CN112906632A - Automatic extraction method of highly adaptive time delay target signal - Google Patents
Automatic extraction method of highly adaptive time delay target signal Download PDFInfo
- Publication number
- CN112906632A CN112906632A CN202110288519.4A CN202110288519A CN112906632A CN 112906632 A CN112906632 A CN 112906632A CN 202110288519 A CN202110288519 A CN 202110288519A CN 112906632 A CN112906632 A CN 112906632A
- Authority
- CN
- China
- Prior art keywords
- bat
- signal
- time
- iteration
- atom
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Signal Processing (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
- Feedback Control In General (AREA)
Abstract
The invention discloses a highly self-adaptive time delay target signal automatic extraction method, which designs a general atom, does not need to construct a huge redundant dictionary in advance when carrying out sparse decomposition, sets atom characteristic parameters as bat individual position vectors, constructs flexible atoms by utilizing an improved bat algorithm to carry out optimal position search, and further realizes standardized optimal matching atom search; using sparse decomposition dual-threshold iteration stopping criterion, wherein residual ℓ -2 norm threshold is determined by background noise intensity without target signal sequencen s ℓ -2 norm adaptive determination; the iteration time threshold is set by a user according to real-time requirements. The application of the double criteria can ensure the signal extraction precision, improve the iteration speed and effectively avoid over decomposition. The method does not depend on complex prior knowledge and subsequent manual judgment and screening, can effectively extract the target signal and simultaneously reserve other components of the signal to the maximum extent, and is completely automaticA method for extracting non-stationary time delay signals is provided.
Description
Technical Field
The invention belongs to the technical field of signal processing, and particularly relates to a method for automatically identifying and extracting a time delay target signal in background noise.
Background
Extraction of non-stationary time delay target signals is an important research content of modern signal processing, and has important application in the fields of vibration signal detection, earthquake monitoring, biomedical signal detection and the like.
The existing signal extraction method mainly comprises the following steps: wavelet decomposition, hilbert-yellow transform, Variational Modal Decomposition (VMD), and signal sparse decomposition algorithms, among others.
Wavelet decomposition is an important means of processing a time-delayed signal, but requires selection of appropriate wavelet basis functions and decomposition level numbers, which may otherwise result in an extracted signal that is not a target signal. The Hilbert-Huang transform has a good effect of extracting linear steady-state signals, but most natural signals are difficult to satisfy the premise, especially the extraction of time delay signals aliasing in noise, and the application effect of the Hilbert-Huang transform is not ideal. The variational modal decomposition has a good processing effect on non-stationary and non-linear signals under a certain condition, but local jump information is punished excessively in the processing process, and a boundary effect and an aliasing phenomenon of local features in a global range exist, so that the variational modal decomposition is not suitable for time delay local feature signal extraction. In addition, the VMD algorithm requires that the number of modes K is predefined, and it is necessary to manually determine whether each decomposed mode belongs to the target signal, so that the flexibility is insufficient. The sparse decomposition algorithm represents the signal in a self-adaptive manner as a linear combination of a plurality of atoms, can describe the local characteristics of the signal in detail, and is an effective non-stationary time delay signal extraction method. However, the traditional signal sparse decomposition algorithm needs to perform matching pursuit in a previously constructed redundant dictionary to determine the optimal matching atom, and the algorithm efficiency is not high enough.
Disclosure of Invention
The invention aims to improve the flexibility and the adaptivity of the identification and the extraction of a time delay target signal, and provides a time delay signal adaptive identification and extraction method based on a bat algorithm optimization sparse decomposition algorithm. The method and the device have the advantages that the dependence on complex prior information in the extraction process of the non-stationary target signal is eliminated, the flexibility of signal identification and extraction is improved, and the detection and extraction of the fully-automatic non-stationary time delay target signal are realized.
The invention is realized by adopting the following technical scheme:
a highly self-adaptive time delay target signal automatic extraction method comprises general atomic design, a target data segment automatic positioning detection algorithm, a real-time characteristic base atom sparse decomposition method, signal extraction based on improved bat algorithm optimization sparse decomposition and the like.
Firstly, designing a general atomic g atom, wherein the expression of the g atom is as follows:
wherein t is sampling time; t is t0For the moment of atom bilateral boundary, τ1And τ2Respectively, bilateral effective oscillation time; d1And d2Respectively, bilateral attenuation factors; f is the atomic signal frequency, phi is the atomic signal phase; ε is a bilateral scale factor.
The constructed g atoms have strong universality, can be matched with most typical signals through parameter adjustment, and almost cover main signal types in electromagnetic observation data.
The method comprises the following specific steps:
(1) and positioning the target data segment by using a target data segment automatic positioning detection algorithm, and then selecting a target-free signal data segment.
(2) Setting sparse decomposition double-threshold criterion: a residual i-2 norm threshold σ and a maximum iteration number threshold N. Sigma is determined by the intensity of background noise, i.e. the sequence n without target signalsThe l-2 norm of (d):n is set by the user according to real-time requirements, typically to 30.
(3) Initializing, wherein the iteration number n is 1, and the number k of selected atoms is 0; current residual xinX; current optimum fitness Fitn best=1E6。
(4) Determining an initialization time parameter tau according to an automatic positioning detection algorithm of the target data segment1,τ2And t0Extracting a target sample Ei=x(t0-τ1:t0+τ2)。
(5) To EiHilbert transformation to obtain more accurate target parameter information
R(t)=Ei(t)*h(t) (2)
Wherein: h (t) is the Hill transform factor. The following complex analytic signals are constructed:
z(t)=x(t)+iR(t)=A(t)eiφ(t) (3)
wherein A (t) is a function of amplitude
The bilateral scaling factor epsilon is determined by the logarithmic ratio of the amplitude maxima and the two adjacent sub-maxima.
Phi (t) is the phase function:
Eiis given by:
using the time-frequency information of the sample as the reference parameter para of the general atom0=[ fφ d1 d2 t0 τ1 τ2 ε]Characteristic base atoms are obtained, the length of the atoms being the same as the original signal length.
(6) Initializing a bat population by using the time-frequency parameters determined in the step (4) and the step (5);whereinDenotes the initial position of the j-th bat of the nth generation, lambdajIs a random number, satisfies lambdaj∈(-1,1);
(7) Finding the optimal bat position after nth sparse decomposition iteration by using an improved bat algorithmBest matching atomsAnd updating the fitness:
wherein N isiteFor the maximum iteration number of the improved bat algorithm, NpopThe number of bats;for real-time flexible atoms, by real-time positioning of bat individualsObtained as a characteristic parameter by substituting the formula (1), i.e.Normalized coefficientGuarantee xinThe effective signal in (a) has the same energy as its sparse decomposition result, is composed ofThe corresponding decomposition coefficients.
if so: then k equals k +1 and the normalized best matching atom is calculatedOptimum atomic numberUpdating residual errors
(9) Let n equal n +1, judgeOr n<And (3) whether N is true, if not, repeating the steps (1) to (9), and if at least one of N is true, executing the step (10).
(10) Obtaining a sparse decomposition optimal matching dictionarySparsely decomposing the step componentsAnd extracting a target signal estimate
Firstly, determining the starting time and the duration time of a characteristic time delay signal by using an automatic positioning detection algorithm of a target data segment, and extracting the target data segment as a characteristic sample by using initial positioning; the target data segment automatic positioning detection algorithm is characterized in that a target signal is determined by searching an effective envelope window of a cross-correlation function of data to be processed and a standard sine atom by utilizing short-time Fourier transform and a related detection technologyThe time information obtained can determine the time domain parameters of the universal atoms, and intercept the target sample Ei. Thirdly, processing the characteristic sample by using Hilbert transform to obtain a target signal time-frequency parameter which is used as a characteristic base of the sparse decomposition dictionary, and constructing a general atom by taking the characteristic base as a center; the real-time flexible atom sparse decomposition method is to perform optimization solution in a characteristic parameter redundancy range by taking an atom parameter as a characteristic value, so as to avoid constructing a huge redundant dictionary. Finally, converting the matching tracking problem in the signal sparse decomposition process into an optimization problem by using an optimized bat algorithm, and adjusting the universal atoms in real time in a redundancy range according to the characteristic parameters to carry out sparse decomposition optimization; the improved bat algorithm is based on the classic bat algorithm, the signal residual error l-2 norm is used as a fitness function for optimizing, the influence of the update direction and speed of the optimal solution of a group on an optimization result is considered in the optimizing process, and the searching speed and precision of the algorithm are improved.
The improved bat algorithm optimized sparse decomposition algorithm is characterized in that the improved bat algorithm is utilized to convert a signal sparse decomposition matching tracking problem into an optimization problem, a general atomic characteristic parameter is used as a bat individual position parameter, a real-time atom is constructed by utilizing the position parameter and the fitness is calculated, the bat position is continuously optimized through iterative optimization to further realize the optimization of the general atomic parameter, a modern standardized optimal matching atom and a signal residual error can be obtained after the optimization of one generation of bats is completed, and the target signal extraction is realized through iterative optimization.
According to the method for automatically extracting the highly-adaptive time delay target signal, the sparse decomposition matching tracking problem is converted into the optimization problem by using the swarm intelligence algorithm, and then the time delay target signal is extracted by using the sparse decomposition algorithm. It is characterized in that:
1. the invention designs the universal atom which can be matched with various typical time delay signals, develops the time-frequency domain characteristic parameter initialization algorithm and improves the automation degree of non-stationary time delay signal identification.
2. The method optimizes the classic bat algorithm, increases the group optimizing direction factor, and improves the convergence speed and precision of the algorithm.
3. According to the method, the problem of matching and tracking of the traditional signal sparse decomposition redundant dictionary is converted into the problem of flexible atomic parameter optimization by using a group intelligent algorithm, so that a huge redundant dictionary is avoided being constructed, and the efficiency of signal sparse decomposition is greatly improved.
4. In the signal sparse decomposition process, the signal residual error l-2 norm is used as a fitness function, the atom time-frequency characteristic parameter is used as an optimization target, the dependence of the existing signal extraction method on various complex prior knowledge, preconditions and later manual intervention is eliminated, and other components of the signal can be furthest reserved while the target signal is extracted.
5. The invention provides a sparse decomposition dual-threshold criterion, a key threshold (a residual error l-2 norm threshold) is adaptively determined by an algorithm according to the background noise intensity, an iteration number threshold N is set by a user according to the real-time requirement, and the application of the dual-threshold criterion can ensure the signal extraction precision, improve the iteration speed and effectively avoid excessive decomposition.
The method is reasonable in design, is a fully-automatic time delay signal detection and extraction method, and has the advantages of high adaptivity, high efficiency, high precision and the like.
Drawings
Fig. 1 shows a time-domain waveform of the different parameters g atom.
Fig. 2 shows an automatic target data segment location detection algorithm.
Fig. 3 shows an example of automatic location detection of a target data segment.
Fig. 4 shows a flow chart of a highly adaptive delay target signal automatic extraction method.
Fig. 5 shows the non-stationary delay signal extraction results of different methods.
Figure 6 shows different approaches to non-stationary delay signal extraction error.
Detailed Description
The following provides a detailed description of specific embodiments of the present invention.
Height self-adaptationThe automatic extraction method of the time delay target signal is characterized in that a signal sparse decomposition method is utilized to extract a non-stable time delay signal, and the problem of sparse decomposition, matching and tracking of the signal is converted into an optimization problem through an improved bat algorithm. The method mainly comprises the following steps: first, the generic atoms of the design can be matched to most typical signals by parameter tuning. Secondly, the improved bat algorithm introduces a group optimizing direction factor, the group optimizing direction factor is a vector and normalization result formed by overlapping the speed of the current bat individual and the speeds of n bats nearby, the speed factor is updated in the group optimizing direction, and the bat individual can be effectively guided to jump out of a local optimal solution. Then, optimizing sparse decomposition based on the improved bat algorithm, without constructing a redundant dictionary in advance, adopting real-time flexible atoms, searching for optimal matching atoms by taking a residual error l-2 norm as a fitness function, calculating the optimal matching atoms and updating the residual error. Finally, a sparse decomposition double-threshold criterion is provided, namely a residual error l-2 norm threshold and an iteration number threshold N, so that the circulation is ended in advance after the residual error l-2 norm threshold is reached in the signal sparse decomposition process, excessive decomposition is avoided, and the decomposition efficiency is improved; and when the iteration times are more than N, ending the iteration, and avoiding the low efficiency caused by infinite loop of the algorithm. After iteration is stopped, the original signal is used for subtracting the final residual error to obtain the target signal estimatorMeanwhile, the optimal matching dictionary for signal sparse decomposition is composed of optimal matching atoms obtained by each iteration.
First, a general atomic g atom is designed, and most typical signals can be matched through parameter adjustment. The universal atom is subjected to standardization processing, so that the original signal and the sparsely decomposed signal have the same energy, and the universal atom characteristic parameter vector comprises effective signal start-stop time, attenuation factors, frequency, phase and damping ratio: para ═ f φ d1 d2 t0 τ1 τ2ε]。
Wherein t is sampling time; t is t0For the moment of atom bilateral boundary, τ1And τ2Respectively, bilateral effective oscillation time; d1And d2Respectively, bilateral attenuation factors, usually d1=d2(ii) a f is the atomic signal frequency, phi is the atomic signal phase; ε is a bilateral scaling factor, typically 1.
As shown in FIG. 1, the time domain waveform of atomic g with different parameters is used as the damping factor d1=d2When 0, epsilon is 1, g atoms degenerate into a standard sine wave (r); when ε is 1, τ1=τ2Not equal to 0, the damping factor is increased, the high-frequency g atoms are shown as bilateral oscillation attenuation signals and are matched with the characteristics of a sharp impulse discharge signal ((r)), and the low-frequency g atoms are shown as triangle-like waves (sixth); when tau is2When the number is 0, g atoms show unilateral excited oscillation characteristics (c- (r)); when tau is10, g atoms show unilateral damped oscillation characteristic (c); when the bilateral scale factor is 0<ε<1, and when the atomic frequency is sufficiently low, g atoms represent a charge-discharge-like triangular wave (v).
Secondly, the invention adopts an automatic target data segment positioning detection algorithm based on a relevant detection technology. The correlation detection technology is a strong periodic signal extraction method, and the technology separates target signals by utilizing the irrelevance between signals and noise and signals with different components.
As shown in FIG. 2, the present invention utilizes the cross-correlation function envelope and the peak gradient as the characteristic values to perform the automatic segmentation extraction of the target signal.
Let the original sample sequence x (t) consist of two parts, x (t) xn(t)+s0(t) wherein xn(t) is a nonlinear, non-stationary signal; s0(t) is a target signal in the x sequence, and the expression is shown as 8
Wherein A isi、fi、ΦiAre respectively target signalsAmplitude, frequency and phase of the main frequency component, ti1~ti2Satisfying t for a target frequency signal occurrence periodi1<ti2<T, T is the total sampling time.
Firstly, preliminarily determining an estimated value of a key parameter of a signal to be extracted in original observation data x through short-time Fourier transformThen constructing standard sine atoms
Determination of a target signal s using correlation detection techniques0iAt the time instant that occurs in the x sequence. x (t),The cross-correlation function between the two signals is:
wherein x isnNon-stationary components of the original signal, andnot related;is prepared by reacting withThe less correlated signal component. The first term of equation (9) is therefore approximately 0, i.e.:
as shown in the formula (10), the original sequence x (t) is related to the standard sinusoidal signalThe cross-correlation function envelope period is [ -t [ ]1-1 t2+1]。
For digital signals, equation (10) becomes:
wherein L is1And L2The start and end positions of the target signal are respectively, and N is the length N ═ f of the original data sequences*T,fsFor the sampling rate, K is the length of the standard sinusoidal signal K ═ fs。
In practical application, because the first term in the formula (9) is not absolutely 0, the maximum peak gradient is used as a characteristic parameter, and the gradient Delta is enveloped by the cross-correlation functionp:
Δp(i)=Penv(i)-Penv(i-1) (12)
Wherein, Penv(i) Represents the value of the envelope of the cross-correlation function at the coordinate point i, when i is L1Where the maximum peak gradient of the cross-correlation function occurs.
And extracting the envelope curve of the cross-correlation function by using a soft threshold, wherein the ideal envelope range of the cross-correlation function is given by the following formula:
wherein the content of the first and second substances,andfor enveloping the effective window soft threshold, λ, from top to bottomp,λnE (0,1), related to the signal-to-noise ratio; [ lp1lp2]And [ ln1 ln2]Respectively upper and lower effective envelope window boundaries. The target signal data segment range is given by:
wherein n isshiftIn order to be a factor of the offset,represents rounding down, [ l1l2]For the active envelope window boundary, max () denotes taking the maximum value and min () denotes taking the minimum value. Lambda is a time parameter redundancy factor and satisfies lambda epsilon [ 1.011.2]And ensuring that the determined time parameter can completely cover the target signal segment. Extracting a target data segment Ei=x(L1:L2) By using EiMaximum extreme point AmaxAnd its two side symmetric minor extreme pointsRatio judgment EiTime-domain oscillation characteristic of (2):the single-side attenuation oscillation is carried out,the single side is excited to oscillate,bilateral oscillation. From the above information, the time parameter is determined using equation (15):
Eithe characteristic of single-side oscillation attenuation is as follows:Eisingle-side stimulated oscillation characteristic:
as shown in fig. 3, in the example of automatic positioning detection of a target data segment, the upper graph is a positioning result, and the lower graph is an original signal time domain waveform. The upper and lower effective envelopes are shown as "at" and "at" scatter points in the figure, and "O" marks the time-localized position.
Thirdly, the improved bat algorithm flow adopted by the invention is as follows:
(1) initializing bat populations, wherein the bat number is NpopMaximum number of iterations NiteWhen the current iteration number m is equal to 0, searching a space dimension q, and initializing a pulse frequency f0={ f i 01,2, q } and a maximum value f of pulse frequencymaxAnd minimum value fminSound wave loudnessAnd maximum pulse transmission frequencyRandomly generating a bat group initial position according to the following formula Wherein: lambda [ alpha ]jTo satisfy the random number of the Gaussian observation matrix, Pmax=1.2*para0,Pmin=0.8*para0。
(2) According to bat group positionCreating NpopFlexible atom formationn. For residual signal xinCalculating the bat individual fitness according to formula (16):
wherein the content of the first and second substances,in order to be a function of the normalization,
determining and storing the optimal bat individual position of the current group according to the formula (17)And optimal fitness.
(4) speed of updating bat individualAnd positionThe bat individual moving direction and speed have certain relevance to the global optimal solution. Thus introducing a population-optimizing direction factor. The bat individual direction update factor is as follows:
in order to avoidGuiding the bat individual to move to the local optimal solution, and comparing the speed of the current bat individual with the nearby nrThe optimal directions of bat are added:
wherein, | | | represents the euclidean distance, and R is the group velocity factor threshold.
by usingThe bat group is guided to rapidly move to the direction of the global optimal solution, and the updating formula of the speed and the pulse emission frequency is as follows:
wherein r is1Is a random number, satisfies r1∈[0,1];fiIs the search pulse frequency of the ith bat;respectively representing the speed of the jth bat when the iteration times are m and m + 1; μ is a weighting factor.
(5) Generating a random number r for each bat2j∈[0,1]Updated bat position according to equation (21):
wherein λ isrjIs a random number satisfying lambdarj(-0.3 0.3),Is the mean value of the sound wave loudness of the m-th generation bat swarm,respectively representing the position of the jth bat at the iteration times of m and m +1, and calculatingDegree of adaptability
(7) Judging whether the algorithm meets the iteration termination conditionIf not, executing the step (8); if m is m +1, k is k +1, and jump to step (11) after updating the residual error according to the following formula
WhereinIs the normalized optimal atom for which the atomic number is normalized,are the corresponding sparse decomposition coefficients.
(8) For each bat, a random number r is generated3jAnd updating the bat position by formula (23)
(9) Updating the loudness of the bats individual sound wave and the pulse emission frequency by a formula (24):
wherein the content of the first and second substances,representing the pulse emission frequency of the jth bat when the iteration number is m + 1;respectively representing the loudness of the sound wave of the jth bat when the iteration times are m and m-1; lambda is the attenuation coefficient of the pulse emission frequency, and belongs to (0, 1); gamma is the pulse emission frequency enhancement coefficient>0; when m → ∞ is reached,
(10) and judging whether the algorithm meets an iteration termination condition m or not by making m equal to m +1<NiteIf not, repeating the step (3-10); and (4) if the condition is met, executing the step (11).
designing a general atom, setting atom characteristic parameters as bat individual position vectors without constructing a huge redundant dictionary in advance when sparse decomposition is carried out, and constructing flexible atoms by utilizing an improved bat algorithm to carry out optimal position search so as to realize optimal atom search; adopting sparse decomposition double-threshold iteration stopping criterion, wherein the threshold of residual error l-2 norm is determined by background noise intensity, namely the threshold does not contain target signal sequence nsThe l-2 norm of (1) is determined adaptively; the iteration number threshold value N is set by a user according to real-time requirements. The application of the double criteria can ensure the signal extraction precision, improve the iteration speed and effectively avoid over decomposition.
The specific implementation flow is shown in fig. 4, and is as follows:
(1) and positioning the target data segment by using a target data segment automatic positioning detection algorithm, and then selecting a target-free signal data segment.
(2) Setting sparse decomposition double-threshold criterion: residual l-2 norm threshold σAnd a maximum iteration number threshold N. Sigma is determined by the intensity of background noise, i.e. the sequence n without target signalsThe l-2 norm of (d):n is set by the user according to real-time requirements, typically to 30.
(3) Initializing, wherein the iteration number n is 1, and the number k of selected atoms is 0; current residual xinX; current optimum fitness Fitn best=1E6。
(4) Determining an initialization time parameter tau according to an automatic positioning detection algorithm of the target data segment1,τ2And t0Extracting a target sample Ei=x(t0-τ1:t0+τ2)。
(5) To EiHilbert transformation to obtain more accurate target parameter information
R(t)=Ei(t)*h(t) (2)
Wherein: h (t) is the Hill transform factor. The following complex analytic signals are constructed:
z(t)=x(t)+iR(t)=A(t)eiφ(t) (3)
wherein A (t) is a function of amplitude
The bilateral scaling factor epsilon is determined by the logarithmic ratio of the amplitude maxima and the two adjacent sub-maxima.
Phi (t) is the phase function:
Eiis given by:
using the time-frequency information of the sample as the reference parameter para of the general atom0=[f φ d1 d2 t0 τ1 τ2 ε]Characteristic base atoms are obtained, the length of the atoms being the same as the original signal length.
(6) Initializing a bat population by using the time-frequency parameters determined in the step (4) and the step (5);whereinDenotes the initial position of the j-th bat of the nth generation, lambdajIs a random number, satisfies lambdaj∈(-1,1)。
(7) Finding the optimal bat position after nth sparse decomposition iteration by using an improved bat algorithmBest matching atomsAnd updating the fitness:
wherein N isiteFor the maximum iteration number of the improved bat algorithm, NpopThe number of bats;for real-time flexible atoms, by real-time positioning of bat individualsObtained as a characteristic parameter by substituting the formula (1), i.e.Normalized coefficientGuarantee xinThe effective signal in (a) has the same energy as its sparse decomposition result, is composed ofThe corresponding decomposition coefficients.
if so: then k equals k +1 and the normalized best matching atom is calculatedOptimum atomic numberUpdating residual errors
(9) Let n equal n +1, judgeOr n<And (3) whether N is true, if not, repeating the steps (1) to (9), and if at least one of N is true, executing the step (10).
(10) Obtaining a sparse decomposition optimal matching dictionarySparsely decomposing the step componentsAnd extracting a target signal estimate
Fifthly, to verify the reliability of the algorithm, the method is characterized by comprising the following steps of background noise nsAnd a time delay target signal s consisting of g atoms with two different parameters constructs a non-stationary original signal x:
x=s+ns=ns+150·g(c,0.3,10,10,13,2,π*1.1,1)+100·g(c,1,2,1,3,5,π/2,1)。
respectively utilizing wavelet analysis, VMD algorithm, sparse decomposition algorithm based on particle swarm algorithm and the method of the invention to identify and extract the time delay target signal s contained in the signal x. Wherein, the wavelet decomposition adopts an expert experience method to preferably select sym3 wavelet as a mother wavelet, and the decomposition layer number is 3 layers; the number of intrinsic modes of the VMD algorithm is set to be 4, and the first two modes are taken as extraction targets; the particle swarm algorithm and the bat algorithm adopt the same initial parameters of the population scale, the maximum iteration times, the stop threshold value and the like.
As shown in fig. 5 and 6, in the non-stationary delay signal extraction result and the extraction error map of different methods, both wavelet decomposition and VMD delay signal extraction cause global aliasing of local features, and the signal sparse decomposition method (PSO-SD) based on the particle swarm algorithm causes a large deviation between the extracted signal and the target signal due to the fact that the extracted signal falls into a local optimal solution, so that the method (IBA-SD) of the present invention has an ideal non-stationary delay signal extraction effect compared with other methods.
The result shows that the method does not depend on complex prior knowledge and subsequent manual judgment and screening, can effectively extract the target signal and simultaneously reserve other components of the signal to the maximum extent, is a fully-automatic non-stationary time delay signal extraction method, and has the advantages of high adaptivity, high convergence, high precision and the like.
Finally, it should be noted that the above embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the detailed description is made with reference to the embodiments of the present invention, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which shall be covered by the claims of the present invention.
Claims (5)
1. A highly adaptive time delay target signal automatic extraction method is characterized in that: the method comprises the following steps:
firstly, designing a general atomic g atom, wherein the expression of the g atom is as follows:
wherein t is sampling time; t is t0For the moment of atom bilateral boundary, τ1And τ2Respectively, bilateral effective oscillation time; d1And d2Respectively, bilateral attenuation factors; f is the atomic signal frequency, phi is the atomic signal phase; epsilon is a bilateral scale factor;
the method comprises the following specific steps:
(1) positioning the target data segment by using a target data segment automatic positioning detection algorithm, and then selecting a target-free signal data segment;
(2) setting sparse decomposition double-threshold criterion: a residual error l-2 norm threshold value sigma and a maximum iteration number threshold value N; sigma is determined by the intensity of background noise, i.e. the sequence n without target signalsThe l-2 norm of (d):n is set by a user according to real-time requirements and is typically set to be 30;
(3) initializing, wherein the iteration number n is 1, and the number k of selected atoms is 0; current residual xinX; current optimum fitness Fitn best=1E6;
(4) Determining an initialization time parameter tau according to an automatic positioning detection algorithm of the target data segment1,τ2And t0Extracting a target sample Ei=x(t0-τ1:t0+τ2);
(5) To EiHilbert transformation to obtain more accurate target parameter information
R(t)=Ei(t)*h(t) (2)
Wherein: h (t) is the Hill transform factor. The following complex analytic signals are constructed:
z(t)=x(t)+iR(t)=A(t)eiφ(t) (3)
wherein A (t) is a function of amplitude
The bilateral scaling factor epsilon is determined by the logarithmic ratio of the amplitude maxima and the two adjacent sub-maxima.
Phi (t) is the phase function:
Eiis given by:
using the time-frequency information of the sample as the reference parameter para of the general atom0=[f φ d1 d2 t0 τ1 τ2 ε]Obtaining characteristic base atoms, wherein the length of the atoms is the same as that of the original signals;
(6) initializing a bat population by using the time-frequency parameters determined in the step (4) and the step (5);whereinDenotes the initial position of the j-th bat of the nth generation, lambdajIs a random number, satisfies lambdaj∈(-1,1);
(7) Finding the optimal bat position after nth sparse decomposition iteration by using an improved bat algorithmBest matching atomsAnd updating the fitness:
wherein N isiteFor the maximum iteration number of the improved bat algorithm, NpopThe number of bats;for real-time flexible atoms, by real-time positioning of bat individualsObtained as a characteristic parameter by substituting the formula (1), i.e.Normalized coefficientGuarantee xinThe effective signal in (a) has the same energy as its sparse decomposition result, is composed ofThe corresponding decomposition coefficient;
if so: then k equals k +1 and the normalized best matching atom is calculatedOptimum atomic numberUpdating residual errors
(9) Let n equal n +1, judgeOr n<N is true, if not, repeating the steps (1) to (9), and if at least one of the N is true, executing the step (10);
2. The method according to claim 1, wherein the method comprises: in the step (3), the target data segment automatic positioning detection algorithm is as follows:
let the original sample sequence x (t) consist of two parts, x (t) xn(t)+s0(t) wherein xn(t) is a nonlinear, non-stationary signal; s0(t) is a target signal in the x sequence, and the expression is shown as 8
Wherein A isi、fi、ΦiAmplitude, frequency and phase, t, respectively, of the main frequency component of the target signali1~ti2Satisfying t for a target frequency signal occurrence periodi1<ti2<T, T is total sampling time;
firstly, preliminarily determining an estimated value of a key parameter of a signal to be extracted in original observation data x through short-time Fourier transformThen constructing standard sine atoms
Determination of a target signal s using correlation detection techniques0iThe time instants occurring in the x sequence; x (t),The cross-correlation function between the two signals is:
wherein x isnIs original toNon-stationary components in the start signal, andnot related;is prepared by reacting withA signal component with weak correlation; the first term of equation (9) is therefore approximately 0, i.e.:
as shown in the formula (10), the original sequence x (t) is related to the standard sinusoidal signalThe cross-correlation function envelope period is [ -t [ ]1-1 t2+1];
For digital signals, equation (10) becomes:
wherein L is1And L2The start and end positions of the target signal are respectively, and N is the length N ═ f of the original data sequences*T,fsFor the sampling rate, K is the length of the standard sinusoidal signal K ═ fs;
In practical application, because the first term in the formula (9) is not absolutely 0, the maximum peak gradient is used as a characteristic parameter, and the gradient Delta is enveloped by the cross-correlation functionp:
Δp(i)=Penv(i)-Penv(i-1) (12)
Wherein, Penv(i) Represents the value of the envelope of the cross-correlation function at the coordinate point i, when i is L1Where cross correlation function occursA maximum peak gradient;
and extracting the envelope curve of the cross-correlation function by using a soft threshold, wherein the ideal envelope range of the cross-correlation function is given by the following formula:
wherein the content of the first and second substances,andfor enveloping the effective window soft threshold, λ, from top to bottomp,λnE (0,1), related to the signal-to-noise ratio; [ lp1 lp2]And [ ln1 ln2]Upper and lower effective envelope window boundaries, respectively; the target signal data segment range is given by:
wherein n isshiftIn order to be a factor of the offset,represents rounding down, [ l1l2]For the effective envelope window boundary, max () represents taking the maximum value, min () represents taking the minimum value; lambda is a time parameter redundancy factor and satisfies lambda epsilon [ 1.011.2]Ensuring that the determined time parameter can completely cover the target signal segment; extracting a target data segment Ei=x(L1:L2) By using EiMaximum extreme point AmaxAnd its two side symmetric minor extreme pointsRatio judgment EiTime-domain oscillation characteristic of (2):
the single-side attenuation oscillation is carried out,the single side is excited to oscillate,bilateral oscillation; from the above information, the time parameter is determined using equation (15):
3. the method according to claim 1, wherein the method comprises: in the step (6), the improved bat algorithm is as follows:
(1) initializing bat populations, wherein the bat number is NpopMaximum number of iterations NiteWhen the current iteration number m is equal to 0, searching a space dimension q, and initializing a pulse frequency f0={fi 01,2, q } and a maximum value f of pulse frequencymaxAnd minimum value fminSound wave loudnessAnd maximum pulse transmission frequencyGeneration of the bat randomlyInitial position of bats groupWherein: lambda [ alpha ]jTo satisfy the random number of the Gaussian observation matrix, Pmax=1.2*para0,Pmin=0.8*para0;
(2) According to bat group positionCreating NpopFlexible atom formationn(ii) a For residual signal xinCalculating the bat individual fitness according to formula (16):
wherein the content of the first and second substances,in order to be a function of the normalization,
determining and storing the optimal bat individual position of the current group according to the formula (17)And the optimal fitness;
(4) speed of updating bat individualAnd positionThe moving direction and speed of the bat individual have certain relevance with the global optimal solution, so a group optimizing direction factor is introduced; the bat individual direction update factor is as follows:
in order to avoidGuiding the bat individual to move to the local optimal solution, and comparing the speed of the current bat individual with the nearby nrThe optimal directions of bat are added:
wherein, | | | represents the euclidean distance, and R is the group velocity factor threshold;
normalizing the vector sum to obtain a normalized population optimizing direction factor:by usingThe bat group is guided to rapidly move to the direction of the global optimal solution, and the updating formula of the speed and the pulse emission frequency is as follows:
wherein r is1Is a random number, satisfies r1∈[0,1];fiIs the search pulse of the ith batThe impulse frequency;respectively representing the speed of the jth bat when the iteration times are m and m + 1; mu is a weight factor;
(5) generating a random number r for each bat2j∈[0,1]Updated bat position according to equation (21):
wherein λ isrjIs a random number satisfying lambdarj(-0.3 0.3),Is the mean value of the sound wave loudness of the m-th generation bat swarm,respectively representing the position of the jth bat when the iteration times are m and m +1, and calculating the fitness
(7) Judging whether the algorithm meets the iteration termination conditionIf not, executing the step (8); if m is m +1, k is k +1, and jump to step (11) after updating the residual error according to the following formula
(8) for each bat, a random number r is generated3jAnd updating the bat position by formula (23)
(9) Updating the loudness of the bats individual sound wave and the pulse emission frequency by a formula (24):
wherein the content of the first and second substances,representing the pulse emission frequency of the jth bat when the iteration number is m + 1;respectively representing the loudness of the sound wave of the jth bat when the iteration times are m and m-1; lambda is the attenuation coefficient of the pulse emission frequency, and belongs to (0, 1); gamma is the pulse emission frequency enhancement coefficient>0; when m → ∞ is reached,
(10) and judging whether the algorithm meets an iteration termination condition m or not by making m equal to m +1<NiteIf not, repeat the first step(3-10); if yes, executing the step (11);
4. the method according to claim 3, wherein the method comprises: the real-time flexible atom is a bat individual real-time position vectorConstructing a universal atom in real time as a universal atom characteristic parameter vector; finding the optimal solution of the bat position by using the residual error l-2 norm as a fitness function to further obtain a standardized optimal matching atom gk bestThe method is characterized in that real-time flexible atoms which are obtained by one-time iteration optimization and enable residual error l-2 norm to be minimum are obtained after standardization processing; the optimal atomic coefficients are obtained from the projection of the corresponding residual signals on the normalized optimally matched atoms, i.e.
5. The method of claim 4, wherein the method further comprises: the sparse decomposition double-threshold criterion is that the residual error l-2 norm threshold sigma is determined by the background noise intensity, namely the target signal sequence n is not containedsThe l-2 norm of (d):the iteration number threshold value N is set by a user according to the real-time requirement, and is typically set to be 30; the target signal estimatorFrom the original signal and the last iteration residualξnAnd (3) calculating:the signal sparse decomposition optimal matching dictionary is a dictionary formed by all optimized normalized optimal matching atoms when iteration stopsAll components of the sparse decomposition are obtained by calculating normalized optimal matching atoms and optimal atom coefficients:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110288519.4A CN112906632B (en) | 2021-03-18 | 2021-03-18 | Automatic extraction method of highly adaptive time delay target signal |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110288519.4A CN112906632B (en) | 2021-03-18 | 2021-03-18 | Automatic extraction method of highly adaptive time delay target signal |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112906632A true CN112906632A (en) | 2021-06-04 |
CN112906632B CN112906632B (en) | 2022-11-11 |
Family
ID=76105708
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110288519.4A Active CN112906632B (en) | 2021-03-18 | 2021-03-18 | Automatic extraction method of highly adaptive time delay target signal |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112906632B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115659162A (en) * | 2022-09-15 | 2023-01-31 | 云南财经大学 | Method, system and equipment for extracting features in radar radiation source signal pulse |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7561756B1 (en) * | 2005-05-02 | 2009-07-14 | Nanostellar, Inc. | Particle shape characterization from 2D images |
CN103995973A (en) * | 2014-05-27 | 2014-08-20 | 哈尔滨工业大学 | Signal sparse decomposition method based on set partitioning of over-complete dictionary |
US9070012B1 (en) * | 2012-01-11 | 2015-06-30 | Sr2 Group, Llc | System and method for uncued discrimination of bated features in image |
CN108833312A (en) * | 2018-06-08 | 2018-11-16 | 西北工业大学 | It is a kind of based on delay-Doppler domain when be thinned out underwater acoustic channel estimation method |
CN109507304A (en) * | 2018-12-26 | 2019-03-22 | 西安科技大学 | A kind of defect inspection method based on ultrasonic inspection |
CN109507292A (en) * | 2018-12-26 | 2019-03-22 | 西安科技大学 | A kind of method for extracting signal |
CN109632973A (en) * | 2018-12-26 | 2019-04-16 | 西安科技大学 | A kind of ultrasound echo signal extracting method based on Based on Multiscale Matching tracking |
CN111678691A (en) * | 2020-07-27 | 2020-09-18 | 江南大学 | Gear fault detection method based on improved sparse decomposition algorithm |
CN112147571A (en) * | 2020-08-31 | 2020-12-29 | 南京理工大学 | Sound source azimuth angle estimation method based on regular orthogonal matching pursuit and bat algorithm |
-
2021
- 2021-03-18 CN CN202110288519.4A patent/CN112906632B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7561756B1 (en) * | 2005-05-02 | 2009-07-14 | Nanostellar, Inc. | Particle shape characterization from 2D images |
US9070012B1 (en) * | 2012-01-11 | 2015-06-30 | Sr2 Group, Llc | System and method for uncued discrimination of bated features in image |
CN103995973A (en) * | 2014-05-27 | 2014-08-20 | 哈尔滨工业大学 | Signal sparse decomposition method based on set partitioning of over-complete dictionary |
CN108833312A (en) * | 2018-06-08 | 2018-11-16 | 西北工业大学 | It is a kind of based on delay-Doppler domain when be thinned out underwater acoustic channel estimation method |
CN109507304A (en) * | 2018-12-26 | 2019-03-22 | 西安科技大学 | A kind of defect inspection method based on ultrasonic inspection |
CN109507292A (en) * | 2018-12-26 | 2019-03-22 | 西安科技大学 | A kind of method for extracting signal |
CN109632973A (en) * | 2018-12-26 | 2019-04-16 | 西安科技大学 | A kind of ultrasound echo signal extracting method based on Based on Multiscale Matching tracking |
CN111678691A (en) * | 2020-07-27 | 2020-09-18 | 江南大学 | Gear fault detection method based on improved sparse decomposition algorithm |
CN112147571A (en) * | 2020-08-31 | 2020-12-29 | 南京理工大学 | Sound source azimuth angle estimation method based on regular orthogonal matching pursuit and bat algorithm |
Non-Patent Citations (2)
Title |
---|
MING, Z: "Feature extraction of radar emitter signals based on gaussian chirplet atoms", 《JOURNAL OF INFRARED AND MILLIMETER WAVES 》 * |
张菀: "基于信号稀疏表示的滚动轴承微弱故障特征提取方法研究", 《中国优秀博士论文电子期刊网》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115659162A (en) * | 2022-09-15 | 2023-01-31 | 云南财经大学 | Method, system and equipment for extracting features in radar radiation source signal pulse |
CN115659162B (en) * | 2022-09-15 | 2023-10-03 | 云南财经大学 | Method, system and equipment for extracting intra-pulse characteristics of radar radiation source signals |
Also Published As
Publication number | Publication date |
---|---|
CN112906632B (en) | 2022-11-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20220318635A1 (en) | Energy identification method for micro-energy device based on bp neural network | |
CN107703486B (en) | Sound source positioning method based on convolutional neural network CNN | |
CN110865357A (en) | Laser radar echo signal noise reduction method based on parameter optimization VMD | |
CN109192200B (en) | Speech recognition method | |
CN109977724B (en) | Underwater target classification method | |
Yang et al. | A new cooperative deep learning method for underwater acoustic target recognition | |
CN112634882B (en) | End-to-end real-time voice endpoint detection neural network model and training method | |
CN112035696B (en) | Voice retrieval method and system based on audio fingerprint | |
CN112053694A (en) | Voiceprint recognition method based on CNN and GRU network fusion | |
CN113111786B (en) | Underwater target identification method based on small sample training diagram convolutional network | |
CN113657491A (en) | Neural network design method for signal modulation type recognition | |
CN112906632B (en) | Automatic extraction method of highly adaptive time delay target signal | |
CN117609702A (en) | Pipeline leakage acoustic emission signal denoising method, system, equipment and medium | |
Yang et al. | Underwater acoustic signal prediction based on correlation variational mode decomposition and error compensation | |
CN112034434A (en) | Radar radiation source identification method based on sparse time-frequency detection convolutional neural network | |
CN114565828A (en) | Feature countermeasure enhancement underwater target recognition method based on acoustic embedded memory space encoder model | |
CN112052880A (en) | Underwater sound target identification method based on weight updating support vector machine | |
CN102237082B (en) | Self-adaption method of speech recognition system | |
WO2023093029A1 (en) | Wake-up word energy calculation method and system, and voice wake-up system and storage medium | |
Wei et al. | Adaptive single-tone waveform design for target recognition in cognitive radar | |
CN115932808A (en) | Passive sonar intelligent detection method based on multi-feature fusion | |
US7180442B1 (en) | Target indentification method using cepstral coefficients | |
CN114636995A (en) | Underwater sound signal detection method and system based on deep learning | |
CN111563411B (en) | Electrocardiosignal classification method using AdaBoost and weak classifier | |
CN113625101A (en) | Traveling wave signal processing method based on fruit fly algorithm and stochastic resonance |
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 |