46 results about "Tensor form" patented technology

A method and apparatus for spatio-temporal compressive sensing, which allows accurate reconstruction of missing values in any digital information represented in matrix or tensor form, is disclosed. The method of embodiments comprises three main components: (i) a method for finding sparse, low-rank approximations of the data of interest that account for spatial and temporal properties of the data, (ii) a method for finding a refined approximation that better satisfies the measurement constraints while staying close to the low-rank approximations obtained by SRMF, and (iii) a method for combining global and local interpolation. The approach of embodiments also provides methods to perform common data analysis tasks, such as tomography, prediction, and anomaly detection, in a unified fashion.

The invention discloses a Gaussian-pulse hybrid image denoising method. The method comprises the following steps of: detecting a position of a random pulse noise on a target image, filtering the target image to obtained a filtered image, and recording the position of the random pulse noise by using a probability matrix; decomposing the filtered image into a plurality of image blocks, finding corresponding image sub-block iteration tensor forms on the target image, and establishing corresponding tensor low rank-sparse decomposition models; solving an optimization problem in each tensor low rank-sparse decomposition model, and realizing decomposition iterative optimization by adoption of an alternating direction method until convergence is carried out; obtaining a low rank tensor of each tensor low rank-sparse decomposition model and unfolding the low rank tensors in a matrix form, obtaining a denoising result of each image sub-block, and solving an average value of overlapped regions, so as to obtain a final denoising result. By implementing the method disclosed by the invention, the local and non-local statistical characteristics of the images are effectively fused, the problem that the detection precision of self-adaptive median filters is not enough is overcome, and the Gaussian-pulse hybrid noise can be filtered to realize a denoising effect with a higher signal-to-noise ratio.

The invention discloses a hyperspectral image target detection method based on tensor spectrum matched filtering. The invention relates to target detection of a hyperspectral image. The object of the invention is to solve problems of conventional hyperspectral image target detection methods such as low detection precision and inability of carrying out information mining from three-dimensional data integrally. The hyperspectral image target detection method comprises steps that 1, a target and background signal representation model is established under a tensor representation condition; 2, a to-be-detected hyperspectral image is converted into a three-order tensor form based on a predetermine window, and a hollow X-hollow Y-spectrum-sample four-order tensor 4D is established based on a local neighborhood; 3, covariance matrixes in the hollow X direction, the hollow Y direction, and the spectrum direction of the 4D are acquired; 4, a new three-order tensor after mapping is acquired; 5, the inner products of the target spectrum tensor, the hollow X-hollow Y-spectrum three-order tensor, and the new three-order tensor after the mapping are calculated respectively, and whether the pixel of the to-be-detected hyperspectral image is the detection target is determined. The hyperspectral image target detection method is used for the digital image processing field.

The invention discloses a transportation data loss recovery method based on tensor reconstruction. The transportation data loss recovery method based on the tensor reconstruction aims to resolve the problem that precision is low and loss in a plurality of days can not be processed when an existing traditional transportation data loss recovery method based on a vector or a matrix form is used for recovering loss data. The transportation data loss recovery method based on the tensor reconstruction comprises that (a) transportation data are set in a multi-dimensional tensor form, loss tensor data are expressed through marked tensor, (b) the tensor data are spread on each mode, the relevance of all modes is calculated, and the weight of each mode is obtained, and (c) an objective function of loss data value recovery is set up and the loss data value of the objective function is solved according to the set tensor data and the calculation of the weight of each mode. The transportation data loss recovery method based on the tensor reconstruction is based on a multi-dimensional tensor model, all transportation time-space information is contained, the relevance of multi-mode is fully utilized, at the same time the original structure of multi-dimensional properties and the like of the transportation data is maintained, recovery precision is obviously superior to the traditional recovery method based on the vector or the matrix form, and an extreme case of the loss of a plurality of days can be solved well.

The invention discloses a hyperspectral image feature extraction method based on 3-D wavelet transform and a sparse tensor. The hyperspectral image feature extraction method includes step (1), balancing influence of data on feature extraction distinguishing according to a data normalization method; step (2), extracting spectral domain and spatial domain features from the normalized data by 3-D discrete wavelet transform; step (3), keeping good structural dependence between the features by expressing wavelet transform features as a second-order feature tensor form; step (4), achieving feature sparsification according to a sparse tensor distinguishing method; step (5), re-expressing the features subjected to sparsification as a vector form. By the hyperspectral image feature extraction method, classification accuracy of a whole classification system can be improved effectively.

The present invention relates to a multi-layer automatic coding method and system based on deep learning. Combining the principle of deep learning and tensor algorithm, the original data is expressed in the form of tensor, which can fully excavate the original data without destroying the structure of the original data. The original information, and through multi-layer learning, obtain more essential abstract features, so as to overcome the limitations of vector expression, retain the structural information of the original data to a large extent, and obtain more robust feature extraction and pattern learning , which is conducive to the reflection of the essence of the original data and the subsequent pattern classification.

The invention belongs to the technical field of millimeter wave communication, and particularly relates to a time-varying channel estimation method based on millimeter wave time-varying channel blocksparsity and low-rank characteristics. A certain permutation matrix is introduced into a traditional measurement matrix to prove that a millimeter wave time-varying channel has a block sparse characteristic, and then a traditional CP algorithm is compared with a millimeter wave time-varying channel model to prove a low-rank tensor form of the millimeter wave time-varying channel model. On the basis, a two-step algorithm is provided. In the first stage, a traditional block sparse algorithm is used for obtaining non-zero positions and the number, channel angle information is extracted. In the second stage, a tensor decomposition algorithm is used for extracting channel Doppler and gain information, and then millimeter wave MIMO time-varying channels are obtained through integration. Experimental results show that the algorithm provided by the invention is superior to the traditional compressed sensing algorithm and tensor decomposition algorithm, and approaches the theoretical lower limit under certain conditions.

The invention provides a multi-modal emotion analysis method and system based on deep learning for acupuncture, and belongs to the technical field of emotion recognition. According to the method, feature extraction is carried out on facial expression features by adopting Gabor wavelet transform, feature extraction and dimension reduction processing are carried out on electroencephalogram signals by adopting UPLBP, then sparse linear fusion is carried out on the facial expression features and the electroencephalogram signal features to fuse the facial expression features and the electroencephalogram signal features into a unified and normalized feature vector, and the feature vector is transformed into a tensor form; training is carried out in a CNN-LSTM network, redundant information is removed and predicted emotion classification information is obtained, and a loss function and a correct rate of the network ar calculated by comparing the predicted emotion classification information with actual emotion classification information. According to the invention, the expression features and the electroencephalogram signal features are fused, redundant information is removed, training isconducted through the CNN-LSTM network, the predicted emotion classification information is obtained, and the emotion recognition accuracy is improved.

The invention provides a structural nuclear magnetic resonance image processing method used for Alzheimer disease early detection. The method includes the steps of firstly, conducting feature extraction on a structural nuclear magnetic resonance image expressed in a tensor form through a non-correlation multi-linear algorithm, then, conducting selecting on an extracted algorithm through the Laplacian value algorithm, and finally sending the selected feature into an SVM classifier and conducting training, classifying/diagnosing. According to the method, feature extraction is conducted on the structural nuclear magnetic resonance image through a tensor analysis method, spatial structure information of original data is reserved, the step of feature selection is added, redundant information in the feature is further reduced, and the feature with higher discernment is selected. According to experimental data, the method has high classification accuracy, and a possible auxiliary tool is provided for early diagnosis of the Alzheimer disease.

The invention provides an online prediction method for future tool wearing capacity. The tool wearing capacity for some time to come is predicted by taking tool wearing capacity data in some time agoas input. The online prediction method for the future tool wearing capacity comprises the steps that firstly, influences of the historical wearing capacity on the future wearing capacity are calculated through a long-short-term memory unit encoder, and a state tensor is generated; secondly, the state tensor is taken as input of a long-short-term memory unit decoder, and the wearing capacity for some time to come is generated through the decoder; and in the encoding and decoding process, the encoder, the decoder and the state tensor form a recurrent neural network for predicting future wearingcapacity changes, internal parameters of the long-short-term memory unit encoder and the long-short-term memory unit decoder are automatically obtained through an adam algorithm, and influence factorsof the historical wearing capacity are adjusted. According to the online prediction method for the future tool wearing capacity, the tool wearing capacity evolution trend prediction problem is solved, and the online prediction method for the future tool wearing capacity has the characteristics that the process is easy and convenient to operate, the processing speed is high, prediction is accurate, and the generalization performance is good and can be suitable for the cutting process under different working conditions.

The invention discloses a traffic billboard identification method based on an attention mechanism, which is applied to electronic equipment and comprises the following steps of: inputting an originalimage, and enabling the size of the original image and the gray value range of each pixel point to be in a range of 0-255; performing normalization on the original image, and converting each image into a third-order tensor form: [high, width, cannels]; constructing an attention mechanism, and learning the part, needing to be processed, of the original image; constructing a Faster RCNN multilayer neural network, and extracting and processing features of the image; according to a processing result, delineating a surrounding box of the advertising board. The multi-layer neural network is adopted,the attention mechanism is combined, the recognition score is increased nearby the two sides of the highway, the recognition score is reduced for the part away from the road surface, and therefore the recognition accuracy is improved, and the method has wide application prospects in the fields of image classification and target recognition.

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 a method for efficiently tensioning a fully connected neural network, comprising the following steps of S1 expressing the input vector of the fully connected layer of the neural network as a tensor form; S2 expressing and tensioning the weights of all connection layers of the neural network as tensor ring decomposition form; 3 carrying out the forward propagation of the neural network by using a tensor algorithm of the whole connected layer of the neural network; S4 in the back propagation process of the neural network, updating the core matrix of the tensor ring decomposition form of the weight of the whole connecting layer of the neural network; S5 judging whether the data set completes the iteration, and if so, returning the weight tensor parameter of the network; if not, returning to the step S3. The method of the invention can reduce the parameters used by the neural network and the training time of the neural network, reduce the dependence and consumptionof the large-scale neural network on the hardware level such as the memory size of the computer, the GPU calculation ability and the like, and reduce the number of the neural network parameters without reducing the classification accuracy of the network.

The invention discloses a content recommendation method, system and device and a storage medium, and relates to the field of artificial intelligence, and the method comprises the steps: representing the browsing information of an account through a unit of a third-order tensor based on a third-order tensor formed by triples of an account identifier, a content identifier and a label identifier, obtaining a first relation matrix from the decomposition of the third-order tensor, compared with matrix decomposition and third-order tensor decomposition and extracting more hidden information. The relations between the label and the content can be obtained from other tasks or through simple manual annotation so that acquisition cost is low, the first label feature of the target account can be obtained based on the first relation matrix, and the first label feature of the target account can be obtained based on the second relation matrix. According to the embodiment of the invention, the secondlabel feature of the first candidate content can be obtained based on the second relation matrix, and then the content interested by the user is determined based on the label feature, so that comparedwith the related technology, the embodiment of the invention has more decision information, and the content recommendation accuracy can be improved.

The invention relates to a road network travel time prediction method based on a tensor neural network. The method comprises the following steps: (1) constructing a road network travel time tensor based on multi-dimensional features; (2) decomposing a road network travel time tensor to obtain a feature matrix of the tensor in each dimension; (3) predicting and obtaining a time sequence characteristic matrix through a long and short time recording neural network; and (4) reconstructing a road network travel time tensor, and obtaining predicted road network travel time stored in a tensor form. According to the tensor neural network provided by the invention, large-scale road network time sequence prediction can be achieved only by predicting the factor matrix of the tensor through a big data compression technology, and the tensor neural network has strong robustness for noise and data loss.

The invention provides a deep convolutional neural network tensor input construction method for electric power system analysis and evaluation. The deep convolutional neural network tensor input construction method for the electric power system analysis and evaluation comprises the steps that by using a space distribution representing method of electric power system nodes, an electrical distance between the nodes is described through contract impedance between arbitrary nodes of a power grid; and high-dimensional space node distribution represented by the electrical distance is subjected to dimensionality reduction to a two-dimensional plane by using a node space distribution dimensionality reduction method, and meanwhile, a distance relationship between the nodes of original high-dimensional space is kept. The deep convolutional neural network tensor input construction method is used, electric power system operation data is assigned to the nodes in the two-dimensional plane and a two-dimensional tensor characteristic pattern is obtained, and a plurality of two-dimensional tensor characteristic patterns are superimposed to obtain the electric power system operation data with a tensor form. The deep convolutional neural network tensor input construction method for the electric power system analysis and evaluation improves the precision of the electric power system analysis and evaluation.

The invention discloses calculation and detection methods for strength and an index of a metal material based on a generalized Hosford yield function. The calculation method comprises the following steps of firstly, introducing a direction function and carrying out Taylor expansion to a six-order tensor form to improve the Hosford yield function; secondly, considering a microstructure of the material, researching a plastic parameter thereof and establishing an Hosford yield function of a cubic crystallite orthorhombic sheet under any stress state of a texture coefficient containing a constant term, a quadratic term, a quartic term and a sextic item based on a group theory and objectivity of the material; thirdly, combining a theory of plasticity to deduce yield strength of the metal sheet under the different angles and a relation between an anisotropic index q value and the texture coefficient; and lastly, detecting the yield strength of the orthorhombic metal sheet under the different angles and the anisotropic index q value according to the texture coefficient of the metal sheet. According to the methods, the destructive test on the material is avoided, and the necessary theoretical basis is provided for the research of sheet forming.

The invention discloses a DAS data de-noising method based on wavelet base tensor sparse representation, which is applied to the field of seismic data processing and capable of effectively retaining structural information in DAS data and effectively removing noise. By proposing a new sparse representation model, the DAS data is expressed as a sparse tensor form. In the calculation process, an iterative compression threshold algorithm based on tensor product is used to calculate and reduce the computational complexity.

The invention discloses an SAR image compression method based on robust tensor decomposition, and the method comprises the following steps: converting an original multi-channel SAR image into a tensor; performing mode-n expansion on the tensor to obtain a mode-n matrix, and verifying sparse distribution characteristics of SAR multi-channel image tensor decomposition singular values; setting a singular value threshold value, retaining singular values exceeding the threshold value, performing zero setting on the residual singular values, and performing singular value truncation processing; obtaining approximate tensor representation based on singular value truncation and serving as an initial iteration value of multichannel SAR image compression; calculating robust tensor decomposition based on an augmented Lagrangian multiplier method, and performing dimensionality reduction on the singular value matrix; and performing tensor reconstruction by using the data obtained by dimension reduction to obtain a final multi-channel SAR image compression result. According to the method, the SAR multi-channel image is converted into a tensor form, and the robust tensor decomposition technology is utilized to carry out high-dimensional principal component sparse representation on the image, so that remote sensing SAR image compression is realized, and the influence of an outlier on the image compression effect is effectively inhibited.

The invention discloses a photocontrol direct-current conversion device, which comprises a resistor network and photo-resistors and is remotely controlled by illuminating visible light. Most traditional regulatory structure devices are high in working frequency range, narrow in frequency band and not easy to process. The photocontrol direct-current conversion device disclosed by the invention works in an electrostatic field; the excitation mode is simple; and an artificial electromagnetic material unit structure is realized by a negative resistance form in anisotropic distribution, so that the photocontrol direct-current conversion device has novel performance characteristics. According to the photocontrol direct-current conversion device, the resistance values, namely tensor forms of the resistors, of distributed resistors in various directions are firstly calculated by the transforming optics principle; and the photo-resistors are arranged on a tensor resistor in parallel, and can be controlled by visible light, so as to realize a multifunctional direct-current photocontrol unit structure. In addition, profiting from excellent characteristics of the novel artificial electromagnetic material, the unit structure also has excellent performance such as high stability and wide working frequency range.