353 results about "Factor matrix" patented technology

Text searching is achieved by techniques including phrase frequency analysis and phrase-co-occurrence analysis. In many cases, factor matrix analysis is also advantageously applied to select high technical content phrases to be analyzed for possible inclusion within a new query. The described techniques may be used to retrieve data, determine levels of emphasis within a collection of data, determine the desirability of conflating search terms, detect symmetry or asymmetry between two text elements within a collection of documents, generate a taxonomy of documents within a collection, and perform literature-based problem solving. (This abstract is intended only to aid those searching patents, and is not intended to limit the disclosure of claims in any manner.)

Object clustering techniques are disclosed. A nonnegative sparse similarity matrix is constructed for a set of objects. Nonnegative factorization of the nonnegative sparse similarity matrix is performed. Objects of the set of objects are allocated to clusters based on factor matrices generated by the nonnegative factorization of the nonnegative sparse similarity matrix.

A radio communications receiver which detects radio signals and recovers data representative of the radio signals in a presence of contemporaneously detected interfering signals, the radio communications receiver having a detector that detects said radio signals and generates digital base band signals representative of the radio signals, a matrix former that forms a correlation matrix having rows and columns from the base band signals, and a data processor that generates a factor matrix of the correlation matrix, the factor matrix having rows and columns, and recovers data from the factor matrix. The data processor generates the factor matrix by calculating first elements of the factor matrix from the correlation matrix using a factorization algorithm in accordance with a pre-determined structure of the factor matrix, and forming second elements of the factor matrix by selectively copying first elements in accordance with the pre-determined structure.

The invention relates to a GPS, GLONASS and BDS unified solution positioning method. The method includes the steps of conducting static relative positioning based on GPS, GLONASS and BDS three-satellite unified solution, and conducting dynamic relative positioning based on GPS, GLONASS and BDS three-satellite unified solution. By means of the GPS, GLONASS and BDS unified solution positioning method, solution is achieved through three-satellite unification, positioning accuracy and positioning reliability under the severe observation environment are improved, and initialization time is shortened; the partial ambiguity fixing strategies, namely, the partial ambiguity fixing method oriented to statistic relative positioning and based on the RMS and the partial ambiguity fixing method oriented to dynamic relative positioning and based on the coordinated factor matrix, are adopted, and therefore the success rate of ambiguity fixing can be increased, and accuracy and reliability of baseline solution can be improved; a unified observation model is set up, the method can be easily expanded and used, and the method has a wider application range.

The present invention discloses a completion method of road network traffic data based on an adaptive space-time constraint low-order algorithm. The accuracy of the completed data is greatly improvedwhen a data loss rate is large, and the completion method provided by the invention has a good application effect on traffic data restoration in different time-space loss modes. The method comprises the steps of: (1) constructing a time-space data matrix of road network traffic data; (2) performing factor matrix decomposition of the time-space data matrix, and introducing a non-constraint low-order restoration method; and (3) adding time sequence change features and space similar features of the traffic data as time-space constraint terms to more accurately complete missing points.

The invention discloses a human eye positioning method. The human eye positioning method includes the steps of constructing two kinds of convolution kernels, wherein each kind of of convolution kernels comprises convolution kernels with different radii; cutting out eye area image parts on human face images, conducting convolution on the two kinds of convolution kernels with the different radii and the eye area images so as to obtain convoluted images, and dividing the two kinds of convoluted images so as to obtain a convolution factor matrix. The centers and the radii of the human eyes are obtained by calculating the position corresponding to the maximum value position of the factor matrix and the radii of the convolution kernels. The invention further discloses a sight estimation method based on human eye positioning. Human face feature points and head orientations of the human face images are calculated, so that the sight direction is finally determined in cooperation with human eye center coordinates, canthus coordinates and the head orientations. According to the human eye positioning method and the sight estimation method, the convolutions are calculated through the Fourier transformation, the gray values at the positions of the centers of the eyes are considered, interference of eyebrows, eyelids, glasses and illumination in positioning is reduced, the centers and the radii of the human eyes can be rapidly and accurately positioned, and the sight directions of the human eyes can be rapidly and accurately estimated.

The invention provides a tracking device and a method for solving multiple-target meeting dodging, which can solve the problem of multiple-target meeting dodging. The method comprises: extracting target characteristics in a target image, wherein the target characteristics comprise target position, a column diagram and speed, and outputting the characteristics; according to the position and the speed of a target in the previous frame, predicting the position of a target in the current frame; according to the position of each target, judging whether targets are mutually dodged/ separated; enrolling a dodged target into a dodged group, and removing a separated target from the dodged group; judging a single target, calculating the characteristic matching factor of each single target and the single target prospect, and inputting the matching factor into a matching factor matrix so as to bring convenience for a matching decision in the later stage; selecting the optimal matching pair from the matching factor matrix, and outputting the optimal matching pair; according to the matching result, updating target information which comprises size, area and the column diagram; and finishing target state conversion, deleting a lost target, generating a new target and the like.

The invention discloses a matrix decomposition recommending method based on difference privacy protection. The method comprises the steps of converting the collected evaluation or preference of usersfor an object into a user-score matrix as a training set of a recommending method model; utilizing a score average value, a user factor matrix, an object factor matrix, a user bias item and an objectbias item to predict score conditions for the object by the users; through a difference privacy average value calculation method, calculating an average value of the user scores under difference privacy protection; according to a score prediction model, building a minimum square error function; through a difference privacy random gradient lowering method, training the score prediction model, adding difference privacy noise in the training process and achieving the difference privacy protection of parameters; through the score prediction model and trained difference privacy protection model parameters, predicting scores for the object by the users. When a recommending result is provided, information of the users can be subjected to the difference privacy protection, and the recommending accuracy is high.

The invention relates to a method for pushing recommendation based on user historic behavior interaction analysis. The problem that a data platform in the prior art cannot supply an accurate and customized personalized information service to a user can be solved. The method comprises the following steps: presetting behavior and favorite articles of the user and performing weight allocation; collecting the behavior record information of the user in real time, classifying and then storing; establishing a favorite matrix according to the historic behavior with the highest weight of the user and respectively establishing a user factor matrix and an article factor matrix with the user and the data according to the article information contained in the data; performing singular value decomposition, thereby acquiring a similar matrix, comparing the similar matrix with the favorite matrix, selecting the disliked articles with high scores and recommending to the corresponding user. The method provided by the invention has the advantage that the user and the user as well as the data and the data are combined with each other, so as to form a high-precision relation quantitative index. The method provided by the invention is a continuous learning and promoting process.

A collective matrix is constructed, having a diagnostic sessions dimension and a diagnostic state descriptors dimension. The diagnostic state descriptors dimension includes a plurality of symptom fields, a plurality of root cause fields, and a plurality of solution fields. Collective matrix factorization of the collective matrix is performed to generate a factored collective matrix comprising a sessions factor matrix embedding diagnostic sessions and a descriptors factor matrix embedding diagnostic state descriptors. An in-progress diagnostic session is embedded in the factored collective matrix. A symptom or solution is recommended for evaluation in the in-progress diagnostic session based on the embedding. The diagnostic state descriptors dimension may further include at least one information field storing a representation (for example, a bag-of-words representation) of a semantic description of a problem being diagnosed by the in-progress diagnostic session.

The invention discloses a construction method of a deep neural network model and a fault diagnosis method and system, and relates to the technical field of communication. The construction method comprises the following steps: determining an alarm root derivation rule based on a service path, and in a service topology of a target network, taking adjacent service nodes in the service path from a source end to a destination end as a relationship between a client layer and a service layer; a unified diagnosis factor matrix used for diagnosing faults in the service path is constructed based on theexpert fault diagnosis data, the unified diagnosis factor matrix comprises a root node and alarms and performance state indexes on service nodes associated with the root node, and the root node is a service node generating a source alarm; and constructing a deep neural network model by taking the unified diagnosis factor matrix as an input and the probability vector of the fault cause type as an output, and training and verifying by using sample data. According to the invention, based on comprehensive and effective alarm and performance state indexes, the constructed model can quickly and accurately perform fault diagnosis and positioning.

The present invention discloses a video analysis-based automatic detection method for fighting and brawling abnormal behaviors. The method comprises the steps of S1, inputting an original image in a video sequence; S2, adopting the pyramid L-K optical flow method, calculating the motion vectors of all pixel points in the image; S3, on the basis of correcting a motion speed matrix, additionally correcting a viewing adjustment factor matrix DM; S4, subjecting the corrected motion speed matrix to clustering analysis, and dynamically extracting the effective pixel points of each frame of the image under different behaviors; S5, in conjunction with the information of the motion vectors in the aspects of motion speed and motion direction, putting forward the comprehensive feature indexes of abnormal behaviors; S6, determining a judgment threshold of motion characteristic indexes for judging the occurrence of fighting and brawling abnormal behaviors, and judging the occurrence of fighting and brawling abnormal behaviors according to the threshold. According to the technical scheme of the invention, on the basis of a result obtained based on the current optical flow calculating method, improved optical flow vectors can reflect the motion characteristics of a target in the vertical perspective field of view more accurately.

The invention relates to a spindle system operational modal analysis method based on a pulse excitation response signal cross-power spectrum function. The method comprises the following steps: 1) respectively selecting exciting points on a cutter shank of a spindle system and conducting pulse excitation on the spindle system at the selected exciting points by using a steel hammer; 2) acquiring response signals produced by a reference point and response points after pulse excitation; 3) conducting band-pass filtering on the acquired signals; 4) resolving a cross-power spectrum function between the reference point and the response points, and establishing a matrix equation formed by different sampling time data of the cross-power spectrum function; 5) solving a coefficient matrix by using the matrix equation to obtain system poles; 6) recognizing a modal shape and a modal participation factor matrix; 7) conducting modal assurance criterion matrix value calculation, and when the modal assurance criterion matrix value is within a preset reasonable range, obtaining the modal parameters of the spindle system. The spindle system operational modal analysis method based on the pulse excitation response signal cross-power spectrum function has the advantages that quick calculation can be realized, the accuracy is high, the error control effect is better, the testing strength and time can be reduced and the testing efficiency is greatly improved.

Disclosed is an SCMA (Sparse Code Multiple Access) optimization codebook design method, comprising the following steps: firstly, setting an SCMA codebook parameter; secondly, rotating a QPSK (Quadrature Phase Shift Keying) constellation an angle anticlockwise to maximize the minimum Euclidean distance between projection constellation points of QPSK on two dimensions; thirdly, performing dimensional and point extension on C to obtain a mother constellation C+, and then rotating the C+ to construct a total constellation map c of df users on a single resource block to maximize the minimum Euclidean distance between the users; fourthly, rotating the total constellation anticlockwise to maximize the minimum product distance of each user on the resource block, and then mapping the mother constellation C+ to an SCMA codebook of multiple users with the use of the optimized rotating angle and in combination with a factor matrix F; and finally, in a Rayleigh fading channel, interweaving Q channels of QAM (Quadrature Amplitude Modulation) symbols, obtained by mapping on each resource block, of the users. The SCMA codebook designed by the method has good capability of resisting noise, interference and fading.

The invention provides a cantilever beam operational modal analysis experiment method, which comprises the following steps that: (1) an end point of a cantilever beam is selected to be used as an excitation point to implement pulse excitation on the cantilever beam by using a steel hammer at the selected excitation point; (2) response signals generated by a reference point and a responding point after the implement pulse excitation are collected; (3) the collected signals are subjected to band-pass filtering; (4) a cross-power spectrum function between the reference point and the response point is solved, and a matrix equation formed by data of the cross-power spectrum function in different moments is built; (5) a coefficient matrix is solved by using the matrix equation, and a system pole is obtained; (6) a modal shape and a modal participation factor matrix are identified; (7) the modal assurance criterion matrix value calculation is carried out, and a modal parameter of the cantilever beam is obtained when the modal assurance criterion value is in a preset reasonable interval. The invention also provides a cantilever beam operational modal analysis experiment device. The method and the device provided by the invention have the advantages that the fast calculation can be realized; the precision is high; a better error control effect is achieved; the test intensity and the test time can be reduced; and the test efficiency is greatly improved.

The invention relates to an SCMA codebook design method based on the maximization sum distance of constellation points. The method comprises the following steps of: firstly, maximizing the sum distance among projection points of a rotated QPSK constellation on a coordinate axis, selecting an optimal rotation angle represented by a formula as shown in the specification, and taking the QPSK constellation having the rotation angle represented by a formula as shown in the specification as a mother constellation C+ after normalization processing; then, respectively rotating the projection of the mother constellation C+ on an X axis in alpha, beta and gamma, and constructing a constellation diagram that three users are stacked on a single resource block; on the premise that alpha is kept constant, maximizing a sum distance function f(alpha, beta, gamma) among the constellation points of the three users stacked on the single resource block so as to obtain optimized rotation angles beta and gamma; and finally, designing an operation factor matrix delta<j> according to alpha and the corresponding optimized angles beta and gamma, and furthermore, in combination with a mapping matrix V<j>, mapping the two-dimensional mother constellation C+ to obtain codebooks U<j> of six users, wherein j is 1, 2, ..., 6. According to the method disclosed by the invention, the sum distance among four constellation points of each user and the sum distance among the constellation points of three users stacked on each resource block are respectively maximized through optimal calculation for two times; and thus, the bit error rate performance of the system is improved.

Provided are a camera shooting type remote water meter reading system achieving fast recognition and a reading recognition method thereof. The system comprises a camera, a fill light, a single-chip microcomputer, a storage and an RS485 circuit; the camera is arranged to be just opposite to a dial plate of a direct-reading character wheel water meter, the fill light is arranged near the camera, light is emitted to the dial plate of the direct-reading character wheel water meter, the single-chip microcomputer is connected with the camera, the fill light, the storage and the RS485 circuit separately, image information of the dial plate of the direct-reading character wheel water meter is shot by the camera and sent to the single-chip microcomputer, the single-chip microcomputer firstly trains dial plate image training samples to obtain feature vector matrices of numbers from 0 to 9 and form a factor matrix, then the collected dial plate image information is processed to obtain digital pictures of all character wheels, and finally an actual reading of the water meter dial plate is obtained by multiplying the digital picture of each character wheel by the factor matrix. According to the camera shooting type remote water meter reading system achieving fast recognition and the reading recognition method thereof, the problem that an existing water meter reading system single-chip microcomputer is out of memory can be effectively solved, and the cost of the water meter reading system is lowered.

The invention relates to a cross spectral function-based operational modal analysis experiment method. The method comprises the following steps that: (1), a beam is suspended by elastic ropes and a steel hammer is used for exerting pulsed excitation on the beam at a selected excitation point; (2), response signals generated by a reference point and a response point after the pulsed excitation are collected; (3), band-pass filtering is carried out on the collected signals; (4), a cross-power spectral function between the reference point and the response point is calculated and a matrix equation formed by data of the cross-power spectral function at different sampling moments is constructed; (5), a coefficient matrix is calculated by using the matrix equation so as to obtain a system pole; (6), a modal shape matrix and a modal participation factor matrix are identified; and (7), modal assurance criterion matrix value calculation is carried out; and when the modal assurance criterion value is in a preset reasonable range, a modal parameter of the beam is obtained. In addition, the invention also provides a cross spectral function-based operational modal analysis experiment apparatus. According to the invention, the experiment intensity and time are reduced; and the test efficiency as well as the learning effect of students is also substantially improved.

The invention discloses a point cloud registration error processing method, namely, a method for processing errors generated by point cloud registration in the process of surface reconstruction of an object by using a ground laser scanning technique. The method comprises the following steps: carrying out image matching according to a photogrammetric image matching principle, and determining a mathematical model of error propagation of multistation point cloud registrations; obtaining a coefficient matrix according to error equations; calculating the variance of six parameters, the variance of model points and the variance of unit weight in a rotation matrix; according to a co-factor matrix of the six parameters, obtaining an observed-value co-factor matrix and a variance matrix of estimated observed values, and obtaining a diagonal variance matrix consisting of the six parameters and the model points; according to a law of covariance propagation and a coefficient matrix of error equations, obtaining error propagation models of point cloud registrations under two different scanning coordinate systems; and determining the number of sequential registrations, calculating a parameter error equation coefficient matrix of each registration, and multiplying the coefficient matrixes by a variance obtained after the previous registration is completed, so that an error propagation model of multistation registrations can be obtained.

The invention discloses a power grid subarea division method based on dynamic electricity prices. The method comprises the steps of: obtaining load data of certain time dimensions; calculating a power transmission and distribution factor matrix; using an optical trend model with the target of the lowest power purchase cost of a power grid company to calculate a marginal price of each node in a power grid in each time intervals; setting an expected subarea number, and carrying out initialization to ensure each subarea to be empty; improving a K-means clustering algorithm, and considering adjacency relations of nodes in the subarea division process; utilizing the improved K-means clustering algorithm to carrying out subarea division on the power grid based on the dynamic marginal price of each node; carrying out deviation analysis on a subarea division result, if the largest relative deviation is smaller than a set limit value, indicating that the subarea division is completed, and otherwise, changing the subarea number and carrying out subarea division again; and obtaining the nodes and the regional electricity price of each subarea. The subarea division obtained by the method ensures that nodes in the area are mutually connected and the electricity price loss of the nodes is in an allowable range, so that a market participant can conveniently carry out settling.

A feature point positioning apparatus includes a memory storing computer-executable instructions; and one or more processors configured to execute the computer-executable instructions such that the one or more processors are configured to, iteratively update a first form coefficient based on, a nonlinear feature extracted from an image, and a regression factor matrix obtained through training, and detect a position of the feature point of the image based on, the updated first form coefficient, and a statistical form model obtained through training.