafloor rough interface, I v is the volume scattering envelope of the seabed medium, assuming that the two are independent of each other. i is a Kirchhoff approximate solution to the Helmholtz diffraction integral equation. For any incident angle θ i , the rough interface scattering signal envelope can be written as:
[0101]
[0102]
[0103]
[0104] Among them, τ p and a are the length and amplitude of the signal pulse, θ1(t) and θ2(t) are the incident angles corresponding to the leading and trailing edges of the sound source pulse, Ψ is the azimuth, I x is the envelope of the transmitted signal, R is the distance from the sound source to the scattering element, b(θ i ,Ψ) is the directivity coefficient of the transducer, α w is the frequency-dependent absorption coefficient of seawater, s i (θ i ) is the seafloor backscatter coefficient that varies with the angle of incidence:
[0105]
[0106] Where ρ is the ratio of the saturated volume density of the sediment layer to the mass density of seawater, v is the sound velocity ratio of the sediment layer to seawater, α is the spectral index of the model, α = γ / 2-1, γ is the spectral index γ = 3.5, k a is the ratio of 2π to the emission wavelength of the sound source, J0(·) is the zero-order Bessel function, and q is the function expression:
[0107]
[0108] C ξ For a function expression:
[0109]
[0110] Where w2 is the spectral intensity, Γ(·) is the gamma function;
[0111] Volumetric scattering envelope of seafloor medium I v (t) is:
[0112]
[0113]
[0114]
[0115]
[0116] Among them, σ v is the seafloor volume scattering coefficient, β e is the acoustic scattering loss in the sediment medium, in neper / m, β e =α b / 4.343, and the attenuation coefficient α b Satisfy α b =k ρ f, unit dB / m, k ρ is the attenuation constant, in dB / m / kHz, f is the signal frequency, in kHz; V l Bidirectional propagation loss on a rough surface with Gaussian distribution:
[0117]
[0118] V f (θ i )=[1-R 2 (θ i )] 2 ·cos 2 (θ i )·[1-v2 sin 2 (θ i )] -1 / 2
[0119]
[0120]
[0121] Among them, V f is the two-way propagation loss at the water layer-sediment layer interface, assuming that the two-way propagation loss at the small-scale rough interface is consistent. is the minimum inclination of the interface, k c is the cutoff wave number; finally, the acoustic pressure of the seabed backscatter echo envelope is expressed as:
[0122]
[0123] Where ρ w and v w are the density of the water layer and the speed of sound, respectively.
[0124] From this we get p a (t) and parameters w2,γ,σ v ,ρ,v,k ρ The mapping relationship between them;
[0125] According to the sedimentary layer parameter M z With parameters w2,γ,σ v ,ρ,v,k ρ The relationship between:
[0126]
[0127]
[0128] γ=3.25,M z ∈[-1,9.0]
[0129]
[0130]
[0131] σ v =0.004α b ,M z ∈[-1,9.0];
[0132] Thus, the envelope of the seabed backscatter echo signal and the sediment layer parameter M are obtained. z The mapping relationship.
[0133] (3) Kalman filter construction
[0134] ① Establish a state space model of sedimentary layer parameters, assuming that the state change of sedimentary layer parameters within a short distance is regarded as a slow-changing process, and the sedimentary layer parameters X k is the system state, the state noise is Gaussian white noise W(k), the observation noise is Gaussian white noise V(k), its variance depends on k, and the state and noise are unrelated. The state equation and observation equation of the system are expressed as: k =X k-1 +V k
[0135] Y k =h k (X k-1 )+W k
[0136] Where V k represents the random disturbance in the process at time k, which obeys a Gaussian distribution with a mean of zero and a covariance matrix of Q = q I, where I is the unit matrix. For the kth seafloor backscatter echo signal envelope, h k The function expression is:
[0137]
[0138] in represents the sedimentary layer parameters to be estimated, n represents the discrete sampling points, is the ensemble average envelope of the seafloor backscatter echo signal at the nth sampling point, p k To obtain the scattered echo envelope for actual measurement, W k represents the measurement noise, which follows a Gaussian distribution with a mean of zero and a variance of r. According to the process and measurement characteristics, q is set to 0.01, and r is set to 1 and 2. Let R k and Q k is the state noise V k and observation noise W k The covariance matrix of .
[0139] ② Based on this state space model, an unscented Kalman filter is established to predict the system state quantity. In the first estimation of the system, the predicted value of the system state quantity is corrected:
[0140]
[0141]
[0142] Where E k The least square mean square error is defined as follows:
[0143]
[0144] The above formula is used in the predicted value The estimated value of the current state is obtained by adding the scaling amount of the prediction error to the basis, and the Kalman gain K k Determine the size of the error adjustment, and the two together determine the amount of update of the state parameters in the system.
[0145] (4) Continuous estimation of sedimentary layer parameters
[0146] The ensemble average envelopes of all signals obtained in (1) are used as the input of the unscented Kalman filter model constructed in (3). The sedimentary layer parameters of the first group of ensemble average envelopes are estimated by the matching inversion method: the average particle size M of the seafloor sediment is calculated by the simulated annealing algorithm. z Search, the input parameters w2,γ,σ of the model in (1) v ,ρ,v,k ρ According to the empirical formula, the least square mean square error defined in (3) is set as the optimized cost function, and the cost function is minimized through multiple iterations. The inversion result is used as the initial state of the unscented Kalman filter, the initial state covariance of the unscented Kalman filter is set to the unit matrix I, and the sequence length is set to the number of signal scattering envelopes. The unscented Kalman filter estimates and outputs the sedimentary layer parameter prediction results corresponding to the average envelope of each group of signal ensembles in turn.
[0147] Test description:
[0148] This experiment was conducted in the Yellow Sea in August 2018. The main purpose of this experiment was to conduct a feasibility study on the remote sensing of sediment acoustic characteristics by measuring the seabed backscatter echoes by cruise mode. The cruise test started at the E1 station in the test sea area (36°00.1804′N121°36.8040′E). The transceiver was used to transmit a 25kHz single-frequency signal and a linear frequency modulation signal, and arrived at the E2 station the next day. The whole journey was about 130 kilometers. The transducer used in this experiment had a high sound source level and receiving sensitivity, with a 25kHz sound source level of 212.72dB and a receiving sensitivity of -186.1dB. The transceiver had good directivity.
[0149] The envelope of the seabed backscatter echo signal can be obtained by filtering the received seabed backscatter echo signal and performing Hilbert transform. A 25kHz time pulse with a length of 2ms was emitted. The signal sampling frequency was 500kHz, and the power amplifier signal and the transducer signal were recorded at the same time. The signal envelope was obtained by performing Hilbert transform on the original signal and taking the absolute value to obtain the envelope of the scattered echo. Due to the depth change of the seabed and the up and down floating of the transducer, the arrival time of the received envelope varies greatly, such as Figure 2 As shown. Since the transceiver has good directivity, the received signal has a high signal-to-noise ratio, and the signal envelope is aligned, as shown Figure 3As shown in the figure, the signal envelopes obtained during the navigation process are quite different due to the different scattering of the seabed, so 100 samples of the obtained scattering echo envelopes are averaged to obtain 80 groups of signal envelopes.
[0150] The 80 sets of envelopes of the preprocessed signals were matched and inverted, and the seabed acoustic parameters were continuously estimated. First, the traditional matching inversion method was used to invert the sediment layer parameters of the 80 sets of scattering envelopes. The algorithm used was the simulated annealing (SA) method, the search interval was [0, 10], the initial temperature T0 = 100, the temperature change rate ΔT = 0.98, and multiple termination conditions were set. When the average change rate of the objective function within 10 iterations δ < 10 -3 Or the iteration is terminated when the cost function value is less than 0.02, and the maximum number of iterations N max =50. The continuous estimation method is used to estimate the sedimentary layer parameters in sequence for 80 sets of scattering envelopes. The algorithm used is the unscented Kalman filter (UKF). In order to obtain relatively accurate initial state values, only the first set of envelope inversion results is used as the initial state of the unscented Kalman filter, and the initial state covariance of the unscented Kalman filter is set to the unit matrix I. The sequence length is set to 80, corresponding to 80 sets of scattering envelopes of the signal. Since the state change of the sedimentary layer parameters within a short distance can be regarded as a slow-changing process, let the state equation in formula (10) be X k =X k-1 +V k . Where V k represents the random disturbance in the process at time k, which obeys a Gaussian distribution with a mean of zero and a covariance matrix of Q = q I, where I is the unit matrix. The measurement equation is set to Y k =h k (X k-1 )+W k ,h k is the function given by formula (5), W k represents the measurement noise, which follows a Gaussian distribution with a mean of zero and a variance of r. According to the process and measurement characteristics, q is set to 0.01, and r is set to 1 and 2.
[0151] Figure 4(a) shows the stem plot of the least square mean square error between the predicted envelope and the measured envelope obtained by the matching inversion method as the distance changes, and Figure 4(b) shows the stem plot of the least square mean square error between the predicted envelope and the measured envelope obtained by the unscented Kalman filter as the distance changes. Since the matching inversion method terminates the search when the error converges to a certain threshold, its prediction accuracy is relatively stable. The unscented Kalman filter adaptively adjusts the excitation of the estimated parameter based on the error feedback between the predicted and measured data. It can be seen from the figure that the unscented Kalman filter can timely adjust the update amount of the state parameters near the inflection point to reduce the prediction error, but the measurement error increases in the area of dramatic bottom changes. This is because the state transfer equation assumes that the dynamic model is a slowly changing process. The least square mean square errors of the matching inversion method and the unscented Kalman filter are both controlled below 0.05, indicating that the prediction results are credible.
[0152] Figure 5(a) is the measured seabed backscattering envelope, and Figure 5(b) is the forward prediction of the seabed backscattering envelope using the method of the present invention using the seabed acoustic parameters obtained by continuous estimation. By comparison, it can be seen that the predicted seabed backscattering envelope corresponds well to the measured result.
[0153] Figure 6 The matching inversion method and the continuous estimation method are given to estimate the average particle size M of the seabed sediment. z The prediction and classification results of the seabed sediment. It can be seen from the figure that the classification results of the sediment by the continuous estimation method and the matching inversion method are basically consistent. The average particle size of the seabed sediment increases from about 60 kilometers, indicating that the seabed sediment changes from soft to hard. According to the Wentworth classification standard, the average particle size of the seabed sediment is in the range of very fine silt to medium silt.
[0154] Figure 7 The distribution map of the seabed sediment types in the test sea area is given. The E1-E2 survey line is the test navigation survey line. The sediment type at the beginning of the test is clayey silt, which is consistent with the average particle size obtained by inversion. Figure 6 As shown in the left box in the middle; and quickly transition to sandy silt, the latter half of the line is silty sand, Figure 6 The continuous estimation results are consistent with the experimental prediction results.
[0155] When using the unscented Kalman filter for one-step estimation, since the unscented Kalman filter does not need to calculate the Jacobian matrix of the nonlinear function, its computational complexity only comes from the forward function calculation called when generating the Sigma point set in the UT transformation and when predicting and correcting. Therefore, the time cost required by the continuous estimation method is determined by the number of calculations of the forward program, just like the matched inversion method. For the continuous estimation method, each set of envelopes is fixed to perform 2L+1 forward model calculations, where L is the number of parameters to be estimated; for the matched inversion method, the number of forward modeling operations required for each set of envelopes is determined by the number of iterations N of the optimization process, and varies according to the predetermined termination conditions. The number of iterations required for each set of inversions varies depending on the specific convergence situation, the number of pseudo-peaks and saddle points, and can range from dozens to hundreds of times. The calculations were performed using MATLAB in the PC environment of the Intel Core i5-9400F 2.90Ghz processor, and the comparison of the calculation time used by the two is shown in Table 1. The average calculation time of the continuous estimation method in a single estimation is shortened by 83.4% compared with the matched inversion method, and the overall time overhead is shortened by 79.6%.
[0156] Table 1 Comparison of computational time costs between SA and UKF
[0157] SA UKF Single estimated forward calculation times 2L+1 N Average calculation time for a single estimate (s) 74.1 12.3 Total operation time (s) 4,848.0 989.0
[0158] Example 2
[0159] According to the method of embodiment 1, embodiment 2 of the present invention proposes a continuous estimation system of sedimentary layer parameters based on Kalman filtering, characterized in that the system comprises: a preprocessing module, a backscattering model establishment module, an unscented Kalman filter establishment module and a sedimentary layer parameter estimation module; wherein,
[0160] The preprocessing module is used to preprocess the seabed backscatter echo signal collected by the single-beam sonar to obtain the envelope of the seabed backscatter echo signal corresponding to different sampling moments, and obtain the ensemble average envelope of several groups of seabed backscatter echo signals through average processing;
[0161] The backscattering model building module is used to build a time-domain high-frequency seabed backscattering model to obtain a mapping relationship between the envelope of the seabed backscattering echo signal and the sediment layer parameters;
[0162] The unscented Kalman filter establishment module is used to establish an unscented Kalman filter of sediment layer parameters according to a time-domain high-frequency seabed backscattering model;
[0163] The sediment layer parameter estimation module is used to input the ensemble average envelope of the seabed backscatter echo signal into the unscented Kalman filter, dynamically correct the parameters of the time domain high-frequency seabed backscatter model according to the error feedback between the predicted and measured data, and obtain the adaptive estimation of the sediment layer parameters corresponding to the envelope of each group of seabed backscatter echo signals.
[0164] The specific implementation process of each of the above modules can be found in the specific steps of Example 1.
[0165] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention is described in detail with reference to the embodiments, it should be understood by those skilled in the art that any modification or equivalent replacement of the technical solutions of the present invention does not depart from the spirit and scope of the technical solutions of the present invention and should be included in the scope of the claims of the present invention.
Claims
1. A method for continuous estimation of sedimentary layer parameters based on Kalman filtering, the method comprising: The seabed backscatter echo signals collected by the single-beam sonar are preprocessed to obtain the envelopes of the seabed backscatter echo signals corresponding to different sampling moments, and the ensemble average envelopes of several groups of seabed backscatter echo signals are obtained through average processing; A high-frequency seabed backscattering model in the time domain is established to obtain the mapping relationship between the envelope of the seabed backscattering echo signal and the sediment layer parameters; An unscented Kalman filter for sediment layer parameters is established based on a high-frequency seafloor backscatter model in the time domain; The ensemble average envelope of the seabed backscatter echo signal is input into the unscented Kalman filter, and the parameters of the time-domain high-frequency seabed backscatter model are dynamically corrected according to the error feedback between the prediction and measured data, so as to obtain the adaptive estimation of the sediment layer parameters corresponding to the envelope of each group of seabed backscatter echo signals; The method pre-processes the seabed backscatter echo signals collected by the single-beam sonar to obtain the envelopes of the seabed backscatter echo signals corresponding to different sampling moments, and obtains the ensemble average envelopes of several groups of seabed backscatter echo signals through averaging; specifically includes: The seabed backscatter echo signal collected by the single-beam sonar is filtered, and the two-dimensional matrix p[m,n] of the seabed backscatter echo signal envelope is obtained by Hilbert transform, where m is a different signal pulse, m=1,2,3,…,M, M is the number of signal pulses; n is the sampling point, and each signal pulse has N sampling points; The minimum threshold method is used to align the envelope of the seabed backscatter echo signal. The alignment subscript of the mth row of the two-dimensional matrix p[m,n] is j m , 1≤j m ≤N, the average aligned sampling point integer j is expressed as: By the delay d of the mth row m =jj m , and obtain the aligned two-dimensional matrix p[m,(nd m )]; The aligned seabed backscatter echo signal envelopes are averaged to obtain the ensemble average envelope p of the seabed backscatter echo signal. al [n]: The method of establishing a time-domain high-frequency seabed backscattering model to obtain a mapping relationship between the envelope of the seabed backscattering echo signal and the sediment layer parameters specifically includes: A high-frequency seabed backscattering model in the time domain is established to obtain the seabed backscattering echo envelope sound pressure p a (t), and then we get p a (t) and parameters w2,γ,σ v ,ρ,v,k ρ The mapping relationship between them; where w2 is the spectrum intensity, γ is the spectrum index, σ v is the seafloor volume scattering coefficient, ρ is the ratio of the saturated volume density of the sediment layer to the mass density of the seawater, v is the sound velocity ratio of the sediment layer to the seawater, k ρ is the attenuation constant; According to the sedimentary layer parameter M z With parameters w2,γ,σ v ,ρ,v,k ρ The relationship between the backscattered echo signal envelope and the sediment layer parameter M is obtained. z The mapping relationship; The unscented Kalman filter for establishing sediment layer parameters according to the time-domain high-frequency seabed backscattering model specifically includes: Establish the system state equation of sediment layer parameters: X k =X k-1 +V k Among them, X k is the sedimentary layer parameter at time k, X k-1 is the deposition layer parameter at time k-1, V k is the random disturbance in the process at time k, which obeys a Gaussian distribution with a mean of zero and a covariance matrix of Q = qI, where I is the identity matrix; Establish the system observation equation of sedimentary layer parameters: Y k =h k (X k-1 )+W k Among them, Y k is the observed signal at time k, W k is the measurement noise, which follows a Gaussian distribution with a mean of zero and a variance of r. k (·) is the observation function: in, represents the sedimentary layer parameters to be estimated, n represents the discrete sampling points, is the ensemble average envelope of the seafloor backscatter echo signal at the nth sampling point, p k (n) is the seabed backscatter echo envelope actually measured at the nth sampling point; Based on the state space model, an unscented Kalman filter is established to estimate the Kalman filter value at the k-1 moment of the system. Correction is performed to update the sedimentary layer parameters and obtain the estimated value of the Kalman filter at time k in, is the scaling amount of the prediction error, K k is the Kalman gain, E k is the least square mean error: The method comprises: inputting the ensemble average envelope of the seabed backscatter echo signal into the unscented Kalman filter, dynamically correcting the parameters of the time-domain high-frequency seabed backscatter model according to the error feedback between the prediction and the measured data, and obtaining an adaptive estimation of the sediment layer parameters corresponding to the envelope of each group of seabed backscatter echo signals; specifically comprising: The sediment layer parameters are estimated by matching inversion method for the ensemble average envelope of the first group of seafloor backscatter echo signals, and used as the initial state of the unscented Kalman filter; The initial state covariance of the unscented Kalman filter is set to the identity matrix I, and the sequence length is set to the number of signal scattering envelopes; The ensemble average envelopes of the seabed backscatter echo signals of the other groups except the first group are sequentially input into the unscented Kalman filter model; For each group of seabed backscatter echo signal, the average envelope of the ensemble is used to calculate the average particle size M of the seabed sediment using the simulated annealing algorithm. z Search, and according to M z With parameters w2,γ,σ v ,ρ,v,k ρ The functional relationship between M z Each search value of gets a set of parameters w2,γ,σ v ,ρ,v,k ρ The values of M are substituted into the time domain high frequency seabed backscattering model to calculate M z The least square mean square error corresponding to the search value; the least square mean square error is used as the optimized cost function, and the cost function is minimized through multiple iterations to obtain the average particle size M of the seabed sediment corresponding to the ensemble average envelope of the group of seabed backscatter echo signals z , i.e., the sediment layer parameter prediction result; repeat this step until the sediment layer parameter prediction result corresponding to the ensemble average envelope of each group of seabed backscatter echo signals is obtained.
2. A continuous estimation system for sedimentary layer parameters based on Kalman filtering, characterized in that: The system comprises: a preprocessing module, a backscattering model building module, an unscented Kalman filter building module and a sediment layer parameter estimation module; wherein, The preprocessing module is used to preprocess the seabed backscatter echo signal collected by the single-beam sonar to obtain the envelope of the seabed backscatter echo signal corresponding to different sampling moments, and obtain the ensemble average envelope of several groups of seabed backscatter echo signals through average processing; The backscattering model building module is used to build a time-domain high-frequency seabed backscattering model to obtain a mapping relationship between the envelope of the seabed backscattering echo signal and the sediment layer parameters; The unscented Kalman filter establishment module is used to establish an unscented Kalman filter of sediment layer parameters according to a time-domain high-frequency seabed backscattering model; The sediment layer parameter estimation module is used to input the ensemble average envelope of the seabed backscatter echo signal into the unscented Kalman filter, dynamically correct the parameters of the time-domain high-frequency seabed backscatter model according to the error feedback between the prediction and the measured data, and obtain the adaptive estimation of the sediment layer parameters corresponding to the envelope of each group of seabed backscatter echo signals; The specific implementation process of the preprocessing module is: The seabed backscatter echo signal collected by the single-beam sonar is filtered, and the two-dimensional matrix p[m,n] of the seabed backscatter echo signal envelope is obtained by Hilbert transform, where m is a different signal pulse, m=1,2,3,…,M, M is the number of signal pulses; n is the sampling point, and each signal pulse has N sampling points; The minimum threshold method is used to align the envelope of the seabed backscatter echo signal. The alignment subscript of the mth row of the two-dimensional matrix p[m,n] is j m , 1≤j m ≤N, the average aligned sampling point integer j is expressed as: By the delay d of the mth row m =jj m , and obtain the aligned two-dimensional matrix p[m,(nd m )]; The aligned seabed backscatter echo signal envelopes are averaged to obtain the ensemble average envelope p of the seabed backscatter echo signal. al [n]: The specific implementation process of the backscatter model establishment module is as follows: A high-frequency seabed backscattering model in the time domain is established to obtain the seabed backscattering echo envelope sound pressure p a (t), and then we get p a (t) and parameters w2,γ,σ v ,ρ,v,k ρ The mapping relationship between them; where w2 is the spectrum intensity, γ is the spectrum index, σ v is the seafloor volume scattering coefficient, ρ is the ratio of the saturated volume density of the sediment layer to the mass density of the seawater, v is the sound velocity ratio of the sediment layer to the seawater, k ρ is the attenuation constant; According to the sedimentary layer parameter M z With parameters w2,γ,σ v ,ρ,v,k ρ The relationship between the backscattered echo signal envelope and the sediment layer parameter M is obtained. z The mapping relationship; The specific implementation process of the unscented Kalman filter establishment module is as follows: Establish the system state equation of sediment layer parameters: X k =X k-1 +V k Among them, X k is the sedimentary layer parameter at time k, X k-1 is the deposition layer parameter at time k-1, V k is the random disturbance in the process at time k, which obeys a Gaussian distribution with a mean of zero and a covariance matrix of Q = qI, where I is the identity matrix; Establish the system observation equation of sedimentary layer parameters: Y k =h k (X k-1 )+W k Among them, Y k is the observed signal at time k, W k is the measurement noise, which follows a Gaussian distribution with a mean of zero and a variance of r. k (·) is the observation function: in, represents the sedimentary layer parameters to be estimated, n represents the discrete sampling points, is the ensemble average envelope of the seafloor backscatter echo signal at the nth sampling point, p k (n) is the seabed backscatter echo envelope actually measured at the nth sampling point; Based on the state space model, an unscented Kalman filter is established to estimate the Kalman filter value at the k-1 moment of the system. Correction is performed to update the sedimentary layer parameters and obtain the estimated value of the Kalman filter at time k in, is the scaling amount of the prediction error, K k is the Kalman gain, E k is the least square mean error: The specific implementation process of the sedimentary layer parameter estimation module is as follows: The sediment layer parameters are estimated by matching inversion method for the ensemble average envelope of the first group of seafloor backscatter echo signals, and used as the initial state of the unscented Kalman filter; The initial state covariance of the unscented Kalman filter is set to the identity matrix I, and the sequence length is set to the number of signal scattering envelopes; The ensemble average envelopes of the seabed backscatter echo signals of the other groups except the first group are sequentially input into the unscented Kalman filter model; For each group of seabed backscatter echo signal, the average envelope of the ensemble is used to calculate the average particle size M of the seabed sediment using the simulated annealing algorithm. z Search, and according to M z With parameters w2,γ,σ v ,ρ,v,k ρ The functional relationship between M z Each search value of gets a set of parameters w2,γ,σ v ,ρ,v,k ρ The values of M are substituted into the time domain high frequency seabed backscattering model to calculate M z The least square mean square error corresponding to the search value; the least square mean square error is used as the optimized cost function, and the cost function is minimized through multiple iterations to obtain the average particle size M of the seabed sediment corresponding to the ensemble average envelope of the group of seabed backscatter echo signals z , i.e., the sediment layer parameter prediction result; repeat this step until the sediment layer parameter prediction result corresponding to the ensemble average envelope of each group of seabed backscatter echo signals is obtained.
Citation Information
Patent Citations
Method for estimating small-scene radar target range images on basis of compression Kalman filtering
CN103235295A
Real-time evaluation method for submarine topographic survey quality of multi-beam sonar based on amplitude method
CN109471114A
Method for simulating noisy radar target echoes
US8334803B1
