91 results about "Matrix completion" patented technology

A computing cluster operated according to a resource allocation policy based on a predictive model of completion time. The predictive model may be applied in a resource control loop that iteratively updates resources assigned to an executing job. At each iteration, the amount of resources allocated to the job may be updated based on of the predictive model so that the job will be scheduled to complete execution at a target completion time. The target completion time may be derived from a utility function determined for the job. The utility function, in turn, may be derived from a service level agreement with service guarantees and penalties for late completion of a job. Allocating resources in this way may maximize utility for an operator of the computing cluster while minimizing disruption to other jobs that may be concurrently executing.

The invention relates to the technical field of 3D point cloud data processing and 3D scene reconstruction. A fusing method of a 3D laser point cloud and a 2D image comprises the following steps: (1) making a black-and-white grid vertical hole calibration board; (2) collecting 3D laser point cloud and 2D images; (3) calculating the space coordinates of the centers of the vertical holes; (4) calculating the pixel coordinates of the centers of the vertical holes; and (5) calculating a mapping matrix to complete data fusion. The method has the advantages as follows: firstly, linear point cloud data of a laser scanner and image data shot by a camera are used, and the amount of data sampled is small; secondly, for solving of the mapping relation between 3D laser point cloud and 2D images, the amount of calculation is small, the calculation process is simple, and the calculation efficiency is high; thirdly, real-time 3D laser point cloud and 2D image fusion can be realized; fourthly, the space coordinates of the centers of the vertical holes are calculated using an interval matrix, there are multiple constraints, and calculation is more accurate; and fifthly, the homographic relation between the laser scanning plane and the camera imaging plane is directly constructed based on the imaging principle of the camera to solve the mapping matrix, and the calculation method is simple and reliable.

The invention relates to a method for error suppression of inertial concretionary coarse alignment of a strapdown inertial navigation system and belongs to the technical field of inertial navigation. The method comprises the following steps of 1, according to current longitude and latitude of a carrier, carrying out transformation of an earth coordinate system into a navigation coordinate system, 2, driving two shafts of a rotation-type inertial navigation system to periodically rotate according to a certain rotating scheme, 3, according to angular velocity information output by a gyroscope in coarse alignment, updating a direction cosine matrix of a posture change of an inertial measurement unit (IMU) coordinate system relative to a p0 coordinate system through calculating a numerical solution of a matrix differential equation, and 4, orderly calculating solutions of matrixes to obtain a posture matrix of the carrier so that coarse alignment is finished. The method has the advantages that through an integral operation in coarse alignment, swinging and shaking-caused error components of outputs of an accelerometer of a ship in alignment of storm-caused swinging and shaking can be compensated; and through IMU double-shaft rotation modulation, an accelerometer zero-bias error accumulated over time is suppressed so that the precision of inertial concretionary coarse alignment is improved.

The invention relates to a superpixel segmentation method based on the fuzzy theory. The superpixel segmentation method improves fitness of superpixel edges and original image edges, and segmented superpixels have an undiversified internal structure and uniform gray scale. The superpixel segmentation method comprises steps of 1) processing images and initializing cluster centers; 2) adding coordinate distances of pixel points into the fuzzy C-means clustering algorithm and establishing an objective function; 3) updating a membership matrix; 4) updating cluster centers with gray scale values and coordinates included; 5) repeating steps 3) and 4) until a termination condition that the change amount of the cluster center gray scale values is smaller than an artificially set threshold value or iteration times are larger than some artificially set value; 6) completing initial superpixel division based on the finally obtained membership matrix; and 7) performing post-processing to complete the final superpixel division by combining sets of points ,which are isolated and unavoidably exist in superpixels generated after the above six steps, with adjacent superpixels which have the maximum similarity.

The invention belongs to the field of signal processing, and particularly relates to a DOA (Direction of Arrival) estimation method through second-order statistics reconstruction based on matrix completion. Based on the matrix completion theory, the DOA estimation method through second-order statistics reconstruction based on matrix completion is put forward. Firstly, based on the matrix completion method, an elastic regularization factor is introduced to reconstruct a received signal covariance matrix to a noise-free covariance matrix; under the second-order statistics domain, through matrixsummation averaging, a multi-vector problem of the noise-free covariance matrix is converted to a single-vector problem; and finally, a sparse reconstructed weighted l1 norm is used to realize DOA parameter estimation. Non-uniform noise influences can be significantly suppressed, the DOA parameter estimation performance is good, and in a condition of a low signal-to-noise ratio, the provided algorithm has high angle estimation precision and high resolution.

The invention discloses a waveform optimized design method for a frequency controlled array MIMO radar system, aiming to solve the problem that too much test data is to be processed for obtaining distance, angle, and resolution of high dimension Doppler. The implementation process is: 1) according to the target environment requirement, determining the transmitting carrier frequency, frequency increment, antenna quantity, and waveform code element length of the frequency controlled array MIMO radar; 2) according to a received data model, constructing decoupling between transmitted waveforms and virtual steering vectors; 3) according to a constructed data matrix, determining the coherence of the data matrix; 4) according to the coherent characteristics of matrix completion, designing transmitting waveform of the frequency controlled array MIMO radar by use of optimized method. By use of the method, through the transmitting waveform design for the frequency controlled array MIMO radar, the coherence of the received data is made as low as possible. Therefore, the method has an advantage that smaller data quantity is needed when accurate target detection and estimation are obtained; the method can be applied in frequency controlled array MIMO radar waveform design.

The invention discloses a Chinese term automatic correction method in an input process, comprising the following steps: (1) using each element input by users as a Term, which is one of a Chinese character string, a letter string or a mixing string of Chinese characters and letters; (2) quickly judging whether the Term is a valid term by structure of a Bloom filter; if the Term is a valid term, carrying out term completion on the Term; otherwise, carrying out term correction on the Term; the term completion means to provide completion prompting to a right short term and the term correction means to correct the incorrect input to obtain the valid term prompting closing to the intention of a user; the term completion and the term correction both adopt term clustering and a Trie tree construction and search mechanism; the judging mechanism of the Bloom filter is adopted in the user input judgment; for the right input, the valid term completion mechanism is adopted to carry out term completion prompting; to input error by users, the original input is restored based on operation characteristics of all sorts of Pinyin input method software and the correction prompting is carried out on the error correction mechanism compiled aiming at the original input.

The invention discloses a sensor network indoor fingerprint positioning method based on matrix completion is applicable to the robust indoor positioning method of the wireless sensor network WSN. The sensor network indoor fingerprint positioning method disclosed by the invention comprises steps of utilizing a matrix completion theory to recover a complete fingerprint database by only sampling part of signal fingerprints through utilizing the low-rank property of a fingerprint matrix, adopting a classic KNN algorithm to perform object positioning of an online phase after the fingerprint database is constructed, in the process of the fingerprint matrix complementation and for effectively eliminating the outlier noise and the Gauss noise contained by the signal fingerprint, respectively introducing an L1 norm regularized item and an F norm regularized item, modeling a fingerprint data recovery question into a norm regularized matrix complementation question and solving the question through an alternative direction mutiplier method algorithm The sensor network indoor fingerprint positioning method based on matrix completion can effectively reduce the workload of the information fingerprint database construction and can acquire positioning precision which is higher than the positioning precision obtained from the same kind method under various noise conditions.

The invention discloses a realization method of an FDA MIMO radar system based on matrix completion, belongs to the field of radar technology, and aims at reducing measuring data which needs processing and obtaining higher distance, angle and Doppler-dimension resolutions. The method comprises the following steps that 1) according to practical requirements and characteristics of the FDA MIMO radar system, an emission carrier wave, an frequency array, the amount of emission antennas and the code-element length of the emission carrier wave are determined, and an emission model of a FDA MIMO radar is established; 2) according to a target space environment, the type of reception antennas and the amount of antennas of the FDA MIMO radar are determined, and a reception signal model is constructed; 3) based on obtained partial reception data, a Lq norm space processing technology to obtain recovery of a data matrix; and 4) the data matrix is combined with a parameter estimation method to obtain combined estimation of a target distance, an orientation 2D space and Doppler.

The invention discloses a Hankel tensor decomposition-based seismic signal reconstruction method, which comprises the steps of constructing a Hankel objective function; converting the Hankel objectivefunction into a matrix completion objective function; and solving the function by using an alternate minimization method in an alternating manner. According to the Hankel tensor decomposition-based seismic signal reconstruction method, the problem of overhigh low-rank requirements on data due to the fact that regularization is carried out only through low-rank tensor decomposition is solved through introducing low-rank tensor decomposition and effectively combining a Hankel structure with a low-rank tensor, meanwhile, the decomposition process of solving a singular value of the tensor is avoided through the low-rank tensor decomposition method and the solving speed is greatly improved.

The invention discloses a method for estimating the angle of MIMO (Multiple-Input Multiple-Output) radar with array element faults based on block Hankel matrix completion. According to the method, firstly, a matrix with a 2-fold Hankel structure is constructed by using column vectors of a virtual array covariance matrix, a block Hankel matrix with a 4-fold Hankel structure is constructed by takinga matrix formed by column vectors as a sub-matrix for enabling each row and each column of the constructed block Hankel matrix to have sampling elements and meet the low-rank property, and missing data in the block Hankel matrix is completed by utilizing matrix completion; then, the corresponding redundant elements are averaged in the recovered block Hankel matrix; and finally, inverse transformation is carried out on the block Hankel matrix without data missing to obtain a complete virtual array covariance matrix, and the DOD (Direction of departure) and DOA (Direction of arrival) of a target are estimated by adopting an array covariance matrix-based algorithm (such as an ESPRIT algorithm). According to the method, a large amount of missing data in the covariance matrix of the MIMO radarvirtual array with the array element faults can be effectively recovered, and the angle estimation performance when the array element faults are transmitted or received is improved.

The invention discloses a non-local image inpainting method based on low-rank matrix recovery. The method comprises the following steps: pre-completing a low-rank texture and a natural image respectively; partitioning the pre-completed image into a plurality image blocks, and calculating the matching block matrix of each image block on the basis of block matching and a grouping method; inpainting each image block in the matching block matrix by using low-rank matrix completion; integrating the repairing results of all the image blocks, and repairing an image to be repaired according to an integration result. Two image completion methods based on examples and numerical calculation are used comprehensively. The method has a wider application scene range than an existing image completion method. In particularly, the image can be reconstructed from random sampling in an almost perfect way. Moreover, pre-completion is performed firstly, and block completion is performed secondly in a grouping way, so that the repairing accuracy is increased greatly. The method can be conveniently applied to a plurality of practical applications such as rendering acceleration, image inpainting and object removal.

The invention belongs to the field of computer vision, and aims to provide a simple and practical background extracting method. The technical scheme adopted by the invention is as follows: a video background recovery method based on movement information and matrix completion comprises the steps of adopting movements of a moving object between frames and frames through optical flow detection, detecting the movement information to generate weight matrixes, and arranging the weight matrixes according to column vectors to generate a total weight matrix; then, arranging all sampled frames according to the column vectors to form original data array matrixes; multiplying observation matrixes and the total weight matrix according to elements to obtain the observation matrixes; after that, obtaining the completed observation matrixes through the matrix completion; finally, recovering each array of the matrixes according to the size of original sampled frames to obtain a background image. The method is mainly applied to background extraction.

In a method and magnetic resonance (MR) apparatus for MR image reconstruction, an image reconstruction algorithm is used that operates on one calibration matrix that is formed by reorganizing a number of individual, undersampled k-space data sets respectively acquired by multiple reception coils in a parallel MR data acquisition from a subject exhibiting motion. The motion causes the k-space data sets to exhibit errors. In order to use the calibration matrix in the reconstruction algorithm, it is subjected to an iterative rank reduction procedure in which, in each iteration, a residual is calculated for each data point that represents how poorly, due to motion-induced corruptions, that data point satisfies the low rank constraint, and non-satisfying data points are removed from the data point for the next iteration. The resulting low rank matrix at the end of the iterations is then used to produce images with fewer motion-induced errors.

The invention discloses a matrix completion based WMSN (wireless multimedia sensor network) image compression acquisition method. The method comprises the steps as follows: after a system is initialized, a camera node obtains original image data A, the original image A is divided into block images of equal size, the sampling rate r is set, and sampling matrix numbers are selected according to r and are distributed to ordinary nodes for processing; the ordinary nodes extract a random sampling matrix Wij from a random sampling matrix library M, compressed data of a block image ak is obtained, the ordinary nodes send the compressed data and random matrix numbers i and j to a cluster head node, and the numbers are transmitted to a base station node for reconstruction after summarization; a base station generates an incomplete block image Xk, and a block matrix ak is obtained with a matrix completion reconstruction algorithm; after splicing, a complete reconstructed image B is obtained and can approximately represent the original image A at this time. With adoption of the method, energy consumption of sensor nodes can be effectively reduced, work efficiency of the sampling end is improved, and the life cycle of the whole network is prolonged.

An optical coherence tomography (OCT) image composed of a plurality of A-scans of a structure is analyzed by defining, for each A-scan, a set of neighboring A-scans surrounding the A-slices scan. Following an optional de-noising step, the neighboring A-scans are aligned in the imaging direction, then a matrix X is formed from the aligned A-scans, and matrix completion is performed to obtain a reduced speckle noise image.

The invention discloses a wireless sensor network positioning method based on matrix completion. The method comprises the following steps: collecting a part of distance information through the low rank property of a node Euclidean distance matrix to restore a relatively complete node Euclidean distance matrix via the matrix completion theory; and then calculating a conversion matrix for converting a relative coordinate into a real coordinate by using the classical multidimensional scaling mapping algorithm according to the relationship between a real position coordinate and a relative position coordinate corresponding to the real position coordinate of an anchor node, and converting the relative position coordinate of an unknown node into the real position coordinate. In a Euclidean distance matrix completion process, the regularization technique is imported to model the Euclidean distance matrix restoration problem as a norm regular matrix completion problem, and then the norm regular matrix completion problem is solved by using the alternating direction multiplier method. By adoption of the method, the workload of constructing the Euclidean distance matrix can be reduced, and positioning precision higher than similar methods is obtained in all kinds of noise scenes.

The invention discloses an image completion method based on adaptive rank estimation Riemannian manifold optimization. The method is characterized by on the basis of an L1 norm regularization matrix completion algorithm (L1MC), introducing an adaptive rank estimation penalty term and a Riemannian manifold of a fixed rank; using a penalty item R; correcting an iterative optimization function f (R)and a sequence of the matrix determined by an empirical formula continuously to estimate the sequence Rq of the missing matrix, and utilizing Rq to construct a Riemannian manifold of a fixed rank formatrix completion. In the process, a soft threshold algorithm is avoided, the operation rate is increased, and the operation space is saved. Riemannian manifold optimization is applied, the rank is limited in a range of R < = Rq to find an estimation value closest to an original matrix, and the introduction of a shrinkage operator for reducing the rank in each iteration is avoided, so that the calculation efficiency and the image recovery rate are greatly improved. Besides, the algorithm uses the convolutional neural network to preprocess and construct the low-rank matrix, so that the missingvalue is effectively estimated in a short time, the image recovery process is accelerated, and the image recovery accuracy is improved.

The invention discloses an image splicing method based on projection geometry and SIFT characteristic matching point pairs. According to the method, firstly, two images with overlapping areas are captured, SIFT characteristic points are extracted from to-be-spliced images, characteristic point matching is carried out through the K-D tree search algorithm, the RANSAC algorithm is further utilized to carry out characteristic point purification to eliminate wrong matching point pairs, a transformation matrix is calculated if the number of characteristic matching point pairs after purification isgreater than 8, if the number of characteristic matching point pairs after the purification is less than 8, the corresponding number of projection matching point pairs are extracted according to the known overlapping areas of the two images to complete the 8 pairs of matching points, a transformation matrix is calculated, image registering is completed, image fusion is carried out through employing the multi-resolution analysis method for the images after registering, and lastly, the spliced image is outputted. The method is advantaged in that a problem of image registration failure due to thefewer pairs of characteristic matching points can be solved through image splicing, and the image splicing effect is excellent.

A computing cluster operated according to a resource allocation policy based on a predictive model of completion time. The predictive model may be applied in a resource control loop that iteratively updates resources assigned to an executing job. At each iteration, the amount of resources allocated to the job may be updated based on of the predictive model so that the job will be scheduled to complete execution at a target completion time. The target completion time may be derived from a utility function determined for the job. The utility function, in turn, may be derived from a service level agreement with service guarantees and penalties for late completion of a job. Allocating resources in this way may maximize utility for an operator of the computing cluster while minimizing disruption to other jobs that may be concurrently executing.

The invention discloses an approximation optimization and signal acquisition reconstruction method for a 0-1 sparse cyclic matrix, belongs to the technical field for the design and optimization of measurement matrix in compressive sensing, and provides a posteriori optimizing method which is easy to implement by hardware and can ensure signal reconstruction effect, wherein the 0-1 sparse cyclic matrix is adopted in the measurement stage, and a Gaussian matrix is adopted in the reconstruction stage. The method comprises the following steps: orthonormalizing the row vector and unitizing the column vector of the measurement matrix obtained by the i-1th iteration by the ith iteration; and optimizing the 0-1 sparse cyclic matrix by taking the maximum value of the absolute value of the correlated coefficient between each row and column vector, the convergence stability of each row vector module and the row and column number of each row and column subjected to Gaussian distribution as the criteria. The posteriori optimization of the measured data and measured matrix is completed by solving a transition matrix and an approximate matrix. The method establishes the foundation for the compressive sensing to be practical from the theoretical study.

A method and system are proposed to obtain a reduced speckle noise image of a subject from optical coherence tomography (OCT) image data of the subject. The cross sectional images each comprise a plurality of scan lines obtained by measuring the time delay of light reflected, in a depth direction, from optical interfaces within the subject. The method comprises two aligning steps. First the cross sectional images are aligned, then image patches of the aligned cross sectional images are aligned to form a set of aligned patches. An image matrix is then formed from the aligned patches; and matrix completion is applied to the image matrix to obtain a reduced speckle noise image of the subject.

The present invention belongs to the field of network management and network measurement and provides an SDN-based network flow matrix measurement method, which aims to meet the constraint of the flow-aggregation feasibility and the constraint of limited resources for network measurement and lowers the estimation error of a flow matrix. According to the technical scheme of the invention, the flows of the same destination node are aggregated together, and a convergence matrix is designed based on the de-agglomeration and routing rules. The flow matrix is estimated through the two-step process. During the first step, a routing rule (namely called an old rule) composed of most aggregated flows is selected to generate a new rule, so that the difference between the number of aggregated flows in the old rule and the number of aggregated flows in the new rule is minimum. After that, a flow matrix is estimated through the matrix completion technology. During the second step, the flow matrix estimated during the first step is adopted as a basis, and the largest front k flows are selected from the flow matrix to be measured directly. According to the technical scheme of the invention, the measurement period is shortened and the difficulty in directly measuring each flow is reduced. Meanwhile, the constraint of the flow-aggregation feasibility is met, and the estimation error is relatively low.

The invention relates to a recommendation system data completion method based on convex optimization local low-rank matrix approximation. The method includes the steps of 1) constructing a recommendation system data matrix M according to the user's rating of a product in a recommendation system, and representing the data on unrated products by the user as 0 element in M; 2) selecting anchor points, dividing the recommendation system data matrix into a plurality of local matrices using a kernel smoothing method, the number of local matrices being the same as the number of the anchor points; and 3) solving a matrix completion model according to a convex optimization local low-rank matrix approximation algorithm, and completing the 0 element in the matrix M according to the matrix completion model to obtain a recommendation system data matrix X after completion. Compared with the prior art, the invention can finish the recommendation system matrix data completion while guaranteeing the operation speed and the accuracy.

The invention relates to a registration method for an inspection tour map and a reference map based on SURF characteristics. The registration method comprises the following steps: acquiring the inspection tour map and the reference map of a target area; Extracting features in the image by using an SURF algorithm; carrying out Euclidean distance sorting on the feature points after feature extraction, calculating Euclidean distances between all SURF feature points of the inspection tour map and the reference map, and finding out a minimum distance value; Arranging the inspection images subjectedto feature extraction and the reference image in parallel, and performing second screening of registration point pairs near an average value of registration lines; And for the screened matching pointpairs, calculating an affine transformation matrix, and utilizing the matrix to complete registration of the inspection tour map and the reference map. According to the method, the inspection tour map and the reference map are registered by using the SURF features, the SURF operator has scale and rotation invariance, and has very high robustness on image noise, light change, affine deformation and the like, so that accurate feature descriptor extraction can be carried out for the same object in the inspection tour map and the reference map.

The invention discloses a drug repositioning method based on low-rank matrix completion. The method comprises the steps that firstly, relevant data of drugs and diseases is integrated to build a drug-disease heterogeneous network, wherein the elements in a matrix comprise drug pairs, disease pairs, known drug-disease pairs and unknown drug-disease pairs; secondly, the matrix is completed by usinga quick matrix completion algorithm, predicted values are assigned to the unknown drug-disease pairs, and according to the completed predicted values, new indications are predicted for all drugs. Themethod is simple and effective, and through the tests on a plurality of data sets and the comparison with other method, it is shown that the method has good predicting performance in drug repositioning.

An image fusion method based on contrast analysis comprises the following steps of: 1, constructing the morphological gradient graph of an input image and the morphological gradient graph of an enhanced image of the input image separately; 2, subjecting the morphological gradient graphs of the input image and the enhanced image thereof to subtraction to obtain a contrast difference graph of the unenhanced and enhanced input images; 3, creating a fusion coefficient matrix according to the contrast difference graph; and 4, fusing the input image and the enhanced image thereof according to the fusion coefficient matrix, subjecting fused processed image pixels to range control in order that the pixels are within a reasonable image range and are output. The single-graph fusion of the present invention aims at the fusion of the input image and the enhanced image thereof which have the same image information essence and is not fusion of different images. The image output by the image fusion method has uniform grayscale distribution and has completely retained details. The method is effective on not only grayscale images, but also color images, and is higher in fidelity of the restoration of the color images in associated color channels.

The invention discloses an automatic large-scale social image annotation algorithm based on inductive matrix completion, and mainly aims at solving the label deficiency and noise containing problems of a large scale of social images to improve the accuracy of automatic image annotation. According to the method, on one hand, automatic annotation is conducted on newly-added images or label-free images by introducing an inductive matrix completion technology; on the other hand, prediction or error correction is conducted on deficient or noise-containing image labels by integrating the low rank performance of an image label matrix with the inherent sparsity of the image label matrix, and therefore the accuracy of image annotation is enhanced; and furthermore, a question model is solved by adopting a popular optimization solving method in machine learning, and therefore large-scale application is achieved. The accuracy and the scalability of image annotation are well improved.

The invention discloses a compressed sensing imaging device and method based on an entangled two-photon signal, and solves the problem of not high imaging quality under the condition of low sampling number. According to the method, a deterministic determinacy random matrix is loaded in a spatial light modulator of an original quantum imaging device to complete imaging of a target object; and during preparation of entangled light, a telescoping system and a 0-degree incident heat mirror and a low-pass narrow-band filter which are placed along a laser transmission direction among the telescopingsystem are adopted to process laser. The imaging method includes producing a picture for being input to a spatial light modulator; inputting the picture to the spatial light modulator; preparing entangled light; obtaining a matched result value; loading a plurality of sets of modulated observation matrices and obtaining M matched results; constructing a quantum imaging mathematical model; and adopting a compressed sensing algorithm to solve a quantum imaging relational expression to restore an image of the target object. The entangled light prepared by the compressed sensing imaging device and method provided by the invention is high in purity and can well restore the image of the target object under the condition of a low sampling rate.

A feedback application may be executing on a data processing system to transmit a screen update to at least two computers. The feedback application receives an indication of completion from a first computer, in response to transmitting the screen update. The feedback application updates a completion gauge to indicate a partial completion among participants, in response to receiving the indication of completion from the first computer. The feedback application receives a second indication of completion from a second computer, in response to transmitting the screen update. The feedback application updates the completion gauge to indicate a completion among participants, in response to receiving the indication of completion from the second computer.