39 results about "Underdetermined blind source separation" patented technology

The invention discloses a full-wave nuclear magnetic resonance signal denoising method based on independent component analysis. The full-wave nuclear magnetic resonance signal denoising method is mainly used for power frequency harmonic interference or certain single-frequency interference in full-wave nuclear magnetic resonance signals. The method comprises the steps that firstly, utilizing a nuclear magnetic resonance sounding water detecting instrument for collecting MRS signals, obtaining the frequency of power frequency harmonic interference or certain single-frequency interference contained in the collected signals through spectral analysis, and adopting a digital orthogonal method for constructing input channel signals to solve the underdetermined blind source separation problem; then, using the constructed input channel signals and the collected MRS signals as input signals for conducting independent component analysis to obtain separated MRS signals; finally, adopting the spectrum correcting method for solving the problem that the separated MRS signals have uncertain amplitude, and then extracting denoised MRS signals. According to the full-wave nuclear magnetic resonance signal denoising method, the problem of the underdetermined blind source and the problem of the uncertain amplitude usually existing in independent component analysis are solved, the power frequency harmonic interference or the certain single-frequency interference in the noised MRS signals is effectively filtered out, and compared with a traditional MRS signal denoising method, the full-wave nuclear magnetic resonance signal denoising method has the advantages of being high in arithmetic speed, high in signal to noise ratio, high in practicability and the like.

The invention discloses a matrix-transformation-based method for underdetermined blind source separation. The method comprises the following steps of: gradually transforming a mixed matrix to obtain a transformed matrix, then acting the transformed matrix on an observation signal to gradually remove each source signal, and constructing multi-stage binary shielding templates by zero points generated in each observation signal to gradually separate the source signal. By the method, the requirement on sparsity of the source signal is reduced; aliasing of at most M-1 paths of source signals is prevented (wherein M is the number of sensors); source signals for forming each time frequency point can be clearly known; the underdetermined separation problem of music signals and noise signals is solved; the statistical property requirement on the source signal is low; the underdetermined separation problem of Gaussian signals and related signals is solved; processing stages are adjusted according to the requirement of separation precision; when the processing level is more, the separation result is better; the separation process is realized by aid of the matrix transformation; and the complexity is relatively low.

The invention discloses an underdetermined blind source separation (UBSS) method based on a maximum matrix diagonal rate. The method comprises the following steps of: constructing inverse matrixes of C*M / N M*M-dimensional sub matrixes of a mixed matrix (wherein M and N are respectively the number of sensors and the number of source signals); multiplying the inverse matrixes by observation signal vectors to acquire initial estimation signal vectors; and sequentially calculating the covariance matrix, the solid matrix, the absolute value matrix and the diagonal rate of each initial estimation signal vector, selecting the initial estimation signal vector corresponding to the maximum diagonal rate as estimation of a source signal vector, and thus realizing underdetermined separation of sourcesignals. By the method, the requirement for source signal sparseness is reduced, aliasing of road source signals is allowed at each time frequency point at most, and the underdetermined separation problem of music signals and noise signals is solved. The requirement for the statistical property of the source signals is low, and the underdetermined separation problem of Gaussian signals and related signals is solved. In addition, by the method, processing of each time frequency point and each sub matrix can be executed in parallel, and hardware implementation is facilitated.

The invention discloses an underdetermined blind source separation method based on density, mainly solving the problems that the computation complexity of the prior art is high, the prior art is easily influenced by an initial value and the number of source signals needs to be given. The method comprises the following steps: after the low energy sampling data of an observed signal is removed, projecting the observed signal onto a right half unit hypersphere; computing the density parameters of all projective points, and deleting the projective points with smaller density; utilizing an improved K-mean value clustering algorithm to cluster the surplus projective points and determining the optimal clustering number and clustering center; removing the cluster with less number of data objects, wherein the number of the surplus clusters is the estimated value of the number of source signals, and the corresponding clustering center is the estimated value of each column vector of a confusion matrix; and according to the observed signal and the estimated confusion matrix, adopting a linear programming method to recover the source signals. According to the method, the computation complexity is reduced, the influence of the initial value on the estimated performance is reduced, the confusion matrix can be estimated when the number of the source signals is unknown, and the estimating precision of the confusion matrix and the source signals can be increased.

The invention belongs to the technical field of array signal processing and discloses an improved underdetermined blind source separation source number estimation method based on wavelet analysis. The method is characterized by carrying out decomposition on observation signals through wavelet transform; carrying out single-layer reconstruction on wavelet coefficients obtained through decomposition; reconstructing an observation signal matrix and constructing a covariance matrix of the matrix; calculating characteristic values of the matrix; and calculating source number. The method can enable whitened observation signals and a part of wavelet decomposition coefficient reconstruction signals thereof to be combined into new multidimensional observation signals, can accurately carry out estimation under the condition of large signal-to-noise ratio change range of mixed signals to obtain the number of source signals, and is simple in principle and high in operability.

The invention discloses a single source point detection-based underdetermined blind source separation method. The method comprises the steps of firstly performing short-time Fourier transform on to-be-analyzed observation signals to obtain corresponding time frequency domain complex matrixes; secondly vectorizing and normalizing the time frequency complex matrix of each observation signal; thirdlydetecting out all equal column vectors in the normalized time frequency matrix by utilizing cosine included angle criteria of the vectors, wherein the extracted column vectors are namely single source points; fourthly performing hierarchical clustering on the extracted single source points to obtain a clustering center, thereby realizing estimation of a mixed matrix, wherein the center of each type corresponds to a column of the mixed matrix; and finally by utilizing the estimated mixed matrix, realizing time frequency estimation of all sources at all time frequency points through a least square method, and obtaining time domain forms of the sources through time domain inverse transform. According to the method disclosed by the invention, a linear relationship among different single source points is considered, and the detection of the single source points can be realized only by judging whether the vectors are same or not, so that efficient and high-precision estimation of the mixedmatrix and source signals can be realized under the underdetermined condition.

The invention discloses a hybrid matrix recognition method in underdetermined blind source separation based on tensor regular decomposition. The method mainly solves the problem that in the prior art, hybrid matrix estimation is limited by specific conditions. The method includes the implementation steps that (1), source signals are sampled and observation data are acquired; (2), four-order covariance matrixes in different time delays are calculated through four-order cumulants of the observation data; (3), the four-order covariance matrix in different time delays are expanded into a three-order tensor mode; (4), tensor regular decomposition is conducted on three-order tensors, and a Khatri-Rao product matrix of a hybrid matrix to be recognized is acquired; (5), the product matrix is processed through an eigenvalue decomposition method, and an estimated value of the hybrid matrix is acquired. The method has the advantage of being high in recognition accuracy and can be used in the fields of voice, communication, radar and biomedicine and used for underdetermined blind source separation under the time-frequency aliasing condition.

The invention discloses an underdetermined blind source separation method applying multiple constraints, which is improved on the basis of a traditional blind source separation method to realize a better signal separating effect. The method disclosed by the invention aims to separate mixed signals and comprises the steps of firstly carrying out equalization and whitening to improve the robustness of initial conditions, carrying out multi-constraint limit on a non-negative matrix factorization algorithm and carrying out optimization on an objective function and further improving the signal separating property through a feedback mechanism finally. Therefore, the method disclosed by the invention has the advantages of good factor interpretability and high separated signal purity.

The present invention relates to a mixing matrix estimation method aiming at underdetermined blind source separation. The method comprises: respectively performing short-time fourier transform on two paths of received observation signals to obtain short-time fourier coefficients X1 (t, f) and X2 (t, f) of two paths of observation signals, forming a plurality of scattered points and removing the low-energy scattered points; obtaining ratios of the remained scattered points, classifying the scattered points according to the ratios to obtain an M class with the largest number of scattered points, calculating an average value of the ratios of the scattered points in each class, then transforming each average value into a slope angle of a straight line, converting each slope angle into a column vector, obtaining a rotation matrix T according to the column vectors of which the directions are approximate, and performing rotation conversion to obtain two paths of new observation signals X'1 (t, f) and X'2 (t, f); and aiming at the two paths of observation signals X'1 (t, f) and X'2 (t, f) and according to a ratio of two paths of observation signals, classifying the scattered points to obtain an S class with a largest number of scattered points, calculating an average value of the ratios of the scattered points in each class, then obtaining an approximate slope angle, obtaining corresponding column vectors by the slope angles and finally, obtaining an integral mixing matrix.

The invention discloses a method for restoring a source signal in undetermined underdetermined blind source separation based on a single-layer perceptron. According to the method for restoring the source signal in the undetermined underdetermined blind source separation based on the single-layer perceptron, an observation signal matrix is first obtained; afterwards, the clustering is carried out on all column vectors in the observation signal matrix to obtain a hybrid matrix; the column vector of a to-be-restored source signal is calculated according to the observation signal matrix and the hybrid matrix; the column vector of the to-be-restored source signal is updated by utilizing a single-layer perceptron method; a restored source signal is finally obtained. By using the method for restoring the source signal in the undetermined underdetermined blind source separation based on the single-layer perceptron, the defects that the restoration precision of the source signal is easily influenced by a noise error and the calculation complexity is slightly high, existing in the prior art, are overcome; the method for restoring the source signal in the undetermined underdetermined blind source separation based on the single-layer perceptron is enabled to have the advantages of quickly restoring the source signal while the higher restoration precision can be maintained.

Provided is an underdetermined blind source separation RFID anti-collision method based on frame slots. The invention provides a parallelism recognition anti-collision algorithm based on the constraint NMF and the frame-slot ALOHA (FSA) in order to solve the problems that when the number of labels recognized by an RFID system is largely increased, the separation effect of an underdetermined blind source separation algorithm becomes poor, even source label signals can not be separated out, and the whole performance of the system declines rapidly. By analyzing the constraint NMF algorithm steps in the RFID system, building an underdetermined blind source separation antenna system model and selecting the reasonable number of frame slots, the number of labels in each slot is controlled within the optimal range, the source signal separation effect and the throughput capacity of the RFID system are still kept in the optimal state even if the number of the labels is increased largely, and therefore the system performance is greatly improved, the recognition time is short, and the underdetermined blind source separation RFID anti-collision method based on the frame slots has very good application value in the field of a large number of targets and need of high efficiency.

The invention discloses an underdetermined blind source separation method based on density, mainly solving the problems that the computation complexity of the prior art is high, the prior art is easily influenced by an initial value and the number of source signals needs to be given. The method comprises the following steps: after the low energy sampling data of an observed signal is removed, projecting the observed signal onto a right half unit hypersphere; computing the density parameters of all projective points, and deleting the projective points with smaller density; utilizing an improved K-mean value clustering algorithm to cluster the surplus projective points and determining the optimal clustering number and clustering center; removing the cluster with less number of data objects, wherein the number of the surplus clusters is the estimated value of the number of source signals, and the corresponding clustering center is the estimated value of each column vector of a confusion matrix; and according to the observed signal and the estimated confusion matrix, adopting a linear programming method to recover the source signals. According to the method, the computation complexity is reduced, the influence of the initial value on the estimated performance is reduced, the confusion matrix can be estimated when the number of the source signals is unknown, and the estimating precision of the confusion matrix and the source signals can be increased.

The invention discloses a two-channel optical three-dimensional interference method based on underdetermined blind source separation. On the basis of interference data collected by a CCD camera, interference signals between surfaces of tested pieces of glass slides are processed by blind separation and thus interference signal parameters are calculated, wherein the interference signal parameters include an interference signal amplitude frequency and an interference signal phase frequency; according to a two-channel optical three-dimensional Michelson interference experiment, estimation of a mixed matrix is obtained based on a K-means clustering algorithm and source signal recovery is realized by an L1 norm shortest path method; and then laser wave number scanning is performed and four-surface interference signals are obtained by blind separation accurately by using light intensity values collected by the CCD camera, thereby realizing blind separation of the four-surface interference signals. Compared with the ordinary interference measurement method not being capable of identifying all peak locations accurately, the disclosed two-channel optical three-dimensional interference method has the following advantages: with two-channel detection and sampling, blind source separation is performed and locations of all interference signals can be detected even when the signals face the strong noises.

The invention discloses an underdetermined blind source separation method, system and device based on minimization and maximization. The method comprises the steps of converting a time-domain instantaneous mixed signal into a frequency-domain instantaneous mixed signal on each frequency band through short-time Fourier transform, extracting a real part vector and an imaginary part vector of a complex value hybrid matrix based on the sparsity of a source signal, performing single source point judgment and extraction through cosine angles, performing normalization and modular arithmetic on singlesource points, and achieving hybrid matrix estimation through hierarchical clustering of the single source points; and converting a source signal recovery problem into an l1 norm optimization problemby utilizing the hybrid matrix and the sparsity hypothesis of the source signal, and converting the source signal of the time-frequency domain back to the time domain by adopting short-time inverse Fourier transform after the source signal is recovered in the time-frequency domain so as to realize underdetermined blind source separation. The non-convex penalty function is constructed, so that theoptimization of the objective function is close to the optimization of the l0 norm, the systematic underestimation of the conventional l1 norm regularization and the source signal recovered accordingto the sparsity hypothesis are avoided, and the convexity of the objective function is ensured.

The invention discloses an analog circuit early fault feature extraction method based on underdetermined blind source separation. The method comprises the following concrete steps that (1) different test nodes are selected for different test circuits and the corresponding test excitation is inputted, and the circuit test point signals are acquired to form an observation signal matrix X(t); (2) theobservation signals X(t) of the time domain are converted into time-frequency domain signals by using short-time Fourier transform; (3) a hybrid matrix A is estimated through the fuzzy C-means clustering algorithm in the time-frequency domain; (4) the estimation values of the source signals in the time-frequency domain are obtained by using the hybrid matrix A obtained in the step three throughthe weighted minimum L1 norm method and then transformed to the time domain so as to obtain the recovered source signals S(t); and (5) the kurtosis of each source signal is calculated for the source signals S(t) obtained in the step four so as to form the feature vectors. The analog circuit early fault features can be effectively and accurately extracted so that the identifiability of the early fault features can be enhanced and the method plays an important role in later circuit fault diagnosis.

The invention discloses a gradient descent method and Newton method based underdetermined blind source separation source signal recovery method. The method comprises the steps of firstly, obtaining an observation signal matrix; secondly, clustering all column vectors in the observation signal matrix to obtain a hybrid matrix; thirdly, calculating to-be-recovered source signal column vectors according to the observation signal matrix and the hybrid matrix; and finally, updating the to-be-recovered source signal column vectors by utilizing a gradient descent method, a Newton method and a projection method to obtain a recovered source signal. According to the method, the defects that the source signal recovery precision is easily influenced by a noise error and the computing complexity is high in the prior art are overcome, so that the method has the advantage that the source signal can be quickly recovered while relatively high recovery precision is kept.

The invention relates to the field of signal identification, in particular relates to the field of blind source / signal separation and specifically relates to an underdetermined blind identification method based on general covariance and tensor decomposition. The underdetermined blind identification method based on general covariance and tensor decomposition comprises establishing kernel functional equation sets according to the general covariance matrix of observation hybrid sampled data, stacking the kernel functional equation sets into a three-dimensional tensor model, and finally gaining a factor matrix by virtue of tensor Tucker decomposition, thereby identifying underdetermined hybrid system characteristics or hybrid matrixes. The underdetermined blind identification method based on general covariance and tensor decomposition has the technical advantages of effective performance improvement and relatively low computation complexity in contrast with a mixed matrix identification method in traditional underdetermined blind source separation; the underdetermined blind identification method provides a technical foundation for the blind anti-jamming technique of array signal processing and satellite communication.

The invention belongs to the technical field of blind signal separation, and particularly relates to a time-frequency single source point extraction method in underdetermined blind source separation based on compound angle detection. The time-frequency single source point extraction method comprises the steps that a source signal through instantaneous mixing, i.e. an observation signal, is acquired from a receiving sensor; the influence of noise is neglected, and spatial time-frequency distribution of the observation signal is calculated; the compound angle of each sensor receiving signal in the time-frequency domain is calculated; the difference value of the arctangent function of the compound angles of two sensor receiving signal is calculated; and the time-frequency points in a time-frequency single source point set are taken, and noise is removed by a self-adaptive hierarchical clustering method. According to the method, the requirement for the sparsity of the source signal can be reduced, and the extraction precision of the time-frequency single source points can be enhanced so that the extraction problem of the time-frequency single source points in underdetermined blind source separation under the condition of mixing in the time-frequency domain of the source signal can be solved by the method.

The invention discloses a gear vibration signal source underdetermined blind source separation method based on density and compressed sensing, and the method comprises the steps: firstly carrying outthe noise reduction preprocessing of a collected signal through employing a wavelet noise reduction method; secondly, carrying out short-time Fourier transform on the mixed signals, and converting a convolution mixed model in a time domain into a linear mixed model in each frequency band; then, achieving effective estimation of a hybrid matrix through a single-source-point extraction method basedon sparse coding and a vibration source number recognition method based on a density peak clustering method; recovering the source signal through a compressed sensing method to obtain a separation signal; and finally, performing sequence and amplitude correction on the separation signals of each frequency band, and converting the separation signals from a frequency domain to a time domain. According to the method, the robustness to noise is enhanced by estimating the number of the vibration sources through the density peak clustering method, the probability of generating false peaks is reduced, then the estimation error of the number of the vibration sources is reduced, and the separation work of mixed signals can be effectively completed.

The invention discloses an underdetermined voice blind source separation method, which comprises the following steps of acquiring an observation signal, processing the acquired observation signal, andacquiring a voice source signal sparse characteristic; obtaining aliasing matrix estimation by using an ant colony K-means clustering algorithm according to the obtained sparse features (representedby a scatter diagram) of the voice source signal; and recovering the voice source signal by using a greedy sparsification method according to the estimated aliasing matrix and the processed observation signal. According to the underdetermined voice blind source separation method, the underdetermined blind source separation problem under sparse representation of two different types of voice signalscan be solved by using the improved two-step method, and the underdetermined voice blind source separation method is suitable for source signal recovery of more than two observation signals.

The invention provides an underdetermined blind source separation method and device based on structured sparse subspace clustering, and the method comprises the steps of converting a time domain signal to a time-frequency domain through short-time Fourier transform, and screening a single source point with a high energy value through a time-frequency interval windowing method based on the short-time stability of the signal; then, adopting a combined hybrid matrix estimation method of structured sparse subspace clustering and an improved potential function, where the structured sparse subspace clustering can determine the number of source signals and obtain data classification, and the improved potential function method can accurately determine a hybrid matrix vector and estimate a hybrid matrix; and then, reconstructing a source signal by adopting a space projection method based on source number estimation in combination with a minimum l1 norm method, and finally, recovering a time-frequency domain reconstruction signal to a time domain by adopting short-time inverse Fourier transform. According to the invention, the number of the source signals can be visually determined, high precision and high robustness are achieved, and the source signals can be accurately reconstructed from the observation signals.

The invention belongs to the radar detection technology field and discloses an underdetermined blind source separation source signal recovering method based on an improved shortest path method. The method is aimed at m observation signals and comprises the following steps: taking adjacent two observation signals each time referred to as the ith observation signal and (i+1)th observation signal wherein i represents 1, 2,..., and m-1 with the results that for each treatment, the observation signals are the combinations of two adjacent observation signals and that the number of combinations totals C(m-1)1; then, using the selected two channels of signals as observation signals; using the shortest path method to recover the corresponding source signals with the result that the recovered source signals are of C(m-1)1 groups; and summing the C(m-1) combinations of the source signals to seek their average value to obtain the to-be-recovered source signal. According to the invention, the method is suitable for the source signal recovery of more than two observation signals and the recovered and obtained m-1 combinations of source signals, the method deals with how to seek the average to obtain the final source signals.

The invention belongs to the technical field of radar reconnaissance and discloses an improved shortest path underdetermined source signal recovery method. Under the condition that the observed signalis known, Under the condition of the estimated mixing matrix, the problem of restoring the source signal according to the sparse component analysis is transformed into solving the following optimization problem: when the number of the observed signals is greater than or equal to 2, the first and m-th observed signals are combined into a two-dimensional observation signal combination according tothe steps 1 to 10; For each combination, the corresponding source signal is recovered by using the classical shortest path method, and m combinations of two-dimensional observation signals are obtained. A new matrix is formed by separating the signals from m two-dimensional observation signals, The estimation of n source signals is the source signal to be recovered. The invention can process radarsignal, communication signal, biomedical signal and the like, and realizes the recovery of underdetermined blind source separation source signal under the condition that the mixing matrix has been estimated.

The invention discloses a dual-channel optical three-dimensional interference method based on underdetermined blind source separation. Through the interference data collected by a CCD camera, the interference signals between the surfaces of the slide glass under test are blindly separated, and the parameters of the interference signals are solved, including : Interference signal amplitude frequency and interference signal phase frequency; based on the dual-channel optical three-dimensional Michelson interference experiment, the estimation of the mixing matrix is ​​obtained through the K-means clustering algorithm and the recovery of the source signal is realized through the L1 norm shortest path method; finally, the laser The wave number scanning accurately blindly separates the four-surface interference signals through the light intensity values ​​collected by the CCD camera, and realizes the blind separation of the four-surface interference signals; compared with the ordinary interferometric method that cannot accurately identify each peak position, the present invention uses a dual-channel Detection and sampling, through blind source separation, enables the signal to detect the location of each interference signal even in the face of relatively strong noise.

The invention discloses an underdetermined blind source separation method based on AP clustering, and belongs to the field of signal processing, and the method comprises the following steps: (S1) converting an observation signal to a time-frequency domain, so as to screen out a time-frequency single source point; (S2) sampling the time-frequency single source points, and clustering the sampled time-frequency single source points by using an AP clustering algorithm to obtain the number of signal sources and an initial clustering center; (S3) taking the number of the signal sources, the initial clustering center and the time-frequency single source points obtained in the step (S1) as input of a preset target clustering algorithm, correcting the initial clustering center to obtain a corrected clustering center, and estimating a hybrid matrix by taking the corrected clustering center as a column vector of the hybrid matrix; (S4) reconstructing a signal source by using the hybrid matrix, wherein the input parameters of the target clustering algorithm comprise a clustering number and an initial clustering center. According to the invention, the estimation precision of the hybrid matrix in the underdetermined blind source separation system can be improved, so that accurate separation of source signals is realized.

The invention belongs to the technical field of array signal processing and discloses an improved underdetermined blind source separation source number estimation method based on wavelet analysis. The method is characterized by carrying out decomposition on observation signals through wavelet transform; carrying out single-layer reconstruction on wavelet coefficients obtained through decomposition; reconstructing an observation signal matrix and constructing a covariance matrix of the matrix; calculating characteristic values of the matrix; and calculating source number. The method can enable whitened observation signals and a part of wavelet decomposition coefficient reconstruction signals thereof to be combined into new multidimensional observation signals, can accurately carry out estimation under the condition of large signal-to-noise ratio change range of mixed signals to obtain the number of source signals, and is simple in principle and high in operability.

The invention relates to a mixing matrix estimation method for underdetermined blind source separation, which performs short-time Fourier transform on two received observation signals to obtain the short-time Fourier coefficients X1(t,f) of the two observation signals ) and X2(t,f) to form multiple scattered points and remove low-energy scattered points; calculate the ratio of the remaining scattered points, classify the scattered points according to the ratio, and obtain the M category with the largest number of scattered points, and calculate each Then convert the mean value into the slope angle of the straight line, and then convert it into a column vector, and obtain the rotation matrix T according to the column vectors with close directions, and perform rotation transformation to obtain a new two-way observation signal X′1( t,f) and X′2(t,f); for the two observation signals X′1(t,f) and X′2(t,f), classify the scattered points again according to the ratio of the two observation signals , get the S class with the largest number of scattered points, calculate the mean value of the ratio of scattered points in each class, and then get the similar slope angles, get the corresponding column vectors through these slope angles, and finally get the entire mixing matrix.

The invention discloses a curvelet expression-based method for image underdetermined blind source separation. The method comprises the following steps: firstly, performing underdetermined blind sourceseparation under a sparse condition to acquire pre-separated signals in the same number as source signals; secondly, using the pre-separated signals as mixed signals to change an underdetermined state into an even-determined state; finally, adopting the conventional FastICA method to perform even-determined blind source separation to acquire final separated signals to realize the effect of the underdetermined image blind source separation. Aiming at the underdetermined blind source separation problem, the method provides a separation method based on underdetermined and even-determined combination, makes full use of a Curvelet domain coefficient property to complete initial underdetermined blind source separation, creates the mixed signals for conversion into even-determined blind source separation to adopt the FastICA to perform even-determined separation of the pre-separated signals at the same time, thereby improving the image blind source separation accuracy and improving image separation effect. The method has significant application potential in radio communication systems, sonar and radar systems, and audio, acoustic and medical signal processing in military and nonmilitaryfields.

The invention relates to a wind power tower drum vibration signal underdetermined blind source separation method for solving an unknown source number, and the method comprises the steps of firstly converting a mixed signal into a time-frequency domain, and converting the mixed signal into a multi-dimensional row vector; screening the multiple pieces of time-frequency point data through a principal component analysis method and a mixed time-frequency ratio method based on the multi-dimensional row vector to obtain single-source domain feature data meeting a preset screening formula, and performing normalization processing on the single-source domain feature data to obtain feature data represented based on a direction angle; clustering through a preset modified clustering algorithm to obtain a signal source number and an estimation hybrid matrix; and finally, obtaining a source signal through sparse optimization based on the signal source number and the estimation hybrid matrix. Therefore, when the single-source domain feature data is screened, the data of the multiple time frequency points are screened through the principal component analysis method and the mixed time frequency ratio method, the information amount is increased, the screening accuracy is improved, clustering is carried out by modifying the aggregation algorithm, the number of vibration sources and the frequency components of all the sources are accurately recognized, and the data analysis accuracy is improved.