47 results about "Separation matrix" patented technology

The invention belongs to the technical field of environment monitoring, and discloses an ecological environment anomaly detection and early warning method based on artificial intelligence. The emission characteristic fingerprint database is constructed according to enterprise production process characteristics by acquiring multi-point pollutant concentration, meteorological parameters and enterprise space distribution data in a park. By analyzing the space-time diffusion trajectory of pollutants, identifying abnormal areas and calculating reverse traceability parameters, a space-time superposition inversion model of multi-source emission is constructed. A multi-enterprise contribution degree separation matrix is generated based on fingerprint matching degree weighted calculation, and accurate separation of emission contributions of all enterprises in the mixed pollution field is achieved. By monitoring pollutant concentration fluctuation characteristics in real time, contribution degree weight distribution is dynamically adjusted, and the accuracy of a monitoring result is ensured. And accurately determining a responsibility subject with excessive emission and generating an early warning signal. According to the invention, accurate identification and responsibility determination of the pollution source can be realized in a complex multi-source emission environment, and the accuracy and fairness of industrial park environment supervision are effectively improved.

The invention relates to an MIMO system modulation type identification method and system under communication interference, and belongs to the technical field of signal identification, and the method comprises the steps: carrying out the filtering, denoising and whitening preprocessing of a received MIMO signal; a five-dimensional fourth-order mutual cumulant tensor is constructed by using the preprocessed signal, and a separation matrix is estimated through parallel factorization, so that suppressing interference reconstruction separation is realized; constructing a three-dimensional tensor again for the separated and suppressed signal, and performing dimensionality reduction and feature extraction on the tensor by using a generalized tensor discriminant analysis algorithm; and performing Tucker decomposition on the tensor after dimension reduction to establish a kernel tensor and a factor matrix, and identifying the similarity of the kernel tensors of different modulation types according to the Euclidean distance criterion to complete MIMO signal modulation type identification. Simulation results show that the method has good recognition accuracy under the condition of low signal to interference ratio, and prior information such as channel coefficients and noise distribution is not needed.

The invention discloses a voice signal blind source separation method, device and equipment and a storage medium, and relates to the technical field of voice separation, and the method comprises the steps: collecting a mixed sound signal of a plurality of source signals through a plurality of sound collection devices, and obtaining a plurality of time domain observation signals; converting the time domain observation signal into a frequency domain observation signal, and creating a frequency domain separation signal model based on the frequency domain observation signal and a preset separation matrix; obtaining the direction of arrival of the source signal to obtain direction of arrival information, and performing iterative updating on a preset separation matrix in the frequency domain separation signal model based on an independent vector analysis algorithm of an auxiliary function to perform blind source separation on the frequency domain observation signal to obtain a frequency domain separation signal; the geometric constraint of the iterative update of the preset separation matrix is the constraint created based on the direction of arrival information; and converting the frequency domain separation signal into a time domain separation signal. According to the invention, the efficiency of voice signal blind source separation can be improved, and the accuracy of blind source separation is remarkably improved.

The invention relates to the technical field of signal processing, in particular to a mixed partial discharge signal separation method which comprises the following steps: S1, a linear mixed mathematical model is selected to describe the mixing process of partial discharge signals, and a to-be-detected mixed signal x (t) is obtained; s2, performing whitening preprocessing on a mixed signal x (t) to be detected to obtain # imgabs0 # S3, constructing a fourth-order cumulant matrix # imgabs2 # of # imgabs1 # S4, performing joint diagonalization processing on the fourth-order cumulant matrix # imgabs3 # to obtain a unitary matrix U; s5, obtaining a separation matrix W: W = UTQ according to the unitary matrix U and the whitening matrix Q; s6, separating each source signal from the to-be-detected mixed signal x (t) by using the separation matrix W; and S7, correcting the output signal vector of each separated source signal. According to the invention, accurate separation of local mixed discharge signals can be realized.

A method for extracting effective information of a communication signal in a low signal-to-noise ratio complex environment relates to the field of communication signal processing, and comprises the following steps: firstly, obtaining a plurality of intrinsic mode functions mixed with effective signals with different frequencies and complex noise interference by using an adaptive noise complete empirical mode decomposition method, and carrying out wavelet denoising and reconstruction on the intrinsic mode functions to obtain a plurality of intrinsic mode functions; the method comprises the following steps of: obtaining a de-noised signal with a high signal-to-noise ratio, pre-processing the de-noised signal, calculating a separation matrix of an observation signal matrix after pre-processing according to a blind source separation algorithm, and extracting an effective communication signal from the original observation signal by utilizing the separation matrix so as to complete the effective information extraction of the communication signal. According to the method, noise and invalid signal interference in the signal transmission process can be suppressed under the conditions of strong environmental noise and low signal-to-noise ratio, signals containing effective information can be accurately extracted, and the target detection efficiency can be improved under the condition of high signal-to-noise ratio.

This invention discloses an online speech separation method and apparatus based on the sparsity constraints of the speech spectrum domain, belonging to the field of information processing. The method includes the following steps: S1: Perform a short-time Fourier transform on the time-domain signal of each microphone to obtain a time-frequency domain representation, and construct the original frequency domain vector; S2: Construct an initialization separation matrix of dimension 2×M according to the direction of the target speech relative to the microphone array; S3: Construct an optimization function and calculate the separation matrix used to separate the target speech from the remaining signal; S4: Obtain the frequency domain signal of the target speech based on the calculated separation matrix, and then obtain the time domain signal of the target speech. This invention, based on the sparsity characteristics of speech signals in the spectrum domain, designs a new optimization function that does not require pre-whitening processing, ensuring the real-time performance of conference communication.

The invention discloses a graph signal blind source separation method in a white noise environment, and belongs to the technical field of signal processing. Aiming at the problem that the traditional blind source separation performance is reduced due to additive white noise, the invention provides a joint optimization scheme combining blind compression nonlinear noise suppression and image smoothing regularization. Firstly, a blind compression function is applied to a noisy observation image signal for preprocessing; secondly, carrying out mean value removal and whitening processing on the compressed signal; then, constructing a graph Laplacian matrix as a smoothing operator, establishing a joint diagonalization objective function fusing graph autocorrelation, a FastICA item and a graph smoothing regular item, and adopting a Givens rotation algorithm to iteratively optimize and solve a separation matrix; and finally, reconstructing a source signal through multiplication of the separation matrix and the whitening data. According to the method, noise interference is effectively suppressed through blind compression, signal structure consistency is maintained by using graph smoothing prior, and separation robustness, precision and convergence stability in a strong noise environment are remarkably improved.

The invention discloses a double-constraint second-order blind identification bridge dynamic deflection signal noise reduction method, which comprises the following steps: acquiring dynamic deflection observation signals of bridge target points, and constructing an observation signal matrix by using the observation signals of at least three adjacent target points; carrying out centralization processing on the observation signal matrix, and carrying out whitening transformation to obtain a whitening signal with a covariance matrix as a unit matrix; selecting a group of non-zero time delay values, calculating a time delay covariance matrix of the whitening signal under each time delay, and performing joint approximate diagonalization on all the time delay covariance matrixes to obtain a separation matrix and an initial separation source signal; optimizing the separation matrix based on sparsity constraint to obtain a hybrid matrix; performing dominant frequency optimization on the initial separation source signal; and reconstructing based on the optimized hybrid matrix and the optimized main frequency source signal to obtain a de-noised bridge dynamic deflection signal. The hybrid matrix optimization based on the sparsity constraint enables the reconstruction result to be closer to the real bridge dynamic deformation.

The application relates to the field of radar technology and provides a method and device for resisting towed interference of an airborne radar. The method comprises the following steps: constructing sum channel data, azimuth difference channel data and elevation difference azimuth data of echo signals; constructing sum-difference three-channel data of each pulse according to the sum channel data, the azimuth difference channel data and the elevation difference azimuth data; processing the sum-difference three-channel data by using a blind source separation algorithm to obtain target echo channel data and a separation matrix; pulse compressing the target echo channel data to obtain a target distance estimation value; determining an azimuth angle and an elevation angle of a target according to the target echo channel data and the separation matrix; and constructing a target echo matrix by using target echo channel data of each pulse, and performing Doppler processing on the target echo matrix to obtain a target speed estimation value. The application can resist towed interference, determine the distance, the azimuth angle, the elevation angle and the speed of the target, and realize effective detection of the target.

A mixed electric signal disassembling method and system based on residential electricity load, characterized in that the method comprises the following steps: step 1, collecting the mixed electric signal of the residential electricity node in the power distribution network through the load monitoring device, and adding random noise to the mixed electric signal; step 2, extracting signal fluctuation trend information from the mixed electric signal with random noise added, and obtaining a pre-disassembled signal; step 3, determining whether the pre-disassembled signal meets the general index of alternating current signal, thereby obtaining a first disassembled signal; step 4, sequentially extracting a second disassembled signal and subsequent disassembled signals by using the methods in steps 2 and 3; step 5, extracting first and second characteristic values, and obtaining effective disassembled signals; step 6, using a characteristic separation matrix to segment each of the effective disassembled signals to generate a plurality of single load electric signals, thereby completing the disassembly of the mixed electric signal.

The application provides a few-channel switching direction finding method based on blind source separation, and belongs to the technical field of emergency search and rescue direction finding. The method comprises the following steps: constructing a corresponding original signal matrix of each array element switching, and performing pretreatment to obtain a corresponding standard signal matrix; performing blind source separation on each standard signal matrix to obtain a corresponding separation matrix and a separation signal matrix; constructing a corresponding array manifold matrix of each array element switching, and calculating all waveform correlation coefficients; using all waveform correlation coefficients, completing signal alignment of all array manifold matrices, and using a reference matrix to obtain a calibration array manifold matrix corresponding to each array element switching; and using all calibration array manifold matrices to perform direction of arrival estimation on each rescue target. The application can effectively estimate the noise of a few-channel direction finding system, and improve the direction finding precision, angle resolution capability and robustness of the few-channel direction finding system.

The invention provides a blind noise reduction method for MEMS multi-sensor self-contrast learning, which belongs to the technical field of MEMS, and comprises the following steps: receiving observation data of MEMS multi-sensors, and constructing a whitening matrix to obtain whitening data; initializing a separation matrix by using a random unit vector, solving an optimal separation vector, and outputting a normalized signal source and a noise source after orthogonalization processing; enabling a single observation signal, a signal source and a noise source to share encoder weight extraction features, and inputting a decoder to reconstruct a signal; and calculating the total loss, and outputting a high-precision noise reduction signal when the total loss is minimum. According to the blind noise reduction method for MEMS multi-sensor self-contrast learning, the difficulty of non-Gaussian modeling for unknown environment interference in a traditional method is solved, and the problem of amplitude uncertainty of signal blind separation under the state that a physical quantity to be measured does not have prior information is solved.

The application discloses a kind of sensor fault detection method and device of structural health monitoring system, it is related to structural health monitoring technical field, comprising: obtaining the nonlinear structure monitoring data of sensor system collected by structural health monitoring system, nonlinear structure monitoring data is handled using improved geometric PNL hybrid model, obtain linear to-be-separated mixed signal, using FastICA model to process to-be-separated mixed signal, obtain multiple independent elements, determine the separation matrix of improved geometric post-nonlinear independent component analysis model according to multiple independent elements, using the processing of improved geometric post-nonlinear independent component analysis to real-time collection nonlinear structure monitoring data, determine whether sensor system exists fault and the sensor of fault occurrence.The method can still complete linearization processing to sensor mixed signal under the condition that prior knowledge is unknown, compared with simple linear ICA analysis algorithm, it is more suitable for complex nonlinear structure.

The present invention discloses an underwater one-dimensional direction of arrival (DOA) estimation method based on a cascaded neural network. The cascaded neural network constructed by this method includes a sound speed classifier and a sound speed regressor. The sound speed classifier decomposes the input features into different sound speed domains, and then controls the operation of the sound speed regressor to obtain the sound speed estimation value. The multi-source signals are separated through blind source separation to obtain the separated signals and the separation matrix. After reconstructing the separated signals, feature extraction is performed, and then the cascaded neural network is input to obtain the sound speed estimation value. According to the relationship between the array manifold matrix and the separation matrix, the array manifold estimation matrix is obtained, and then the DOA estimation expression is obtained according to its characteristics. The estimated sound speed is substituted into the expression to obtain the DOA estimation value under the determined sound speed. The present invention does not require obtaining the sound speed pre-estimation value through an additional array, and all sensors can be used for DOA estimation, improving the DOA estimation accuracy and reducing the hardware cost.

This invention belongs to the field of environmental monitoring technology. It discloses an artificial intelligence-based method for detecting and warning of ecological and environmental anomalies. By acquiring data on pollutant concentrations, meteorological parameters, and spatial distribution of enterprises at multiple locations within an industrial park, an emission characteristic fingerprint database is constructed based on the production process characteristics of each enterprise. By analyzing the spatiotemporal diffusion trajectory of pollutants, abnormal areas are identified, and reverse source tracing parameters are calculated to construct a spatiotemporal superposition inversion model for multi-source emissions. A multi-enterprise contribution separation matrix is ​​generated based on fingerprint matching degree weighted calculation, achieving accurate separation of emission contributions from each enterprise in a mixed pollution field. By monitoring pollutant concentration fluctuation characteristics in real time and dynamically adjusting the contribution weight allocation, the accuracy of monitoring results is ensured. The responsible parties for excessive emissions are accurately identified, and early warning signals are generated. This invention can achieve accurate identification of pollution sources and determination of responsibility in complex multi-source emission environments, effectively improving the accuracy and impartiality of environmental supervision in industrial parks.

The invention relates to a superparamagnetic effect identification and separation method in time domain electromagnetic response. The method comprises the following steps: establishing a double-feature superparamagnetic effect identification model based on a fuzzy support vector machine algorithm; fitting the induction-magnetization effect in power law attenuation in the late period by adopting a Levenberg-Marquardt algorithm to obtain a quasi-superparamagnetic effect, and performing de-equalization and whitening on the quasi-superparamagnetic effect and the total effect as input signals of an independent component analysis algorithm; and selecting a mutual information minimization criterion and a natural gradient algorithm to update and optimize the separation matrix according to the statistical characteristics of the transient electromagnetic signal until the convergence condition is met, and finally separating to obtain the induction response and the superparamagnetic effect. According to the method, the superparamagnetic effect can be effectively identified and accurately separated, the inversion interpretation precision of the time-domain electromagnetic method in superficial superparamagnetic layer exploration is effectively improved, and technical support is provided for the development of the time-domain electromagnetic method in refined detection.

Embodiments of the present application disclose a voice signal processing method and device and electronic equipment. The device comprises a dereverberation module and a blind source separation module; wherein: the dereverberation module is configured to utilize a prediction coefficient matrix G(n) of the n th moment to perform dereverberation processing on a delay signal group Y(n-D) corresponding to an input signal y(n) of the n th moment, to obtain a dereverberation actual signal x(n) of the n th moment; the blind source separation module is configured to utilize a separation matrix W(n) of the n th moment to process the dereverberation actual signal x(n) of the n th moment, to obtain a separation actual signal z(n) of the n th moment; wherein the device further comprises a calculation module configured to calculate the prediction coefficient matrix G(n) of the n th moment by the following method, comprising: obtaining a separation matrix W(n-1) of the (n-1) th moment; obtaining the prediction coefficient matrix G(n) of the n th moment according to the separation matrix W(n-1) of the (n-1) th moment and a prediction coefficient matrix G(n-1) of the (n-1) th moment.

To provide a technique capable of emphasizing a sound signal with high estimation accuracy even if there is a state where the number of allocated time frames is small.SOLUTION: A sound signal enhancement device for acquiring a signal in which a target sound is enhanced from an observation signal in which the target sound and noise are mixed includes a separation matrix application unit 3 for applying a plurality of separation matrices to the observation signal to acquire a plurality of estimation signals, and a switch unit 4 for acquiring the signal in which the target sound is enhanced by using any of the plurality of estimation signals. The state regularization weight is further used when the plurality of separation matrices are updated using the observation signals, and the state regularization weight is used to update the switch weight used by the switch.SELECTED DRAWING: Figure 1

The invention discloses an anti-interference radar signal extraction method and device and electronic equipment, and the method comprises the steps: carrying out the whitening processing of a to-be-processed radar signal through a zero-phase component analysis whitening algorithm, and obtaining a to-be-separated radar signal after the whitening processing. An initial separation matrix is obtained by performing iterative optimization processing on a preset target separation function for multiple times. And performing fine adjustment on the initial separation matrix according to a second-order blind source separation algorithm to obtain a target separation matrix. Performing signal separation processing on the radar signal and the interference signal according to the target separation matrix to obtain an initial radar signal, and performing denoising processing on the initial radar signal according to a target denoising algorithm to obtain a radar signal, the target denoising algorithm being any one of a wavelet transform denoising algorithm and an improved stationary wavelet transform denoising algorithm. The problems of high calculation complexity and low efficiency in the prior art are avoided, and the stability and accuracy of radar signal extraction are improved.

The invention discloses a blind source separation transfer learning monitoring method, a power grid system, a blind source separation transfer learning monitoring device and a storage medium, and the method comprises the steps: obtaining base current data to construct a training data set, firstly controlling each circuit breaker to be conducted for steady-state operation, sequentially controlling the on-off action of each circuit breaker to obtain current waveform base signals collected by a current transformer in each on-off process to construct a training data set; obtaining an initial separation matrix by using the training data set and the branch data set, and establishing a blind source separation model; iteratively executing a transfer learning step for multiple times: acquiring a real-time main circuit current; obtaining an estimated current waveform signal of each execution branch based on a blind source separation model according to the real-time main circuit current; obtaining the on-off action of each execution branch, and analyzing the variation of the separation matrix; iteratively calculating a target separation matrix, and updating a blind source separation model at the next moment by using the target separation matrix; according to the design, the monitoring structure is simplified, the operation cost is reduced, and the diagnosis accuracy is optimized.

The application discloses a kind of multi-channel data synchronous acquisition and optimization processing method, it is related to data processing technical field, comprising: through synchronous trigger signal control multi-channel sensor synchronous acquisition data, the multi-channel data collected is preprocessed, and multi-channel data matrix is generated;Initial separation matrix is constructed, based on mutual information value dynamic adjustment learning rate, iteration optimization separation matrix is carried out through negative entropy maximization criterion, realize the preliminary separation of multi-channel mixed signal, and the source signal of target frequency band is screened;Modal decomposition is carried out to target source signal, extracts extreme point and fits envelope line, iteration is screened modal component that satisfies preset condition, and effective modal component set is obtained;Modal component is classified as cross-channel common mode or channel exclusive mode based on cosine similarity, and according to signal-to-noise ratio, weight is distributed and weighted fusion, after evaluating the quality of fusion signal, output result, the signal-to-noise ratio and stability of fusion signal are significantly improved.

To separately collect a sound field radiated from a sound source in an area and a sound field coming from the outside of the area by using a distributed spherical microphone array.SOLUTION: A separation matrix calculation part 10 of a sound collection device 1 calculates a separation matrix in which position information and size are reflected and which is expressed by a spherical harmonic function on the basis of preset distributed spherical microphone array position information. The frequency analysis unit 11 obtains a time-frequency spectrum by performing Fourier transform on a sound collection signal from the distributed spherical microphone array M in a time domain. A spherical harmonic spectrum vector calculation part 12 calculates a spherical harmonic spectrum vector having a spherical harmonic spectrum of a sound field radiated from a sound source S in an area and a spherical harmonic spectrum of a sound field due to an interference wave coming from the outside of the area as elements on the basis of a separation matrix and a time frequency spectrum of a sound collection signal.SELECTED DRAWING: Figure 3

A convolution blind source separation method for identifying a gas turbine related noise source is disclosed, in which an observation signal of a gas turbine noise is collected, and a corresponding time-frequency domain complex value signal X is obtained by performing a synchronous compression transformation on the observation signal; an instantaneous blind source separation is performed on the time-frequency domain complex value signal X at each frequency point, a complex value normalized boundary target function is established, and an iteration optimization is performed at each frequency point based on a sub-gradient optimization method, a complex value separation matrix and a separation signal are obtained by performing a plurality of blind extraction estimations, a power spectrum spectral density distance method is used to perform a permutation calibration on the separation signal, an amplitude calibration is performed using a minimum distortion algorithm to obtain a time-frequency domain separation signal, and the time-frequency domain separation signal is restored to a time domain by using a synchronous compression inverse transformation to obtain a reconstructed time domain separation signal.

The invention provides a missing value-containing time sequence analysis method based on independent components, which belongs to the signal processing related field, and comprises the following steps: S1, obtaining a time sequence of multi-dimensional observation data, preprocessing the time sequence, and then stacking the time sequence into an observation data matrix; s2, decomposing the observation data matrix X, and solving to obtain a whitening component matrix Z; s3, a separation matrix is introduced, the whitening component matrix Z is further decomposed, and a time component and a space mode which are independent of each other are obtained; and S4, reconstructing a physical source signal of the time sequence of the observation data by adopting a time component and a spatial mode. The invention relates to an independent component-based missing value-containing time sequence analysis method, which is dedicated to non-interpolation processing of an incomplete observation data time sequence, does not need to interpolate a missing value in advance, and can effectively improve the precision of physical signal separation and extraction.

The present application relates to the technical field of industrial pipe network leakage detection, and discloses a printing and dyeing industry water pipe network leakage rapid identification and positioning system, comprising: a sensing module for collecting multi-dimensional data of the printing and dyeing workshop water pipe network; a processing and construction module for obtaining vibration conduction coefficients and pseudo-signal characteristic ranges; an extraction module for blind source separation of the pre-processed pressure-acoustic wave mixed signal; a verification module for verifying the extracted features in combination with a preset threshold, separation confidence and working condition data; and a calculation output module for calculating leakage point coordinates based on the propagation time difference of the real leakage signal. Through multi-dimensional data preprocessing and the construction of a multi-body dynamics model containing pipe network aging factors, the blind source separation of the pressure-acoustic wave mixed signal is realized by constraining the separation matrix with the pseudo-signal characteristic range, thereby realizing the efficient separation of the pseudo-signal and the real leakage signal generated by the dynamic deformation of the printing and dyeing workshop pipe network.

The invention relates to the technical field of radars, and provides a method and a device for resisting pull-type interference of an airborne radar. The method comprises the following steps: constructing sum channel data, azimuth difference channel data and pitch difference azimuth data of echo signals; constructing sum-difference three-channel data of each pulse according to the sum channel data, the azimuth difference channel data and the pitch difference azimuth data; processing the sum-difference three-channel data by using a blind source separation algorithm to obtain target echo channel data and a separation matrix; performing pulse compression on the target echo channel data to obtain a target distance estimation value; determining an azimuth angle and a pitch angle of the target according to the target echo channel data and the separation matrix; and constructing a target echo matrix by using the target echo channel data of each pulse, and performing Doppler processing on the target echo matrix to obtain a target speed estimation value. The method can resist pull-type interference, determines the distance, azimuth angle, pitch angle and speed of the target, and achieves the effective detection of the target.

The invention discloses a multi-channel data synchronous acquisition and optimization processing method, and relates to the technical field of data processing, and the method comprises the steps: controlling a multi-channel sensor to synchronously acquire data through a synchronous trigger signal, carrying out the preprocessing of the acquired multi-channel data, and generating a multi-channel data matrix; constructing an initial separation matrix, dynamically adjusting a learning rate based on a mutual information value, iteratively optimizing the separation matrix through a negentropy maximization criterion, realizing initial separation of multi-channel mixed signals, and screening source signals of a target frequency band; performing modal decomposition on the target source signal, extracting an extreme point and fitting an envelope line, and iteratively screening modal components meeting a preset condition to obtain an effective modal component set; the modal components are classified into cross-channel common modals or channel exclusive modals based on cosine similarity, weights are distributed according to the signal-to-noise ratio and weighted fusion is carried out, a result is output after the quality of the fusion signal is evaluated, and the signal-to-noise ratio and stability of the fusion signal are remarkably improved.

This invention discloses a method, apparatus, device, and medium for separating mixed signals. The method includes: acquiring the weights and encoding results of a mixed signal; performing feature aggregation on the encoding results to obtain aggregated features, and determining at least one similarity matrix between the aggregated features and the encoding results based on the aggregated features and the encoding results; weighting each of the similarity matrices using the weights to obtain at least one separation matrix; fusing the encoding results and each of the separation matrices, and decoding to obtain a separated clean signal. The embodiments of this invention can improve the mixed signal separation effect, reduce the dependence on signal type when separating mixed signals, and improve the versatility of the mixed signal separation method.

The present disclosure provides a sound source separation method and device, and relates to the technical field of intelligent vehicles, which comprises the following steps: performing sound source separation on a plurality of sound collection signals in a previous observation window according to a separation matrix corresponding to a previous round of sound source separation to obtain a plurality of separation estimation signals corresponding to the previous round of sound source separation, and obtaining a speech existence probability corresponding to the separation estimation signals; updating the separation matrix corresponding to the previous round of sound source separation according to the speech existence probability; and performing sound source separation on a plurality of sound collection signals in a current observation window according to the updated separation matrix to obtain a plurality of separation estimation signals corresponding to the current round of sound source separation. The speech existence probability of the separation estimation signals obtained based on the previous round of sound source separation is used to update the separation matrix, so that the speech existence information of the sound collection signals in the previous observation window is added when the separation matrix is updated, and the separation coefficients of various sound sources are adjusted accordingly, thereby enhancing the separation effect of the sound sources.

The invention discloses a feature extraction method and system based on an electrocardiogram signal. The method comprises the steps that S1, the electrocardiogram signal of an electrocardiogram image is collected; s2, pre-processing the electrocardiogram signal to obtain a pre-processed signal, and decomposing the pre-processed signal to obtain a preliminary decomposition feature of a signal waveform; s3, based on the preliminary decomposition characteristics, obtaining a separation matrix and reconstruction parameters of electrocardiogram wave-form components by using a waveform independence judgment model; s4, obtaining a waveform boundary positioning result based on the separation matrix and the reconstruction parameters; and S5, performing feature extraction on the electrocardiogram signal based on the waveform boundary positioning result to obtain electrocardiogram signal features.