21 results about "Matrix projection" patented technology

The invention discloses a million-frame-level industrial vision system and method based on event driving and compressed sensing, and the system is characterized in that an event camera imaging module in the system captures the brightness change of each pixel in a field of view of the event camera imaging module in an asynchronous manner, and generates an event containing a pixel coordinate, a timestamp and change polarity for each change; a compressed sensing coding module constructs sparse image vectors for events in a time window, and the sparse image vectors are projected to low-dimensional observation vectors through an observation matrix phi; the sparse image reconstruction module is used for optimizing an objective function through sparse constraint and total variation regularization; a dynamic ROI compression module controls a compression mask function according to the event density and a gradient threshold. The method can break through the limitation of the traditional frame rate, has the advantages of high precision, high efficiency, low power consumption, strong robustness and the like, and has a wide industrial application prospect.

The application discloses a kind of based on random projection's physical property parameter inversion method of potential field data, comprising the following steps: S1: measured potential field data is obtained, according to survey area and depth range is profiled in underground space, sensitivity matrix in inversion is calculated based on potential field data forward theory, and then the forward calculation relationship of full space is established;S2: random projection matrix is designed, and sensitivity matrix is projected to multiple low-dimensional subspace, and the subspace forward calculation relationship is established;S3: based on the regularization equation of subspace forward calculation relationship, and the physical property parameter in subspace is solved using conjugate gradient algorithm;S4: the final physical property parameter inversion result is obtained by the weighted average calculation of multiple physical property parameters in subspace.The physical property parameter inversion method of potential field data based on random projection has higher depth resolution and inversion reliability, and improves the practicability of physical property inversion method in actual data processing.

The application discloses a target screening method and system based on a low-rank sparse joint tensor of an enhanced latent space, relates to the technical field of target screening, and comprises the following steps: introducing a projection matrix to project a first matrix into a latent representation matrix, and performing separation of a noise matrix once to obtain a second matrix and a first noise matrix; performing self-representation on the second matrix in a latent space, and performing separation of a noise matrix twice to obtain a self-representation matrix of multi-view data and a second noise matrix; performing separation of a block diagonal and a non-block diagonal on the self-representation matrix, and constructing a block diagonal tensor and a non-block diagonal tensor; respectively calculating norms of the noise matrix, the block diagonal tensor and the non-block diagonal tensor, and performing weighted summation on the respective norm results to obtain a target function; solving the target function by using an alternating direction multiplier method; and obtaining a multi-view self-representation coefficient matrix when the solution result of the target function is the minimum value.

The present application belongs to the field of computer vision, artificial intelligence and intelligent security technology, and specifically relates to an unsupervised visible light-infrared personnel re-identification method based on cyclic pair identity learning. In the training stage, a backbone network is used to extract multi-modal features from pedestrian images, and a cross-modal similarity matrix is calculated according to the multi-modal features. The cross-modal similarity matrix is projected into a pair matrix through a pair matrix projection network. A multi-modal relationship matrix is calculated by fusing the pair matrix and the cross-modal identity matching information. An invertible mapping model is used to align the modes, and the network and model are trained using loss in the training stage. In the identification stage, the cosine similarity of the query image feature set and all features in the gallery feature set is calculated, and the images corresponding to the top-ranked features are taken as the cross-modal identity re-identification results of the query image. The present application can realize accurate matching and identity re-identification of visible light and infrared modal pedestrian images under unsupervised conditions.

The invention provides a nuclear power embedded part intelligent detection method based on data fusion, belongs to the technical field of intelligent detection, is based on multi-source data fusion of 2D images and 3D point clouds, combines AI deep learning and homography matrix projection transformation technologies, and aims to break through the technical bottleneck of three-dimensional positioning and accurate measurement of embedded parts in complex scenes. A high-precision sub-pixel angular point extraction algorithm based on Hough straight line fitting is designed, and precise recognition of the edge of the embedded part is achieved; high-precision data fusion of the 2D image and the 3D point cloud is realized through joint calibration parameters of the camera and the laser radar; accurate conversion from 2D image angular points on the surface of the embedded part to 3D space coordinates is realized based on homography matrix projection conversion; according to the method, the BIM design drawing data is combined, automatic calculation and evaluation of the installation deviation of the embedded part are achieved, the detection precision and efficiency are improved, the detection cost is reduced, and the detection safety is guaranteed.

The invention relates to the technical field of industrial automatic control and fluid surface treatment, in particular to an automatic test panel spraying system based on PLC control. Comprising a virtual feature analysis module which is used for acquiring three-dimensional digital model data and a preset spraying track of a virtual target curved surface and constructing a virtual geometric feature data set containing curvature features of discrete points and a normal vector field; the equivalent flux calculation module is used for generating a target three-dimensional flux distribution data matrix; the projection mapping compensation module is used for generating a composite compensation instruction sequence containing dynamic attitude deflection data, variable-speed motion data and variable-fan-amplitude air pressure data; and the execution driving control module is used for converting the digital instruction into a multi-axis servo driving signal and an analog quantity adjusting signal, driving an execution mechanism to execute a time-space coordination action on a physical plane base material, and reproducing fluid deposition characteristics of a virtual target curved surface. According to the system, the consumption of trial and error materials is greatly reduced, and the conversion time from a laboratory to a production line is shortened.

Computer-implemented method for encoding a list of minutiae of a dactylogram, said method taking, as input data, the coordinates associated with each minutia from a list of minutiae of a dactylogram, and providing, as output datum, a fixed-size encoding vector which is representative of the list of the minutiae of said dactylogram, the method comprising the following steps:(a) concatenating the coordinates of each minutia from the list of minutiae in the form of a source matrix of dimension;(b) projecting the source matrix of dimension into a space of a dimension greater than the dimension of said source matrix using a projection model previously trained to form a projected matrix of dimension;(c) inferring an inference matrix of dimension by applying, to the intermediate matrix, a previously trained graph neural network;(d) aggregating the values of the inference matrix into a fixed-size vector using a previously defined aggregation model, said fixed-size vector being the fixed-size encoding vector which is representative of the list of the minutiae of the dactylogram.

The invention belongs to the technical field of computer vision, artificial intelligence and intelligent security and protection, and particularly relates to an unsupervised visible light-infrared person re-identification method based on cyclic pairwise identity learning, and the method comprises the steps: extracting multi-modal features from a pedestrian image through employing a backbone network in a training stage, calculating a cross-modal similarity matrix according to the multi-modal features, and carrying out the recognition of the cross-modal similarity matrix; projecting a cross-modal similarity matrix to a pairwise matrix through a pairwise matrix projection network, calculating a multi-modal relation matrix by fusing the pairwise matrix and cross-modal identity matching information, aligning modals by using a reversible mapping model, and training the network and the model by using loss in a training stage; in the recognition stage, the cosine similarity of all features in the query image feature set and the image library feature set is calculated, and an image corresponding to the feature ranked in the front is taken as a cross-modal identity re-recognition result of the query image; according to the method, visible light and infrared modal pedestrian image accurate matching and identity re-identification can be realized under the unsupervised condition.

The invention provides a navigation method for adaptively optimizing filtering. The method comprises the following steps: respectively establishing error models of an inertial navigation system (SINS) and a Doppler velocimeter (DVL) by utilizing Lie group geometric characteristics; transmitting the output data of the SINS and the DVL into an adaptive Kalman filter; a Kalman filter framework is established according to a system model, and a state error and a noise covariance matrix are projected to a Lie group SE2 (3) to be converted into a low-dimensional matrix. When the DVL data is available, a process noise covariance matrix and a measurement noise covariance matrix are explored through DQN training; and when the DVL data is lost, independently optimizing the process noise covariance matrix and the measurement noise covariance matrix by using the trained model. According to the technical scheme provided by the invention, the noise covariance matrix of projection conversion under the Lie group is optimized through the deep reinforcement learning method, Kalman filtering can be adaptively optimized, and high-precision and high-real-time underwater navigation is achieved.

The application provides a virtual reality-based display method, device and computer readable medium. The method applied to a first terminal device comprises: obtaining a center point coordinate and four corner point coordinates of a real screen; obtaining position information and angle information of a real camera relative to the real screen; wherein the field of view angle of the real camera covers the real screen; determining a center point coordinate and a default matrix of a virtual screen according to the position information and the angle information, and determining a projection matrix and a view matrix of a virtual camera; determining a position coordinate of the virtual camera relative to the center point of the virtual screen according to the default matrix, the projection matrix and the view matrix; determining an offset value according to the position coordinate, the center point coordinate of the real screen and the center point coordinate of the virtual screen; performing calibration processing on the position coordinate according to the offset value; and determining a target image according to the four corner point coordinates and the result of the calibration processing, so as to reduce the error of the position offset of the picture displayed by the extended screen.

The invention provides an SF6 equipment anomaly detection method, which comprises the steps of vectorization feature extraction, anti-leakage model training and incremental updating, and is characterized in that the feature extraction efficiency is improved through parallel computing based on a memory view, data leakage is avoided based on statistic physical isolation, anti-forgetting updating is realized based on parameter space decoupling, and the detection accuracy is improved. And efficient and accurate anomaly detection of the SF6 equipment is cooperatively realized. According to the method, the computing power requirement of mass data trend analysis is reduced by utilizing the memory view and the matrix projection operator, a physically isolated training pipeline is constructed to eliminate data leakage, and a parameter decoupling strategy is designed to realize anti-forgetting increment updating.

The invention relates to a method for constructing ultrasonically a confocal image of an object contained in a medium, the method comprising the following steps: a) acquiring a canonical reflection matrix Rui(t) by insonifying the object via a scattering screen present in the medium; b) determining a focused reflection matrix R' ξξ (ω) by projecting the canonical reflection matrix into a spatio-frequential basis ξ: (I); c) determining a corrected reflection matrix R'' ξξ (ω); d) determining a reflection matrix R xx (zM) associated with the object by projecting the corrected reflection matrix R'' ξξ (ω) into a plane x of the object, e) constructing an image from the object reflection matrix R xx (zM). The spatio-frequential basis ξ ensures an independence between the spatial and sequential data and makes it possible to demonstrate a chromato-angular memory effect associated with the field reflected by the object. The scattering screen makes it possible to increase the base resolution of the acquisition probe.

Various aspects relate generally to wireless communication and more particularly to beamforming techniques in wireless communication networks. Some aspects more specifically relate to beamforming techniques for uplink MU-MIMO communications, and the techniques can include UL MU-MIMO precoding using per-STA feedback. In some examples, an AP can support STA-side precoding for UL MU-MIMO communications by providing STAs with feedback representative of unitary matrices that correspond to block diagonal components of a matrix projection of channel matrices for the STAs in accordance with a linear equalizer. In some examples, the AP can trigger the STAs to transmit sounding packets concurrently, but can use the block diagonal matrix projection to process the various STAs' channels independently.

The application relates to a flow timing model continuous learning method, device and equipment based on orthogonal rotation. The method comprises the following steps: locking a pre-training timing model deep layer parameter, sampling a sub-data set from each time window of historical traffic data, and inputting the sub-data set into a shallow layer multiple times for light gradient updating according to time; after each updating, a key matrix is extracted and spliced, projected to a shallow layer weight matrix row space, decomposed to obtain an input orthogonal matrix; a covariance matrix is constructed, decomposed to obtain an output orthogonal matrix, and shallow layer weight updating is iterated until a condition is met, so that a congestion prediction model is obtained; and real-time traffic data is input to obtain a future road condition result. The method does not damage an original feature space, realizes continuous learning with high efficiency and low consumption, and the model is accurate in prediction.

The application discloses an overlapping speech separation method and system based on feature space orthogonal projection, a device and a medium, and relates to the technical field of speech recognition. The method comprises the following steps: real-time detection of a collected speech sequence is performed through a pre-trained segmentation model, speech active segments and non-active segments are divided, and multi-speaker overlapping speech segments are accurately located. For the overlapping segments, the identity of the main speaker is determined and the voiceprint features thereof are extracted by using context tracking information, and a known speaker feature subspace is constructed. Then, the mixed voiceprint features of the overlapping speech are extracted, the mixed voiceprint features are decomposed into a subspace parallel component and an orthogonal vertical component through matrix projection operation, so that residual features without the main speaker information are obtained, the cosine similarity with candidate speakers is calculated, the speaker with the highest similarity is selected as the secondary speaker, and the logical separation of the overlapping speech is realized. The known speaker features are eliminated through mathematical projection operation, and unknown speakers can be accurately identified in the residual space.

The invention provides a data order reduction method and equipment for a digital twin model of power grid equipment. According to the implementation scheme, a sampling data matrix is constructed based on a temperature sampling data set of power grid equipment; carrying out main mode interception on the sampling data matrix and each sub-matrix to obtain a global basis and each local basis, carrying out combination and orthogonalization on the global basis matrix and the local basis matrix to obtain an orthogonalized mode basis matrix, projecting the sampling data matrix to the orthogonalized mode basis matrix, and carrying out singular value decomposition to obtain a singular value decomposition matrix; obtaining a first orthogonal matrix, a first singular matrix and a first transpose matrix; and performing main modal interception on the first orthogonal matrix based on the orthogonalization modal basis matrix to obtain a second orthogonal matrix, performing singular value reconstruction on the second orthogonal matrix, the first singular matrix and the first transpose matrix to obtain a reduced-order temperature data matrix, and performing prediction through a power grid equipment digital twin model to obtain a fault prediction result of the power grid equipment. According to the invention, the state prediction efficiency and accuracy of the power grid equipment can be improved.

This invention provides an intelligent calculation method and system for the spatial resolution of X-ray digital images, comprising: generating a target image using a bilinear image quality meter (BMI) with a X-ray detector; matching the target image to the BMI specifications and retrieving a template image of the same specifications; calculating the homography matrix between the template image and the target image; projecting and mapping the extreme point combinations of each line pair of the BMI in the template image to calculate the average modulation degree of all line pairs of the BMI in the target image; constructing a continuous relationship curve corresponding to the wire diameter of the average modulation degree, thereby obtaining the spatial resolution of the X-ray digital image, and simultaneously displaying all calculation processes and results. This invention can actively identify and match different standard BMI types, eliminate manual operation in the calculation process, avoid human error factors, and automatically calculate and output accurate spatial resolution results. It features intelligent calculation, high accuracy, high confidence, good transferability and scalability, and strong robustness.

The application discloses a skill label generation method, device and equipment of a knowledge tracking model, comprising the following steps: constructing a knowledge tracking model, inputting interactive exercises and answer results into the knowledge tracking model; based on a skill label matrix, the comprehensive codes corresponding to the interactive exercises and the answer results are respectively obtained in the form of matrix projection, and the monotone self-attention mechanism processing is respectively performed to obtain the cross codes corresponding to the interactive exercises and the answer results, and then the evaluation result vector of the student knowledge state is obtained; based on the evaluation result vector of the student knowledge state and the comprehensive code of the current interactive exercise, the prediction result of whether the answer result is correct or not is obtained; the loss of the prediction result relative to the true answer result is obtained, the skill label matrix and the knowledge tracking model are trained; the skill label matrix after training is subjected to binaryzation processing, and the skill label result is stored. The application realizes high-performance and full-automatic labeling of the skills associated with the knowledge tracking exercises.

The application discloses a high-energy flash X-ray image nonlinear reconstruction method based on MCMC. According to the high-energy flash X-ray imaging principle, a discrete form of a nonlinear forward model is constructed, and a corresponding Jacobian matrix form is derived, the solution and uncertainty quantification of the inverse problem are considered in combination with the Bayesian theory, a hyperparameter based on weak information prior is introduced to construct a nonlinear hierarchical Bayesian model. By accelerating the solution of the optimization problem of random disturbance to sample the conditional distribution, the solution of the optimization problem is constrained in combination with the Jacobian matrix projection, and the proposal distribution of the target parameter is designed to reduce the sample statistical bias. Under the minimum variance criterion, the sample values of the linear and nonlinear Bayesian models are fused to obtain the final reconstructed image. The application can improve the sample estimation efficiency while ensuring that the reconstructed result presents clear edges and high precision.

The invention relates to the technical field of power grids, and discloses a time sequence prediction method based on frequency domain processing and a relational attention network. And performing power load prediction through the prediction model, wherein the method comprises the following steps: taking a time sequence of power data with multiple power load variables as input data; applying a frequency domain preprocessing module to the input data to obtain a frequency-enhanced power data time sequence; inputting the frequency-enhanced power data time sequence into a double-self-attention prediction module after passing through an embedded layer, wherein the double-self-attention prediction module comprises a standard dot product attention branch and a tilt attention branch; the oblique attention branch calculates an anti-symmetric score of a query-key pair through a learnable anti-symmetric matrix projection; fusing outputs of the standard dot product attention branch and the oblique attention branch to generate an attention weight; and generating a power load prediction result based on the attention weight. The prediction model provided by the invention obtains leading performance in a long-term prediction task.

The invention relates to a streaming time sequence model continuous learning method, device and equipment based on orthogonal rotation. The method comprises the following steps: locking deep parameters of a pre-training time sequence model, sampling a sub-data set from each time window of historical traffic data, and inputting shallow-layer lightweight gradient updating for multiple times according to time; extracting and splicing a key matrix after each update, projecting the key matrix to a shallow weight matrix row space, and decomposing to obtain an input orthogonal matrix; constructing a covariance matrix, decomposing the covariance matrix to obtain an output orthogonal matrix, and updating shallow weight iteration to a condition to obtain a congestion prediction model; and inputting real-time traffic data to obtain a future road condition result. The method does not damage the original feature space, realizes continuous learning with high efficiency and low consumption, and is accurate in model prediction.