42 results about "FastICA" patented technology

The invention relates to a mechanical vibration fault characteristic time domain blind extraction method, and belongs to the technical field of mechanical equipment status monitor and fault diagnosis. The mechanical vibration fault characteristic time domain blind extraction method includes: firstly, expanding a vibration observation signal into a high dimension signal subspace; then, obtaining a low dimension signal; afterwards, performing FastICA independent component analysis, calculating normalization kurtosis of all independent components, figuring out a component signal corresponding to the minimum normalization kurtosis, and using an orthogonal matching pursuit algorithm to reconstitute periodic signals; subsequently, removing the reconstituted periodic signal from each independent component, and then using an improved KL distance algorithm to calculate a distance matrix among the independent components after the periodic signals are removed from the independent components, and performing dynamic particle swarm clustering so as to obtain an estimation signal; finally, analyzing an envelope demodulation spectrum of the estimation signal, and performing fault diagnosis. The mechanical vibration fault characteristic time domain blind extraction method is suitable for processing a long convolution data problem, can effectively reduce influences from periodic ingredients on a blind separation result, and simultaneously can solve blind separation result order uncertainty problems, and finally achieves bearing fault characteristic extraction.

The invention discloses a pipeline positioning method. The pipeline positioning method comprises the steps that original signals are subjected to CEEMD; then noise-containing signals which are formedafter decomposition is accomplished are subjected to noise removal; cross-correlation calculation is conducted on the signals, and invalid redundancy components are removed according to the cross-correlation degree; and finally, the signals which are subjected to decomposition and noise removal are rebuilt, single leakage source signals are isolated through FastICA, and leakage points are positioned by use of a time difference positioning formula. The experiment result shows that the pipeline positioning method is suitable for detection and positioning of multi-point leakage of a metal pipeline, can improve the accuracy of pipeline leakage positioning to a large extent, and provides the powerful theoretical basis and the practical experience for practical engineering application.

The invention relates to a VMD and FastICA combined rolling bearing fault diagnosis method comprising the steps that the original vibration signals of a rolling bearing are acquired by using the present data acquisition equipment; VMD decomposition is performed on the acquired original vibration signals of the rolling bearing; the original vibration signals are decomposed into k modal components through VMD decomposition, and three continuous modal components act as a sequence combination to perform FastICA analysis so as to obtain reconstruction fault signals; and Hilbert transform is performed on the reconstruction fault signals to obtain the envelope spectrum of the reconstruction fault signals, the characteristic frequency, the modulation characteristic frequency and the side band under each fault characteristic are extracted from the obtained envelope spectrum and thus occurrence of the fault in the rolling bearing and the fault type are judged according to the extracted characteristic frequency component.

The invention discloses a high-precision denoising method for a local ultrasonic array signal, which comprises the following steps of: separating a multi-channel array signal into statistical independent components by using a fast independent component Fast ICA, carrying out autocorrelation analysis on the independent components, and preliminarily eliminating the influence of noise; Further denoising the signal of each channel by using an ensemble empirical mode decomposition (EEMD) algorithm; And recombining the signals subjected to fine denoising to obtain high-precision denoised signals. According to the method, the defects that traditional EMD decomposition can only denoise a single sensor and a single channel signal, and FastICA can only roughly denoise an array signal are overcome, too large direction finding and positioning errors caused by interference of on-site noise signals are avoided, and therefore the ultrasonic array signal denoising effect and the direction finding andpositioning precision are improved.

The invention relates to a mixed-noise-resisting blind image source separating method based on a feedback mechanism. The method comprises the steps of utilizing wavelet transform to enable mixed signals to be sparse, carrying out linear cluster on sparse wavelet coefficients to estimate a hybrid matrix of a system, carrying out primary separation according to mixed images, then respectively calculating mean values of separating branches, taking out the most value to be outputted, utilizing a 0-setting feedback method to remove signals of the branch out of original mixed signals, carrying out next blind separation on the residual mixed signals in the method, and repeatedly carrying out the process until only a noise branch is left, namely completely separating all mixed signals. The method can effectively and blindly separate image mixing where Gaussian white noise participates. Compared with classis FastICA, the method can achieve higher separation accuracy.

The invention discloses a live video monitoring method and system, a storage medium and an electronic device, and relates to the field of live video monitoring. The method comprises the following steps: a server analyzes a live video stream to obtain a voice data signal, and the server performs modal decomposition on the voice data through the CEEMDAN to obtain an n-order IMF component; the server separates the n-order IMF component through the FastICA to obtain m reconstructed voice signals, wherein m is not smaller than 1 and is not greater than n-2; the server identifies all reconstructed voice signals, judges whether a reconstructed voice signal containing poor information exists, if so, performs abnormality mark on the current direct broadcasting room; and otherwise, performs normal marking on the current direct broadcasting room. By adoption of the live video monitoring method and system, the noise of the voice data signals can be greatly reduced so as to maximally and perfectly identify the live broadcast voice data, greatly improve the monitoring precision and provide forceful guarantee for the spread of bad information on the Internet.

A computer executed method for supervised facial recognition comprising the operations of preprocessing, feature extraction and recognition. Preprocessing may comprise dividing received face images into several subimages, converting the different face image (or subimage) dimensions into a common dimension and/or converting the datatypes of all of the face images (or subimages) into an appropriate datatype. In feature extraction, 2D DMWT is used to extract information from the face images. Application of the 2D DMWT may be followed by FastICA. FastICA, or, in cases where FastICA is not used, 2D DMWT, may be followed by application of the l₂-norm and/or eigendecomposition to obtain discriminating and independent features. The resulting independent features are fed into the recognition phase, which may use a neural network, to identify an unknown face image.

The invention discloses a seismic multi-attribute fusion method based on Shearlet-fastICA. The method is characterized by comprising the following steps: S1, preprocessing seismic attribute, then carrying out shearlet transformation, and determining initial low-frequency subband coefficients and initial high-frequency subband coefficients of different seismic attribute data in a spatial domain; S2, respectively mapping the initial low-frequency subband coefficients and the initial high-frequency subband coefficients from the spatial domain to an ICA domain by using an ICA primary function; S3, carrying out fusion in the ICA domain to determine fused low-frequency subband coefficients and fused high-frequency subband coefficients; and S4, mapping the fused low-frequency subband coefficients and the fused high-frequency subband coefficients back to the spatial domain from the ICA domain and carrying out Shearlet reconstitution to obtain seismic attribute fusion results. According to the seismic multi-attribute fusion method based on Shearlet-fastICA, the mutual redundancy is eliminated by the fused attribute data; the relevant characteristics of the geological mass can be more clearly and obviously described; the powerful technical support is supplied to subsequent reservoir prediction and construction interpretation.

The invention relates to a single-channel gearbox multi-fault separation dual-core micro-processing system and belongs to the gearbox fault diagnosis technology and signal processing technology field.The system of the invention is the dual-core micro-processing system of DSP and ARM. A DSP core includes a phase space reconstruction parameter estimation module, a phase space reconstruction module,a reconstruction parameter adjustment module, a reconstruction signal separation module, and a Fourier transform module. An ARM core includes a time frequency domain spectrogram drawing module. A time delay reconstruction method is used to expand, which avoids human intervention of a reconstructed signal taking experience as a judgment standard. A principal component analysis method is used to reduce influences of a noise and redundancy in a gearbox on the reconstructed signal. A limited support sample kernel function and FastICA fusion algorithm is used to the estimate a source signal probability density function, and then a nonlinear function according with a statistical characteristic of a source signal is obtained. And finally the signal is effectively separated. In the invention, theinfluence of improper selection of reconstruction parameters and a separated nonlinear function is greatly reduced on the aspect of gearbox multi-fault separation.

The invention discloses an electric power system low-frequency oscillation mode identification method based on FastICA and TLS-ESPRIT. The method comprises the following steps: obtaining a low-frequency oscillation signal to carry out zero mean and whitening processing; constructing a separation matrix and calculating an inverse matrix of the separation matrix; performing iterative loop, and updating a separation matrix; after convergence is judged, using the signal recovered by FastICA as a new dominant signal for sampling; carrying out TLS-ESPRIT analysis on the new dominant signal, and calculating the frequency, attenuation factor, amplitude and phase of each oscillation mode. According to the method, the original characteristics of the signal can be well reserved under the condition of noise interference, and the signal-to-noise ratio is improved; and moreover, the method is accurate in identification and higher in precision, can reflect the characteristics of the low-frequency oscillation signal more accurately and comprehensively, and has a good application prospect in the design of a low-frequency oscillation early-warning and damping controller.

The invention relates to an electroencephalogram signal preprocessing method based on a self-adapting noise cancellation system. In the method, fast independent component analysis (FastICA) and discrete wavelet transform (DWT) are used for estimating electronystagmogram signals in original electroencephalogram signals, wherein estimated electronystagmogram signals are used as reference electronystagmogram input signals of a self-adapting noise cancellation system (ANC); and a self-adapting filter and a weighting coefficient updating module are used for automatically removing electronystagmogram artifacts in the electroencephalogram signals, so that the automatic electronystagmogram signal component recognition, automatic parameter updating, and automatic electronystagmogram signal removalare achieved.

The invention relates to the technical field of rotary machine fault diagnosis, and concretely relatesto a rotary machine coupling fault diagnosis method based on SCA and FastICA. The method specifically comprises the following steps: 1, acquiring a rotor vibration signal of a coupling fault of the rotary machine through a plurality of sensors; 2, introducing a synchronous accumulation average noise reduction algorithm, and performing noise reduction on the acquired vibration signal in combination with signal equalization and smoothing processing; 3, separating the denoised signals by using aFastICA algorithm, and separating out each single fault characteristic signal; and 4, carrying out corresponding diagnosis on each separated single fault characteristic signal. According to the method, impulse noise and white noise can be effectively filtered out, noise is reduced, the signal-to-noise ratio is increased, effective extraction of fault feature signals is achieved, and the method isan effective diagnosis method for the coupling fault of the rotary machine system.

The invention discloses an electromechanical equipment bearing fault prediction method based on a Bayesian network of transfer learning. The method specifically comprises the following steps: carrying out noise removal processing on an original acquisition signal through CEEMDAN and FastICA technologies to form a reconstructed original signal; introducing migration learning, clsutering original signals by a neural network to obtain a signal set classified according to fault types, taking the signal set as a target domain data set and as input of a Bayesian network, and selecting a reference sample set from a source domain as a training set of the source domain Bayesian network; based on the covariable shift theory, using the training data of the source domain to complete the goal of minimizing the parameter loss value on the target domain, and optimizing the maximum likelihood estimation of the Bayesian network on the target domain; and returning a result by output of the Bayesian network on the target domain and is visually embodied, giving an alarm in an abnormal state, returning parameter information to mechanical equipment in a decline state, and indicating a machine to adjust self-adjusting parameters such as the rotating speed within a certain range.

The invention relates to the technical field of transfer learning and adaptive neural network clustering bearing fault analysis, and discloses an SOM network clustering electromechanical equipment bearing fault analysis method based on transfer learning and manifold distance. The specific process is as follows: original acquisition signals are de-noised by CEEDAN and FastICA technologies to form reconstructed original signals, and the reconstructed original signals are used as the input of the SOM adaptive neural network; in the SOM network, a manifold distance is considered and utilized to calculate a similarity matrix, a transfer learning model mechanism is introduced, weight parameters of SOM neural network nodes of a source domain and a target domain of a data sample are shared, and a target domain clustering task function and neuron weights are optimized through a source domain training result; and a clustering task of the target domain sample is finished through an SOM neural network target domain output layer result, thus obtaining a clustering result, and outputting a mechanical fault diagnosis type.

The invention relates to a multi-channel electroencephalogram signal eye artifact automatic removal method and a storage medium, and the method comprises the steps: S1, processing an original EEG signal through FastICA and a differential evolution algorithm, so as to obtain a plurality of independent components; s2, processing the independent component containing the ocular artifacts by using wavelet transform and the differential evolution algorithm to obtain an ocular artifact component; and S3, based on wavelet reconstruction and inverse transformation, according to the ocular artifact component, obtaining an EEG signal without ocular artifacts. The invention relates to a method for automatically removing ocular artifacts by combining a differential evolution algorithm, FastICA and wavelet transform. According to the method, wavelet decomposition is adopted to decompose artifact components in FastICA, artifacts can be further separated out, useful information in EEG signals is reserved to the maximum extent, and the problem that the useful information is lost in the automatic artifact removing process is solved.

The invention discloses an automatic weather station real-time data quality control method based on PSR-FastICA-OS_ELM. According to the method, firstly, single station real-time air temperature and a fixed length historical air temperature sequence are collected, and a collection signal is formed; then the collection signal is subjected to basic quality control; the phase space reconstruction is carried out; a reconstructed high-dimensional vector space signal is subjected to rapid independent component analysis, purposes of denoising and redundancy elimination are achieved, and an unbiased signal of a source signal in a high-dimensional vector space is obtained; finally, through an online sequential extreme learning machine, the reconstruction model of a time sequence collection sequence is constructed, and a denoised air temperature reconstruction value is obtained. According to the comparison of the denoised air temperature reconstruction value with an actual observation value, the condition correction is carried out, and the quality control of the air temperature of the time is completed. According to the method, the requirement of real-time data collection is satisfied, and the quality of real-time automatic weather station observation data is improved.

The invention provides a speech enhancement method and device based on a neural network, a terminal and a medium. The method comprises the following steps: acquiring multiple paths of noisy speech signals acquired by each array element of a microphone array; unmixing the multiple paths of noisy speech signals by adopting a fast independent component analysis algorithm fastICA to obtain multiple paths of separation signals; predicting an optimal over-subtraction factor by adopting an error back propagation neural network BPNN; and according to the optimal over-subtraction factor predicted by the neural network, carrying out voice processing on the path of separation signal with the highest signal-to-noise ratio after unmixing by adopting a multi-window spectral subtraction method to obtaina voice enhancement signal. According to the invention, the BPNN is adopted to predict the optimal over-subtraction factor, so that the silence zone is smoother, and the speech quality is improved.