39 results about "Higher order tensor" patented technology

The present invention discloses a deep neural network compression method based on block item tensor decomposition. The method comprises the steps of: obtaining a framework of a deep neural network, respectively converting a weight matrix W and an input vector x to a high-order tensor (u)W(/u) and a high-order tensor (u)X(/u), performing block item tensor decomposition processing of the high-ordertensor (u)W(/u), replacing a full-connection layer of the deep neural network with a block item tensor layer, and employing a back-propagating algorithm to perform training of the deep neural networkreplaced in the step D. The deep neural network compression method based on the block item tensor decomposition employs a block item tensor decomposition method to construct the block item tensor layer to replace the full- connection layer of the original deep neural network, the features of symmetry and index expression capability of the block item tensor layer are employed to greatly compress the parameter amount of the full-connection layer and maintain the classification precision of the original network.

The invention discloses a multi-visual angle gait recognition method and a system based on higher-order tensor subspace learning, which belong to the field of intelligent recognition. A gait video is acquired from multiple representational angles, and a gait sequence image is obtained through framing interception; background extraction, background subtraction and binary processing are carried out on the gait sequence image respectively, black and white visual effects are presented, and a contour sequence under the multiple visual angles is obtained; the contour sequence is converted to tensor data; a higher-order discriminant tensor subspace analysis algorithm based on graph embedding obtained after expanding DTSA on the basis of multilinear discriminant analysis and a graph embedding principle is used for carrying out dimension reduction and feature extraction on the tensor data; and according to the extracted and obtained multi-visual angle gait features, the gait features are subjected to similarity measurement, and a recognition result is obtained. The method is simple, the cost is low, person identity authority detection and disguised person identity authentication can be automatically carried out on a particular place, and safety protection on the monitored place and identity authentication in multiple conditions can be effectively improved.

Computer-readable media having computer-executable instructions and apparatuses categorize documents or corpus of documents. A Tensor Space Model (TSM), which models the text by a higher-order tensor, represents a document or a corpus of documents. Supported by techniques of multilinear algebra, TSM provides a framework for analyzing the multifactor structures. TSM is further supported by operations and presented tools, such as the High-Order Singular Value Decomposition (HOSVD) for a reduction of the dimensions of the higher-order tensor. The dimensionally reduced tensor is compared with tensors that represent possible categories. Consequently, a category is selected for the document or corpus of documents. Experimental results on the dataset for 20 Newsgroups suggest that TSM is advantageous to a Vector Space Model (VSM) for text classification.

The invention relates to the technical field of medical instruments, provides a novel practical method for extracting MRI (magnetic resonance imaging) parameters of structural abnormality of white matters of a human brain, and aims to achieve the purpose of objectively evaluate the lesion degree by analyzing anisotropy of tissues through a certain characteristic extraction method. The invention adopts the technical scheme that the method for extracting high-order tensor characteristic parameters of diffusion kurtosis tensor imaging comprises the following steps: a subject collects diffusion weighted signals of tissues along a plurality of directions on a magnetic resonance scanner, then a second-order or fourth-order tensor that reflects diffusion and distribution probability density function of water molecules in the tissues can be obtained through fitting, and lesion characteristics of diffusion and distribution abnormality of the water molecules, caused by structural abnormality of the tissues, can be obtained through certain tensor analysis, wherein parameters from two different angles, including diffusion coefficient and kurtosis coefficient, are mainly included. The invention is mainly applied to design and manufacture of medical instruments.

A method is provided for scalable supervised high-order parametric embedding for big data visualization. The method is performed by a processor and includes receiving feature vectors and class labels. Each feature vector is representative of a respective one of a plurality of high-dimensional data points. The class labels denote classes for the high-dimensional data points. The method further includes multiplying each feature vector by one or more factorized high-order tensors to obtain respective product vectors. The method also includes performing a maximally collapsing metric learning on the product vectors using learned synthetic exemplars and learned high-order filters. The learned high-order filters represent high-order embedding parameters. The method additionally includes performing an output operation to output a set of data that includes (i) interpretable factorized high-order filters, (ii) exemplars representative of the class labels and data separation properties in two-dimensional space, and (iii) a two-dimensional embedding of the high-dimensional data points.

The invention discloses a target tracing algorithm based on a high-order partial least square method. The target tracing algorithm based on the high-order partial least square method comprises an initialization phase, a training phase, a testing phase and an updating phase. A two-dimensional tensor is used for representing image blocks, and internal spatial structure information of the image blocks is stored. Meanwhile, the image blocks of multiple timing sequences and the types of the image blocks are integrated to build a high-order tensor, the relevance between the high-order tensor and the category array of the high-order tensor is analyzed through the partial least square method, and the performance of the target tracing algorithm is largely improved.

The invention belongs to the technical field of color video recovery, and discloses a color video recovery method and system based on high-order tensor singular value decomposition, and the method comprises the steps that a low-rank part and a sparse part are extracted from a p-order tensor based on a high-dimensional robust principal component analysis (HTRPCA) problem; according to the p-order tensor, conjugate properties of a block diagonal matrix are generated, and an improved algorithm of a tensor product and tensor SVD is proposed; T-SVD is implemented on the tensor, and a new tensor dimension rank is defined; a tensor nuclear norm based on a tensor forward rank is utilized, and a near-end operator and an ADMM algorithm are utilized to solve an HTRPCA problem, so that accurate recovery guarantee of the HTRPCA is realized. Experimental results show that on the premise of theoretical optimal parameters, for generating data and color videos, HTRPCA has superiority in the aspect of processing corresponding problems, and theoretical guarantee is provided for accurate tensor recovery.

The invention discloses a tensor decomposition-based channel state information positioning fingerprint construction method. The method comprises the following steps of: firstly, expressing acquired channel state information (CSI) data as a three-dimensional image; regarding the three-dimensional image as a third-order tensor; then, combining a tensor decomposition algorithm based on a Parallet Factor (PARAFAC) analysis model and an ALS (Alternate Least Squares) iterative algorithm for noise reduction processing of the tensor; then, carrying out single-layer tensor wavelet decomposition on three dimensions of the CSI image by using a tensor wavelet decomposition algorithm, and calculating wavelet coefficients of wavelet sub-components by using an angular second moment; and finally, obtaining the CSI positioning fingerprint corresponding to each reference point coordinate. According to the method, the characteristic that high-order tensors can describe data information and structures isfully utilized, complex data is expressed in a tensor form, noise reduction and feature extraction of tensor images are finally achieved, and the data processing and analyzing capacity is improved.

The invention relates to the medical imaging technology, in particular to a diffusion tensor imaging method for processing nerve fiber reconstruction in a nerve complex distribution area, device and magnetic resonance equipment. A multivariate high-order tensor model is adopted to describe the diffusion motion in a voxel, and in combination of a high-order tensor decomposition theory, the directions of all fibers in the voxel are solved. The method disclosed in the invention is particularly suitable for reconstructing nerve fibers in the nerve complex distribution area and can effectively process nerve fiber bundle intersection, bifurcation and other conditions. The method, the device and the equipment have important application value in neuroscience, medical imaging and the like.

The invention relates to an intelligent roadside multi-source data fusion method based on Bayesian tensor decomposition. The method comprises the following steps: (1) data preprocessing: collecting traffic flow information at regular intervals through a video and radar detector, marking time labels on lanes, and carrying out the effective recognition and processing of missing and abnormal data to obtain an original data set; and (2) data fusion: constructing a speed tensor, and solving related parameters by using the Bayesian thought to obtain a fused speed tensor. The method is based on intelligent road side heterogeneous data, meanwhile, the Bayesian statistical idea is utilized, the tensor decomposition theory is introduced, the characteristics of high-order tensor mining data characteristics are fully considered, effective fusion of data between different data sources is achieved, and therefore the precision and quality of the data are improved.

The invention relates to a color image reconstruction method based on a tensor enhancement technology of a local data block, which comprises the following steps: 1, dividing a to-be-reconstructed image into local data blocks, carrying out structured addressing by taking the local data blocks as tensor enhancement units, converting the to-be-reconstructed image from three-dimensional data into high-dimensional data, and obtaining a high-order tensor; 2 reconstructing the high-order tensor by using a tensor chain nuclear norm minimization model to obtain a reconstructed tensor; and 3 carrying out inverse operation of the step 2 on the reconstructed tensor according to an index position to obtain a converted image, calculating the gray value of each pixel point of the converted image, and restoring the size of the converted image to the original size of the to-be-reconstructed image to complete the reconstruction of the color image. According to the method, a local data block is used as a tensor enhancement unit for structured addressing, the local data block is used as a whole for operation, the complete structure of the local data is reserved, and artifact blocks caused by thorough disruption of the data structure on a reconstructed image are reduced.

The invention discloses an image restoration method based on smooth Tacker decomposition and high-order tensor Hankering. The method comprises the following steps: 1) inputting a to-be-restored imageto determine a to-be-restored area of the image; 2) constructing a high-order tensor Hankering and discrete total variation model; and 3) constructing a smooth Tacker decomposition and high-order tensor Hankering image restoration model in combination with the step 2), restoring the color image, and finally reconstructing and outputting a high-quality visual data image. The method has the advantages that the image processing efficiency and the image restoration accuracy are both considered.

The invention provides a compression method and device of a neural network model. The compression method of the neural network model comprises the following steps: acquiring a parameter matrix in an embedded layer of the neural network model; receiving an order raising instruction, and performing order raising on the parameter matrix according to the order raising instruction to obtain a high-order tensor; decomposing the high-order tensor to obtain a decomposition result of the high-order tensor; and updating the parameter matrix according to the decomposition result to obtain a compressed neural network model. According to the method provided by the invention, on the basis of keeping the performance of the neural network model, the parameter quantity in the embedded layer of the neural network model is effectively reduced, and the volume of the neural network model is remarkably reduced.

The invention discloses a method for automatically selecting a sparse tensor storage format based on a convolutional neural network. The histogram indicates that the matrix is ​​scaled to a fixed size; 3) The fixed size matrix is ​​used as the input of the convolutional neural network CNN, where the structure of the CNN is designed and customized for the automatic selection of the sparse tensor storage format; 4) The method of supervised learning is used Train CNN and get the trained network model; 5) Take the new sparse tensor as the input of the network model, and get the optimal storage format of the tensor after forward propagation. The invention utilizes the advantages of CNN in the classification problem, and combines the feedforward neural network FFNN to adapt the prediction of the optimal sparse tensor storage format, and on the premise of fully retaining the tensor characteristics, the sparse tensor is effectively converted into a fixed tensor. Matrix input of size, which can be applied to automatic selection of sparse format for higher-order tensors under arbitrary tensor computation.

The invention discloses a sub-array split L-type co-prime array direction-of-arrival estimation method based on fourth-order sampling covariance tensor denoising, which mainly solves the problems that a signal structure is damaged and high-order virtual domain statistics are interfered by noise items in the existing method, and comprises the following implementation steps of: constructing a linear sub-array split L-type co-prime array; modeling a received signal of the L-type co-prime array and deriving a second-order cross-correlation matrix; deriving a fourth-order covariance tensor based on the cross-correlation matrix; fourth-order sampling covariance tensor denoising is realized based on kernel tensor thresholding processing; deriving a fourth-order virtual domain signal based on a denoising sampling covariance tensor; constructing a de-noised structured virtual domain tensor; and obtaining a direction of arrival estimation result through structured virtual domain tensor decomposition. According to the method, high-order tensor statistical distribution characteristics of the constructed sub-array split L-shaped co-prime array are fully utilized, high-precision two-dimensional direction-of-arrival estimation is realized through de-noising virtual domain tensor signal processing, and the method can be used for target positioning.