104 results about "Robust principal component analysis" patented technology

The invention discloses a genetic algorithm-based support vector regression shipping traffic flow prediction method, comprising the following steps: (1) reducing the dimensional number of factors possibly generating influences on shipping traffic flow by a weighted principal component analysis method, and selecting influencing factors with higher cumulative contribution rate; (2) carrying out normalization preprocessing on original vessel traffic flow time series data to generate a data set and then grouping; (3) selecting a kernel function to determine support vector machine (SVM) regression parameters; (4) constructing a support vector regression prediction model optimized by a genetic algorithm; (5) inputting the data set to generate a prediction function; and (6) predicting according to the prediction function generated in step (5), evaluating and analyzing prediction error, and if the error is relatively large, returning to step (2) and regulating the parameters again, predicting once again. The method of the invention has the advantages of higher prediction accuracy and higher stability of prediction accuracy.

The invention discloses a method for performing anomaly detection by a weighted convolutional autoencoder-long short-term memory network (WCAE-LSTM network). The method is devoted to perform anomaly detection and positioning by learning a generation model of a mobile pedestrian, thereby guaranteeing the public safety. The invention provides a novel double-channel framework, which learns generationmodes of an original data channel and a corresponding optical flow channel and reconstructs data by utilizing the WCAE-LSTM network, and performs the anomaly detection on the basis of a reconstruction error. In addition, for the problem of complex background, it is proposed that a sparse foreground and a low-rank background are separated by adopting modular robust principal component analysis decomposition; and a weighted Euclidean loss function is designed according to obtained background information, so that background noises are inhibited. The designed WCAE-LSTM network can not only perform the anomaly detection globally but also roughly locate an abnormal region locally; and through the joint consideration of global-local anomaly analysis and optical flow anomaly analysis results, finally robust and accurate detection of abnormal events is realized.

A CNC machine shaping tool is efficiently used by monitoring human audible sound during shaping. A sound information set is created for a tool shaping a workpiece. Shaping sounds are recorded and sliced into short term units. A human operator assigns tool condition labels to each slice. Short term units are combined into mid term units. Noise is reduced by profiling. Mid term sound related features of time and frequency domains are extracted. Dimensionality is reduced by robust principal component analysis. The principal component set is balanced, e.g. by SMOTE. A classifier and principal components are selected. An information set of patterns of values of selected principal components for the tool is created. In an industrial setting, shaping sounds are recorded, noise reduced and select principal component vector values are compared to the tool condition labeled patterns of values in the information set to identify tool condition before deterioration.

The invention discloses a multi-focus image fusion method. The method comprises the steps of firstly carrying out robust principal component analysis on a multi-focus image to obtain a spare component matrix corresponding to a source image, then weighting and averaging the sparse component matrix to obtain a temporary spare component matrix and carrying out quadtree decompression on the temporary sparse component matrix, calculating the gradient energy of matrix subblocks corresponding to the sparse component matrix of the source image according to QT decompression results and comparing the gradient energy to construct a fusion decision matrix, and finally combining the image subblocks corresponding to the source image according to the decision matrix to obtain a fusion image. The method solves the problem that image blocks cannot be divided in a self-adaptive mode by combining RPCA with QT decompression, can effectively restrain block effects and improves the image fusion quality.

A set of input images are acquired sequentially as image tensors. A low-tubal rank tensor and a sparse tensor are initialized using the image tensor, wherein the low-tubal rank tensor is a tensor product of a low-rank spanning tensor basis and corresponding tensor coefficients, and for each image, updating iteratively the image tensor, the tensor coefficients, and the sparse tensor using the image tensor and the low-rank spanning basis from a previous iteration. The spanning tensor basis is updated using the tensor coefficients, the sparse tensor, and the low rank tubal tensor, wherein the low rank tubal tensor represents a set of output images and the sparse tensor representing a set of sparse images.

A computer program product and method for performing automated defect detection of blades within an engine is disclosed. The method may include providing a storage medium for storing data and programs used in processing video images, providing a processing unit for processing images, receiving from a borescope an initial set of images of a plurality of members inside of a device, and using the processing unit to apply Robust Principal Component Analysis to decompose the initial set of images into a first series of low rank component images and a second series of sparse component images, wherein there are at least two images in the initial series.

A computer program product and method for performing automated defect detection of blades within an engine is disclosed. The method may include providing a storage medium for storing data and programs used in processing video images, providing a processing unit for processing images, receiving from a borescope an initial set of images of a plurality of members inside of a device, and using the processing unit to apply Robust Principal Component Analysis to decompose the initial set of images into a first series of low rank component images and a second series of sparse component images, wherein there are at least two images in the initial series.

The present invention discloses a PRCA-based NSST image fusion method. The method comprises the following steps of S1, respectively subjecting to-be-fused images to multi-scale decomposition by using a non-down-sampling redundancy lifting non-separable wavelet (RLNSW), completing the directional localization and then obtaining the low-pass sub-band components and the band-pass sub-band components of the to-be-fused images; S2, fusing the low-pass sub-band components of the to-be-fused images based on the first fusion rule of the robust principal component analysis (RPCA), and fusing the band-pass sub-band components of the to-be-fused images based on the second fusion rule of the robust principal component analysis (RPCA); S3, after the fusion process of the step S2, subjecting the low-pass sub-band components and the band-pass sub-band components of the to-be-fused images to inverse NSST conversion, obtaining a fused image and outputting the image. The method is simple in implementation, good in image fusion effect, short in consumed time and good in practicability.

The invention discloses a data dimension reduction method based on a parallel principal component analysis (PCA) algorithm. The method comprises the steps of: S1, constructing a sample data matrix D<nxm> by high-dimensional data of which dimensions are to be reduced; S2, calculating a covariance matrix C<mxm> of the sample data matrix D<nxm>; S3, calculating m feature values of the covariance matrix C<mxm> and m corresponding feature vectors; S4, determining the number k of principal components according to the feature values and the feature vectors; and S5, utilizing the feature vectors, which correspond to the top-k greater feature values, to construct a transformation matrix, and utilizing the transformation matrix to calculate a principal component matrix, wherein the principal component matrix is data of which the dimensions are reduced. According to the method, the problem that according to a traditional stand-alone principal component analysis algorithm, the data cannot be loaded into a memory at once because a data size is too large is overcome, I / O operations are reduced, and the processing efficiency of data dimension reduction is improved.

The invention discloses an area-of-interest detection method based on full convolution neural network and low-rank sparse decomposition. The method comprises steps of 1) carrying out super-pixel clustering on an original image, extracting color, texture and edge characteristics of each super-pixel and forming a characteristic matrix based on the characteristics; 2) in an MSRA database, based on agradient descent method, learning to obtain a characteristic conversion matrix; 3) in the MSRA database, by use of the full convolution neural network, learning to obtain high-level semantic prior knowledge; 4) by use of the characteristic conversion matrix and the high-level semantic prior knowledge, converting the characteristic matrix; and 5) by use of the robust principal component analysis method, carrying out low-rank sparse decomposition on the converted matrix, and according to sparse noise obtained through the decomposition, calculating a saliency map. According to the invention, themethod is used in an image preprocessing process, and can be widely applied in visual working field like visual tracking, image classification, image segmentation and target re-positioning.

The invention discloses an infrared weak small target detection method based on tensor robust principal component analysis, and relates to the field of infrared image processing and target detection.The method comprises the steps of 1 traversing an original image to construct a third-order tensor; 2 calculating a second-order structure tensor of the original image, and constructing a structure weight tensor; 3 using the tensor robust principal component analysis for constructing an objective function, inputting a third-order tensor and a structure weight tensor into the objective function, and using an ADMM for solving the objective function to obtain a background tensor and an objective tensor; 4 reconstructing a background image and a target image according to the background tensor andthe target tensor; 5 segmenting the target image and outputting a target detection result. According to the method, the problem that the target detection accuracy is low due to the fact that the nuclear norm and the local structure weight adopted in an existing method easily cause local optimal solution and detection target distortion is solved, and the effects of improving the target detection and background inhibition capability and enhancing the target shape keeping capability are achieved.

The invention belongs to the technical field of multichannel clutter suppression, and particularly relates to a multichannel clutter suppression method based on robust principal component analysis. The multichannel clutter suppression method based on the robust principal component analysis comprises the following steps of obtaining a distance-Doppler image corresponding to each reception channel of radar; obtaining corresponding distance-Doppler data according to each distance-Doppler image; conducting column vectorization on distance-Doppler data matrices, and combining all column vectors into a matrix X; decomposing the matrix X through the robust principal component analysis so as to obtain a corresponding sparse matrix S0; according to the sparse matrix S0, obtaining 1 to m, wherein 1 to m are submatrices of X1 to Xm; according to 1 to m, obtaining the optimal weight wopt,j corresponding to a jth Doppler unit; conducting clutter suppression according to the optimal weight corresponding to each Doppler unit, and obtaining a result Xout after clutter suppression.

The invention discloses a polarization SAR terrain classification method based on deep RPCA. The polarization SAR terrain classification method comprises the following steps: (1) reading a polarization SAR image; (2) filtering; (3) extracting characteristics; (4) normalizing characteristic groups; (5) selecting a training sample and a testing sample; (6) training a first layer of deep robust principal component analysis RPCA; (7) training a second layer of deep robust principal component analysis RPCA; (8) training a support vector machine SVM; (9) generating superpixels; (10) classifying; (11) calculating classification precision; (12) outputting a result. Compared with scattering characteristics of the polarization SAR image, the image characteristics extracted according to the method comprise relatively rich terrain information, when the terrain information is classified, the classification precision is effectively improved, and the polarization SAR terrain classification method can be applied to detection and recognition of polarization SAR image targets.

The invention discloses a face identification method based on low-rank decomposition and an auxiliary dictionary under complex environments; the method comprises the following steps: 1, using a non-salient robust principal component analysis method to make low-rank decomposition for an inputted face image, solving a target function based on a norm, and obtaining low-rank contents with complex environment influence primarily removed; 2, correlation-free low-rank decomposition based on a nuclear norm: adding regular terms with inter-class correlation removed into the target function, carrying out singular value decomposition for the low rank contents obtained by previous step for an initialization matrix, and using an ADMM algorithm alternative iteration to solve the low rank dictionary foridentification; 3, classification identification based on auxiliary dictionary learning: obtaining the auxiliary dictionary that simulates complex environment changes, simultaneously using the auxiliary dictionary with the low rank dictionary, and carrying out face classification identification via RADL. The method uses the low rank decomposition target function to fully remove interference information, thus enabling the decomposed face image to have more ID identification capability and anti-environment interference ability.

The invention discloses a thick cloud detection and removal method of time series remote sensing image from coarse to fine. Firstly, the image is pre-processed, including super-pixel segmentation andtransformation of the temporal image to form a matrix. The low rank theory and structural sparseness theory are used to model the background (ideal cloudless surface information) and the foreground (cloud and its shadow) respectively, and the time-series images are separated into the foreground and the background by using the robust principal component analysis framework and introducing affine transformation model, and the cloud and cloud shadow regions at super pixel level are obtained. Then, different scaling factors are set for cloud and non-cloud regions, and the original robust principalcomponent analysis is used to decompose them to remove thick clouds from remote sensing images. The invention greatly improves the precision and efficiency of removing thick clouds from remote sensingimages according to unregistered multi-temporal remote sensing image sequences, and can generate high-precision clouds and cloud shadow detection products, and has extremely high multi-temporal remote sensing image research and application value.

The invention discloses an infrared and visible light image fusion method based on robust principal component sparse decomposition and belongs to the image processing technology field. The method is used to mainly solve a problem that spectral information of a visible light image loses too much during existing infrared and visible light image fusion. The method comprises the following steps of carrying out a robust principal component analysis on images which are accurately registered respectively so as to acquire respective sparse matrixes; carrying out non-downsampling Contourlet conversion on the images respectively so as to acquire respective high and low frequency sub-band images; using the sparse matrixes to fuse the low frequency sub-band image and the high frequency sub-band image; and carrying out non-downsampling Contourlet inverse transformation so as to acquire a fusion image. The method can be applied to registered infrared and visible light image fusion processing.

The invention discloses a tracking method based on a sparse subspace, comprising the steps of firstly learning a plurality of initial frame images by using a random projection matrix and a robust principal component analysis (RPCA) method, and obtaining a low-rank matrix of the images; and extracting the sparse subspace where a tracked target is located from the low-rank matrix. The sparse subspace obtained according to the method has the characteristics of low complexity and high robustness. Compared with the traditional particle filter method based on the target color, target texture or target template, the algorithm based on the target features mentioned in the text has the characteristics of being less in the number of required particles, high in timeliness and strong in stability.

The invention relates to the technical field of ground penetrating radars, in particular to a ground penetrating radar data background removing method based on robust principal component analysis. The ground penetrating radar data background removing method aims to solve the problems of inaccurate estimation of ground penetrating radar data background signals and poor robustness of the traditional background removing method. The ground penetrating radar data background removing method is implemented by the steps of: 1, standardizing ground penetrating radar B-SCAN data; 2, removing DC offset of ground penetrating radar A-SCAN data; 3, adopting a moving average filtering method to remove random noise of a ground penetrating radar B-SCAN image; 4, and removing background through robust principal component analysis. The ground penetrating radar data background removing method has the advantages of achieving good effect and having high robustness by adopting robust principal component analysis, and can be applied to the technical field of ground penetrating radars.

The invention discloses a non-linear dynamic process monitoring method based on the non-linear principal component analysis of a normalized variable, which comprises the following steps: acquiring a data matrix Y, pre-specifying a value of p and a system order n; the Hankel matrix of the past and future observational side values being combined according to the formula; calculating covariance and cross-variance matrices of past and future observations; singular value decomposition of H matrix; calculating a state vector and a residual vector; the state vector being projected onto the high dimensional feature space by explicit second order polynomial mapping; the first k principal components being determined by eigenvalue decomposition in principal component analysis; finally, the T2 statistic, the combined statistic Qc and their corresponding control limits being calculated. The method of the invention is used for monitoring three different types of faults in the Eastman chemical process of Tennessee, and the simulation results show that the proposed CV-NPCA method has high fault detection rate and relatively low fault false alarm rate.

The invention discloses an SAR image target detection method based on sketch and low-rank decomposition, which mainly solves the problem of inaccurate artificial target positioning in the prior art. The SAR image target detection method is implemented by the steps of: (1) extracting a sketch of an SAR image; (2) extracting artificial target potential regions; (3) eliminating false alarm target regions to obtain an artificial target region; (4) constructing an observation matrix of the artificial target region; (5) adopting a robust principal component analysis RPCA method for decomposing the observation matrix to obtain a sparse graph of the artificial target region; (6) and positioning an artificial target. The SAR image target detection method can achieve a good target positioning result, and can be applied in the fields of SAR image target detection and subsequent target recognition.

A Terahertz image target recognition method based on deep learning and RPCA comprises the following steps of (1) removing the background noise by using a robust principal component analysis (RPCA) method; (2) using shape priori knowledge to reduce the region of interest; (3) generating terahertz image data sets; (4) training deep learning network Faster-RCNN; (5) using the deep learning network Faster-RCNN for target recognition. The method removes the background noise interference of the terahertz image by using the robust principal component analysis (RPCA) method, reduces region of interestby using shape prior knowledge, uses the depth learning network Faster-RCNN to perform the target recognition on the region of interest, so tha the method of the invention uses the robust principal component analysis RPCA and the shape prior knowledge to overcome the influence of the background noise, and can detect the object in the terahertz image very quickly and accurately.

The invention discloses a skywave over-the-horizon radar transient interference suppression method based on robust principal component analysis, and the problem of a poor interference suppression effect caused by the need of previous transient interference positioning in the prior art is mainly solved. The method comprises the steps of (1) using distance unit echo signals to construct Hankel matrixes in a plural form and decomposing the Hankel matrixes as a real part matrix and an imaginary part matrix, (2) calculating the sparse matrixes and the low rank matrixes corresponding to the real part matrix and the imaginary part matrix, (3) obtaining the echo signals to transient interference and noise suppression according to the low rank matrix of the real part and the low rank matrix of an imaginary part, (4) subjecting each distance unit in the step (1), (2) and (3) and carrying out Doppler processing to obtain echo distance Doppler information with transient interference and noise suppression. The complexity of the method is low, it can significantly suppress noise and transient disturbances multiple simultaneous presence can be used for sky wave or ground wave OTH radar interference lightning strike, meteor trails interference echo suppression and human impact of interference.

The present invention relates to a computer-implemented method and apparatus for data processing for the purpose of blind separation of nonnegative correlated pure components from smaller number of nonlinear mixtures of mass spectra. More specific, the invention relates to preprocessing of recorded matrix of mixtures spectra by robust principal component analysis, trimmed thresholding, hard thresholding and soft thresholding; empirical kernel map-based nonlinear mappings of preprocessed matrix of mixtures mass spectra into reproducible kernel Hilbert space and linear sparseness and nonnegativity constrained factorization of mapped matrices therein. Thereby, preprocessing of recorded matrix of mixtures mass spectra is performed to suppress higher order monomials of the pure components that are induced by nonlinear mixtures. Components separated by each factorization are correlated with the ones stored in the library. Thereby, component from the library is associated with the separated component by which it has the highest correlation coefficient. Value of the correlation coefficient indicates degree of pureness of the separated component. Separated components that are not assigned to the pure components from the library can be considered as candidates for new pure components. Identified pure components can be used for identification of compounds in chemical synthesis, food quality inspection or pollution inspection, identification and characterization of compounds obtained from natural sources (microorganisms, plants and animals), or in instrumental diagnostics—determination and identification of metabolites and biomarkers present in biological fluids (urine, blood plasma, cerebrospinal fluid, saliva, amniotic fluid, bile, tears, etc.) or tissue extracts.

This invention provides an analyzing method and analyzing device of OD flow, relates to the technical field of networks, and solves the problem in the prior art that the analyzing result of the OD flow of abnormal flow, burst flow, measurement noise and the like which are shown as impact noise of random values, is not ideal. The analyzing method of the OD flow comprises the steps of: performing periodic sampling to obtain the OD flow of a whole network; converting the OD flow of the whole network into a flow matrix; modeling the flow matrix by using a robust principal component analysis model to acquire the robust principal component analysis model of the flow matrix; performing a convex optimization relaxation on the robust principal component analysis model of the flow matrix to acquire a convex optimization model of the robust principal component analysis; solving the convex optimization model of the robust principal component analysis by using an augmented Lagrange multiplier method to decompose the flow matrix into a low-rank matrix and a sparse matrix; and outputting the acquired low-rank matrix and sparse matrix. The method provided the embodiment is used for analyzing the OD flow.

The invention discloses an image noise level estimation method based on principal component analysis. The method comprises: step 1, dividing a raw image into a plurality of image blocks with the same size; step2, turning the selected image blocks into an image data matrix, and calculating a covariance matrix of the image data matrix; step 3, conducting eigen decomposition of the covariance matrix, setting an obtained minimal eigenvalue as an estimate of the noise level, and using the estimate of the noise level to calculate a threshold; step 4, calculating a gradient matrix corresponding to each image block and the trace of a covariance matrix corresponding to the gradient matrix; and step 5, picking the image block with the trace not greater than the threshold, and repeating the step 2 to 4 until the estimates of the noise level of two continuous calculations are the same or the iteration is repeated for a preset number of times. The method uses principal component analysis to find a smooth block in the image more accurately. The method is precise and robust and is applicable to large-scale noise levels and various scenes.

Aiming at a multi-channel synthetic aperture radar system under a strong clutter background, the invention provides a multi-channel synthetic aperture radar RPCA amplitude-phase joint target detectionmethod and device in combination with a robust principal component analysis (PRCA) method and an amplitude-phase joint detection method. According to the method, firstly, focused images of all channels are subjected to column vectorization and then stacked to obtain a matrix X, then the matrix X is decomposed through an improved RPCA method to obtain a low-rank matrix L and a sparse matrix S, andfinally phase information is introduced for amplitude-phase joint detection. According to the method, specific clutter and moving target information are comprehensively considered, and the moving target detection performance is effectively improved; the advantages of an RPCA and amplitude-phase joint detection method are fully combined through a two-step detection method, and the false alarm rateunder a low signal-to-clutter ratio condition is reduced; and the matrix decomposition technology adopted by the method can greatly reduce the operation complexity and improve the algorithm efficiency.

The invention relates to a ground-penetrating radar clutter removing method. The method comprises the following steps of stepS1, using a Stolt offset method to carry out focus processing on ground-penetrating radar original data so as to obtain a ground-penetrating radar image matrix M; stepS2, using a robust principal component analysis method to decompose a ground-penetrating radar image M so asto acquire a low rank matrix L and a sparse matrix S which are a clutter L and a target S; and stepS3, removing the clutter L, and determining the retained target S to be the final ground-penetratingradar image after the clutter is removed. By using the ground-penetrating radar clutter removing method, the advantage of a Stolt offset technology is combined with the advantage of robust principalcomponent analysis; and compared with a direct robust principal component analysis algorithm, the method can be used to solve a problem that target echo energy is not concentrated, greatly increase sparseness, and avoid a large error due to the improper selection of a principal element.

A set of input images are acquired sequentially as image tensors. A low-tubal rank tensor and a sparse tensor are initialized using the image tensor, wherein the low-tubal rank tensor is a tensor product of a low-rank spanning tensor basis and corresponding tensor coefficients, and for each image, updating iteratively the image tensor, the tensor coefficients, and the sparse tensor using the image tensor and the low-rank spanning basis from a previous iteration. The spanning tensor basis is updated using the tensor coefficients, the sparse tensor, and the low rank tubal tensor, wherein the low rank tubal tensor represents a set of output images and the sparse tensor representing a set of sparse images.

The invention discloses a speech enhancement system and method based on a multi-subband short-time fractional Fourier spectrum random rearrangement robust principal component analysis MFrSRRPCA algorithm. The realization steps are: a time-frequency analysis module generates time-frequency information of noisy speech; the time-frequency analysis module generates time-frequency information of noisyspeech. The time-frequency subband division module divides the time-frequency amplitude spectrum of the noisy speech into a plurality of noisy subbands. Each time-frequency amplitude spectrum enhancement module randomly disrupts the sequence of each frame spectrum element in the corresponding noisy sub-band, and generates the corresponding enhancement sub-band by using a robust principal componentanalysis algorithm according to the noise intensity estimation value in the corresponding sub-band. The time-frequency subband recombination module composes all the enhancement subbands to enhance the time-frequency amplitude spectrum. The time-domain speech reconstruction module reconstructs the enhanced time-frequency amplitude spectrum into enhanced speech. The invention can improve the soundquality and intelligibility of the noisy speech, and can be used for the speech enhancement and noise reduction of the speech receiving system.

The invention discloses an overhead line system insulator state detection method based on a robust principal component analysis method. An insulator sample data set is established according to the acquired images of the overhead line system support and suspension device, and a Mask-RCNN convolutional neural network is adopted to perform target detection and segmentation, and therefore, the positions of insulators in the images can be positioned, and the insulators can be segmented; the minimum external moment of the insulators is calculated according to the positioning result, the inclinationangle is detected, and the obtained picture is rotated according to the inclination angle to obtain a horizontal insulator image; the collected insulator images are cut one by one to obtain a single insulator piece data set with a fixed visual angle; foreground and background segmentation is performed on the insulator sheet data set with the fixed visual angle; and texture feature extraction is carried out on the separated foreground through a gray level co-occurrence matrix, texture features of the image are extracted by adopting energy and entropy, weighted summation is carried out accordingto whether the texture features are positively correlated, and a threshold value is set to identify the states of the insulators. According to the invention, detection and rapid positioning of defective insulators, dirt and other bad states are realized.