Patents
Literature
Patsnap Eureka AI that helps you search prior art, draft patents, and assess FTO risks, powered by patent and scientific literature data.

53 results about "Graph fourier transform" patented technology

Intelligent monitoring system for municipal drainage pipe network

The invention discloses an intelligent monitoring system for a municipal drainage pipe network, and particularly relates to the technical field of drainage pipe network monitoring. The node operation mode identification module carries out real-time classification and confidence evaluation on the operation state of the pipe network, constructs a multi-attribute pipe network weighted graph based on pipe diameter difference, gradient and confluence density, and extracts multi-scale features through graph Fourier transform. A hybrid anomaly detection link is constructed in combination with an LSTM self-encoder, an isolated forest model and chemical oxygen demand and turbidity water quality verification, and the problems that traditional single-index monitoring is prone to false alarm and missing alarm and inaccurate in positioning are solved; sensor data compensation is realized through cooperation with digital twinning, a rapid detection mode is started during rainstorm early warning, key nodes are processed preferentially, and drainage scheduling is controlled in a closed-loop mode; and target nodes which are easy to accumulate grease are screened based on pipe network topology connectivity, accumulation risks are predicted by fusing multi-sensor data, and a preventive clearing instruction is triggered.
Owner:JIAXING JIAYUAN TESTING TECH SERVICE CO LTD

Hybrid wireless eye movement tracking method and system based on multi-sensor fusion, electronic equipment and storage medium

The invention provides a hybrid wireless eye movement tracking method and system based on multi-sensor fusion, electronic equipment and a storage medium, and relates to the technical field of eye movement tracking. The method comprises the following steps: firstly, acquiring various data by using various different types of sensors such as a Q-Var sensor, an inertial measurement unit and a high-speed panoramic camera; secondly, preprocessing the various sensor data by means of graph Fourier transform, a Kalman filter, a spectral clustering technology and the like; fusing the preprocessed sensor data based on a graph theory method; and finally, performing fixation track prediction based on a convolutional neural network and a gating circulation unit. According to the technical scheme, the performance of eye movement tracking under different environment conditions is improved, the eye movement tracking precision is higher, and the stability is better. In addition, the technology of the application also uses a 5G millimeter wave technology or an IEEE 802.11 ac protocol to transmit gaze tracking data, thereby achieving the purpose of bandwidth enhancement and low-delay data transmission.
Owner:南通诺瞳奕目医疗科技有限公司

Unmanned aerial vehicle confrontation action classification method based on SlowFast-graph Fourier transform

The invention provides an unmanned aerial vehicle confrontation action classification method based on SlowFast-graph Fourier transform, and the method comprises the steps: obtaining the training graph data of an unmanned aerial vehicle, carrying out the preprocessing of the training graph data, dividing the graph data into high-frame-rate graph data and low-frame-rate graph data, thereby giving full play to the advantages of different-frame-rate data in feature capture, and achieving the recognition of the confrontation actions of the unmanned aerial vehicle. The perception and decision-making capability of the unmanned aerial vehicle to a complex environment is improved; respectively inputting two data sequences with fast and slow differences into a SlowFast-graph Fourier transform network, and fusing features generated by a fast path into a slow path by adopting a one-way transverse connection mode so as to realize effective fusion and interaction of the features; and inputting the fused features into an action classification network module to realize accurate classification operation of the actions of the unmanned aerial vehicle.
Owner:WUXI UNIV

Wind power prediction method based on frequency domain space-time diagram convolutional network

The invention provides a wind power prediction method based on a frequency domain space-time diagram convolutional network. The method comprises the following steps: firstly, respectively constructing a spatial proximity weight matrix and a similarity matrix based on a fan position and historical feature data, and adaptively fusing the spatial proximity weight matrix and the similarity matrix into an adjacent matrix; performing high-dimensional embedding on the normalized features, and converting a time domain signal to a graph frequency domain by using graph Fourier transform GFT; designing a prediction and reconstruction double-branch network, extracting key frequency features through discrete Fourier transform (DFT), causal convolution and channel attention, and performing inverse transformation reduction; and finally, constructing a joint loss function of prediction and reconstruction loss weighting, and realizing end-to-end training through back propagation and an Adam optimizer. According to the method, spatial topology and time-frequency dynamic characteristics are fused, prediction and reconstruction advantages are combined, and the precision and robustness of wind power prediction are improved.
Owner:HUNAN UNIV

Skeleton data processing method based on graph Fourier transform and space-time synchronous graph convolutional network

The invention provides a skeleton data processing method based on graph Fourier transform and a space-time synchronization graph convolutional network, which comprises the following steps: converting skeleton data to a frequency domain through graph Fourier transform, then extracting space-time characteristics by using the space-time synchronization graph convolutional network, and finally carrying out data smoothing processing through memory playback and Kalman filtering. Therefore, the generalization ability and the recognition precision of the skeleton data are improved.
Owner:WUXI UNIV

Water supply network pressure estimation method based on second-order graph signal reconstruction

The invention discloses a water supply pipe network pressure estimation method based on second-order graph signal reconstruction, and relates to the technical field of water supply pipe network reconstruction, and the method comprises the steps: obtaining water supply pipe network information, constructing a first pipe network graph Laplacian matrix, and carrying out the spectrum analysis to obtain a spectrum base; the method comprises the following steps: presetting a high-frequency term of a graph Laplacian frequency spectrum on a frequency spectrum base, limiting to obtain a low-frequency term of the graph Laplacian frequency spectrum, reconstructing a first pipe network water head on the basis of inverse graph Fourier transform on the low-frequency term of the graph Laplacian frequency spectrum, and obtaining a pseudo monitoring value; and constructing a second pipe network graph Laplacian matrix according to the pseudo monitoring value and the actual monitoring value, and repeating the reconstruction process to obtain a second pipe network water head. The innovation lies in that through a double constraint framework, the low-frequency component and the measured data covariance matrix are fused while the pipe network topological characteristics are reserved, and the problems that a traditional method is sensitive to parameters, the hydraulic relation is ignored and the like are effectively solved.
Owner:HEFEI UNIV OF TECH

Complex pavement mechanical response monitoring device and method

The invention relates to the technical field of mechanical response monitoring, in particular to a complex pavement mechanical response monitoring device and method, and the method comprises the steps: collecting sensor data of each section of pavement of a road; the method comprises the following steps: constructing a graph structure by utilizing the correlation among mechanics, heat and humidity among sensors, and mapping the graph structure into a graph frequency domain feature vector by utilizing graph Fourier transform to form a hypergraph space; constructing an objective function based on the spatial proximity and the time continuity of the graph frequency domain feature vector in the hypergraph space so as to perform global optimization on the graph frequency domain feature vector by using a particle swarm optimization algorithm; and inversely transforming the optimized graph frequency domain feature vector into a data space, generating a sensor signal coupling curve, and identifying road surface damage according to abnormal change of the curve, thereby realizing real-time monitoring of the road surface health state. The invention aims to improve the accuracy and real-time performance of road surface health state monitoring.
Owner:HEBEI ZHUANYE CONSTRUCTION ENGINEERING CO LTD

A power distribution network data compression and reconstruction method based on graph transformation compression sensing

This invention discloses a method for compressing and reconstructing distribution network data based on graph transform compressed sensing, belonging to the field of distribution network measurement data technology. The method includes the following steps: inputting the network topology and historical operating data of the distribution network; constructing a power-voltage relationship parameter matrix; determining the number of principal components and obtaining the projected principal component coefficient matrix; calculating the graph Fourier transform basis; using the graph Fourier transform basis as the sparse basis after projecting the principal component coefficient matrix, calculating the sparsity and initial compression of each principal component under the sparse basis, and obtaining spatially compressed observations; reconstructing the voltage amplitude measurement data of distribution network nodes, and further reconstructing the active power measurement data and reactive power measurement data of branches. This invention employs the above-mentioned method for compressing and reconstructing distribution network data based on graph transform compressed sensing, achieving high data reconstruction accuracy while maintaining a high compression ratio, reducing computational load, and making it more suitable for applications on edge-side devices.
Owner:TIANJIN UNIV

Photovoltaic array insulation state real-time monitoring method based on distributed optical fiber sensing

The invention discloses a photovoltaic array insulation state real-time monitoring method based on distributed optical fiber sensing, particularly relates to the technical field of power cable fault detection, and is used for solving the problem that a plurality of independent insulation defect points in a cable accessory are difficult to effectively distinguish and position from a fused surface temperature field in the prior art. The method comprises the following steps: acquiring temperature distribution data measured by distributed optical fibers along a cable line, identifying a temperature abnormal region, mapping the temperature data of the abnormal region into a graph signal, performing graph Fourier transform to obtain spectral domain representation, analyzing statistical independence among different frequency components in a spectral domain, and calculating the spectral domain representation. And if the possibility exists, identifying candidate positions of potential defect points in a spatial domain through inverse transformation according to spectral domain characteristics related to statistical independence, and carrying out spatial clustering and center calculation on the candidate positions, so as to determine whether a multi-heat-source composite fault exists in the attachment. And a plurality of potential independent insulation defect points in the cable accessory can be identified and accurately positioned.
Owner:TALUGTUG NEW ENERGY JOINT CO

Spectral compression for dynamic mesh encoding

Apparatuses and methods are disclosed for encoding and decoding mesh data. Encoding techniques are disclosed including coding a mesh into a bitstream. The coding includes generating a base mesh from the mesh, obtaining connectivity and geometry data of the base mesh. Then, subdividing the base mesh, obtaining connectivity and geometry data of the subdivided mesh. Coding proceeds by generating GFT coefficients, based on a Graph Fourier Transform (GFT), using displacement data, and then, coding into the bitstream the coefficients and connectivity data of the base mesh. Decoding techniques are disclosed including decoding the mesh from the bitstream. The decoding includes decoding from the bitstream connectivity data of the base mesh and subdividing the base mesh, obtaining connectivity data of the subdivided mesh. Decoding proceeds by decoding from the bitstream the GFT coefficients and reconstructing the mesh based on the decoded connectivity data of the subdivided mesh and the decoded coefficients.
Owner:INTERDIGITAL CE PATENT HOLDINGS SAS

An adaptive reconstruction method for band-limited graph signals with unknown signal support

The present invention discloses an adaptive reconstruction method for a band-limited graph signal with unknown signal support. The method comprises: constructing a corresponding observation graph signal model based on an observation signal affected by impulse noise; establishing an optimization problem model based on the observation graph signal model using the l2 norm and the l0 norm to describe the recovery problem; solving the optimization problem model using an alternating optimization method to obtain a reconstructed signal; and performing an inverse graph Fourier transform on the reconstructed signal to obtain a reconstructed graph signal. The present invention reconstructs observation signals affected by impulse noise by exploiting the sparsity of the graph signal in its frequency domain. This method can effectively process graph signals affected by impulse noise, and even when the signal support is unknown, the reconstructed signal still has very high accuracy.
Owner:NANJING UNIV OF POSTS & TELECOMM

Heart rate anomaly classification method based on transfer learning and dynamic graph convolution

The invention belongs to the technical field of electrocardiosignal detection methods, and particularly relates to a heart rate anomaly classification method based on transfer learning and dynamic graph convolution, which comprises the following steps: collecting electrocardiosignal data and preprocessing the data; performing feature extraction on the electrocardiosignal, and converting the electrocardiosignal into a multi-channel feature matrix; adopting a transfer learning strategy, training a Bi-LSTM network hyper-parameter on the basis of the PTB-XL data set, and migrating to the model to process a feature vector; on the basis of the physiological space relation obtained by the lead signals, a graph structure of the twelve-lead electrocardiosignals is constructed, and a feature matrix serves as node information content; performing graph Fourier transform on the graph structure by adopting spectrogram filtering, and simplifying calculation of graph convolution by utilizing approximate representation of a Chebyshev polynomial; and dynamically updating the adjacent matrix in the prediction process by using a cross entropy function, and further updating the hyper-parameter of the Bi-LSTM by using a Focal Loss function.
Owner:SHENZHEN JINGHAO TECHNOLOGY CO LTD

Graph signal frequency-domain motion compensation for point cloud attributes coding

Systems and methods are disclosed herein for temporally predictive coding of three-dimensional (3D) dynamic point cloud attributes. A first frame and a second frame of point cloud data are accessed. The point cloud data points include 3D spatial coordinates and one or more graphic attributes. A block tree data structure comprising a plurality of blocks is generated based on a tree partitioning of the second frame of point cloud data. Matching block pairs between the first frame and the second frame are identified from the plurality of blocks based on block-wise searching. Frequency-domain projections are generated for each matching block pair via a graph Fourier transform (GFT) algorithm. A bitstream of motion-compensated residuals is generated based on differences in the frequency-domain projections for each matching block pair.
Owner:ADEIA GUIDES INC

Graph signal frequency-domain motion compensation for point cloud attributes coding

PendingUS20260080576A1Image codingGraphicsPoint cloud
Systems and methods are disclosed herein for temporally predictive coding of three-dimensional (3D) dynamic point cloud attributes. A first frame and a second frame of point cloud data are accessed. The point cloud data points include 3D spatial coordinates and one or more graphic attributes. A block tree data structure comprising a plurality of blocks is generated based on a tree partitioning of the second frame of point cloud data. Matching block pairs between the first frame and the second frame are identified from the plurality of blocks based on block-wise searching. Frequency-domain projections are generated for each matching block pair via a graph Fourier transform (GFT) algorithm. A bitstream of motion-compensated residuals is generated based on differences in the frequency-domain projections for each matching block pair.
Owner:ADEIA GUIDES INC

Graph data compression method based on linear canonical transformation of two-dimensional graph

The invention discloses a graph data compression method based on linear canonical transformation of a two-dimensional graph, and belongs to the technical field of digital data processing. According to the method, the underlying structure features of the to-be-compressed data are utilized, the to-be-compressed data are modeled into the two-dimensional diagram data on the Cartesian product diagram, the two-dimensional diagram data are distributed to two different dimensions, representation of the to-be-compressed data is achieved, and local structure information of the to-be-compressed data is fully utilized. According to the method, the linear regular transformation of the two-dimensional diagram is defined through the definition of the Fourier transformation of the diagram and the matrix decomposition property of the linear regular transformation, the flexibility of the linear regular parameters of the two-dimensional diagram is utilized, so that the transformation captures the local features and detail information of the data of the two-dimensional diagram, and the compression and processing are performed in the linear regular domain of the two-dimensional diagram; and effective compression of two-dimensional diagram data under a low compression ratio is realized. According to the method, parameter optimization and selection of linear regular transformation of the two-dimensional diagram are carried out through the normalized root mean square error and the peak signal-to-noise ratio, so that different application scenes are matched, and the compression performance of the two-dimensional diagram is further improved.
Owner:BEIJING INST OF TECH

Spectrum compression for dynamic grid coding

An apparatus and method for encoding and decoding grid data are disclosed. Encoding techniques are disclosed, including encoding a mesh into a bitstream. The encoding includes generating a base grid from the grid, obtaining connectivity and geometry data for the base grid. Then, the basic grid is subdivided, and connectivity and geometric data of the subdivided grid are obtained. The encoding is performed by generating graph Fourier transform (GFT) coefficients based on GFT using the displacement data, and then encoding connectivity data and coefficients of the base grid into the bitstream. Decoding techniques are disclosed, including decoding a mesh from a bitstream. The decoding includes decoding connectivity data of the base mesh from the bitstream and subdividing the base mesh to obtain connectivity data of the subdivided mesh. The decoding is performed by decoding the GFT coefficients from the bitstream and reconstructing the mesh based on the decoded connectivity data and the decoded coefficients that subdivide the mesh.
Owner:INTERDIGITAL CE PATENT HOLDINGS SAS

A Point Cloud Attribute Compression Method Based on Inter-Block Prediction and Graph Fourier Transform

The present invention proposes a point cloud attribute compression method based on inter-block prediction and graph Fourier transform to solve the problem of improving compression performance while maintaining the "bit rate-distortion" balance. First, clustering is performed on the point cloud to divide the point cloud into different point cloud blocks. The present invention will encode the point cloud attribute information based on the point cloud blocks. Specifically, the color attribute of the point cloud block after region division is split into two parts: the attribute mean and the attribute residual. According to the different characteristics of the two parts of information, they are processed separately. For the attribute residual information, the present invention proposes a graph Fourier transform method to further enhance the sparsity of the transform coefficients, thereby obtaining better compression performance. For the attribute mean information, according to the correlation between the attribute means of adjacent point cloud blocks, prediction is used to eliminate the inter-block redundancy. Then, the attribute mean and the attribute residual after transformation and prediction processing are encoded and decoded, and reconstructed based on inverse transformation and inverse prediction.
Owner:BEIJING UNIV OF TECH

Temperature online reconstruction method of adaptive graph signal based on maximum dustpan line criterion

PendingCN121580000ATime domainAlgorithm
The invention relates to the technical field of signal processing, and particularly discloses an adaptive graph signal temperature online reconstruction method based on a maximum dustpan line criterion, which comprises the following steps: taking a meteorological station as a node of a graph, taking temperature as a signal value of the node on the graph, and constructing a graph signal model according to an observation signal influenced by impulse noise; based on the graph signal model, establishing an optimization model by using a maximum dustpan line function criterion and a minimum mean square algorithm and using an L2 norm and an L1 norm; constructing an MGLMS robust algorithm based on the optimization model, and converting the MGLMS robust algorithm into a time domain update equation by using inverse Fourier transform of a graph; and performing on-line reconstruction on the signal values of the nodes on the graph by using the time domain update equation to obtain reconstructed graph signals, thereby obtaining the temperature on the unknown node. According to the adaptive graph signal temperature online reconstruction method based on the maximum dustpan line criterion, data reconstruction and processing of a GSP in an impulse noise environment are improved.
Owner:GUANGXI UNIV

3D gaussian splatting data compression

Post training compression of 3DGS data is agnostic to training in a traditional signal compression perspective. Gaussian parameters are treated as signals. Pre-processing and transform coding techniques are used to compress the signals effectively. Firstly, lossless / lossy compression is performed on 3DGS geometry (positions, scales, rotations) using a point cloud coding-based (e.g., G-PCC, GeS) framework. Positions are compressed using occupancy tree coding. Scales and rotations are encoded as attributes using transform coding. The widely used block-based graph Fourier transform (GFT) is used to compress the attributes (base colors, spherical harmonic coefficients and opacities). In addition, a graph construction strategy is used for 3DGS data that computes the edge weights based on similarity (or dissimilarity) between the 3D Gaussian distributions using KL-divergence. Alternatively, positions can be encoded using occupancy tree (e.g., G-PCC, GeS) or AI-based PCC methods, and any subset of Gaussian parameters or the transformed coefficients of Gaussian parameters can be mapped into 2D frames and encoded by video coders.
Owner:SONY GROUP CORP +1

Human body falling behavior intelligent identification method and system based on monitoring video

The invention relates to the field of fall identification, in particular to a human body fall behavior intelligent identification method and system based on a monitoring video, and the method comprises the steps: extracting human body key points, calculating a detection variance to generate a local signal-to-noise ratio weight, and mapping a space coordinate containing a vertical projection component through a perspective matrix; fusing the signal-to-noise ratio, the vertical projection and the Euclidean distance to construct an anti-noise skeleton topology network; calculating a spectrum structure entropy by using graph Fourier transform, and converting the speed signal to a graph frequency domain; modulating a speed signal by establishing a frequency domain gain function taking a spectrum structure entropy as an independent variable, and calculating a motion imbalance index; and finally, performing multi-dimensional gating correction on the index by combining the potential energy attenuation coefficient and the global confidence coefficient, and judging falling when the correction value exceeds a threshold value. According to the invention, through fusion of frequency domain analysis and physical constraint, environmental noise is effectively suppressed, and the robustness and accuracy of fall detection in a complex scene are significantly improved.
Owner:SHENYANG HUAANXIN TECH CO LTD

Motion coding for dynamic meshes using intra- and inter-frame graph fourier transforms

Apparatuses and methods are disclosed for encoding and for decoding mesh data. Disclosed encoding techniques include receiving a mesh sequence, including geometry data of vertices of meshes in the sequence and coding motion data into a bitstream of coded mesh data. The motion data coding comprises transforming, based on a Graph Fourier Transform (GFT), the geometry data to obtain GFT coefficients representative of the motion data, and then coding the GFT coefficients into the bitstream. Disclosed decoding techniques include receiving a bitstream of coded mesh data, including coded motion data and decoding the motion data from the bitstream. The decoding of the motion data comprises decoding GFT coefficients representative of the motion data, and then inverse transforming, based on the GFT, the decoded GFT coefficients to obtain decoded geometry data of vertices of meshes in the sequence.
Owner:INTERDIGITAL CE PATENT HOLDINGS SAS

A Big Data-Based Intelligent Early Warning Method and System for Enterprise Operation Risks

This invention relates to the field of economic management technology, and in particular to a method and system for intelligent early warning of enterprise operational risks based on big data. The method includes the following steps: Step S1, extracting a first time series of material flow and a second time series of capital flow from enterprise transaction details, and constructing a two-layer graph structure containing enterprise nodes and edge weights; Step S2, determining the resampling frequency according to a benchmark period, and reconstructing the first and second time series using a time warping algorithm to obtain time-aligned material flow and capital flow sequences; Step S3, calculating the rate of change of the flow phase angle as a graph signal; Step S4, extracting high-frequency components and calculating the high-frequency energy integral of each enterprise node; Step S5, performing a quadratic graph Fourier transform and outputting the results. In this invention, by constructing a two-layer graph structure and using a time warping algorithm to reconstruct and align the material flow and capital flow sequences, the accuracy and timeliness of early identification of individual operational anomalies are improved.
Owner:XINYANG NORMAL UNIVERSITY

Tool wear condition monitoring method based on graph Fourier transform and key order

A tool wear condition monitoring method based on graph Fourier transform and key orders comprises the following steps: step 1: collecting monitoring signals of the tool at different wear stages, and selecting signals of a given length as original signal samples; step 2: converting the one-dimensional signal in the original signal sample into a graph signal according to graph spectrum theory, calculating the graph Laplacian matrix of the graph signal, and extracting eigenvalues and eigenvectors; step 3: converting the original signal sample into a graph Fourier transform (GFT) value related to the order of the eigenvalue; step 4: arranging the GFT values obtained from the monitoring signals at different wear stages in ascending order, taking the eigenvalue orders corresponding to the first k GFT values as key orders, and obtaining h (h≥k) key orders; step 5: obtaining all order combination patterns based on the h key orders and a dimension parameter m, and obtaining a typical pattern library; step 6: matching the order combination pattern of the monitoring signal processed by steps 1 to 5 with the typical pattern library to identify the tool wear condition.
Owner:BEIHANG UNIV

Three-dimensional pre-training model fine tuning method based on spectral domain

The invention provides a three-dimensional pre-training model fine tuning method based on a spectral domain, and the method comprises the steps: calculating a feature vector matrix of the Fourier transform of a graph through point cloud key points by using a point cloud spectral domain adapter for a downstream task, providing additional features through a spectral domain fine tuning strategy, and explicitly introducing the inherent geometric information of point cloud data. According to the scheme provided by the invention, the adjustable parameter quantity, the memory overhead and the storage cost of the fine-tuning pre-training model in the point cloud analysis downstream task can be greatly reduced, and the performance equivalent to or even better than that of all fine-tuning parameters is achieved.
Owner:冯天瑞

A method, device, storage medium and computing device for detecting OTFS modulation signals based on super-threshold method and visual graph

The present invention provides an OTFS modulation signal detection method, device, storage medium and computing equipment based on super-threshold method and visual graph. The method firstly takes the modulus square of the observed signal; then samples the observed signal after modulus square based on the super-threshold method; then calculates the corresponding power spectrum sequence and performs visual graph conversion on the new sequence obtained after sampling, thereby obtaining the corresponding visual graph signal, and performs a graphic Fourier transform on the obtained visual graph signal, and finally uses the smoothness of the graphic signal as the detection statistic, sets the decision threshold, and judges whether the observed signal is an OTFS modulation signal or an ordinary noise signal. The detection performance of the method of the present invention is better than that of the method of directly performing visual graph conversion on the time domain signal.
Owner:NANJING FORESTRY UNIV +1

A complex road surface mechanical response monitoring device and method

The application relates to the technical field of mechanical response monitoring, in particular to a complex road surface mechanical response monitoring device and method, which comprises the following steps: collecting sensor data of each road surface of a highway; constructing a graph structure by using the correlation among mechanics, heat and humidity among sensors; mapping the graph structure into a graph frequency domain feature vector by using graph Fourier transform to form a hypergraph space; constructing a target function based on the spatial proximity and time continuity of the graph frequency domain feature vector in the hypergraph space, so that the graph frequency domain feature vector is globally optimized by using a particle swarm optimization algorithm; inversely transforming the optimized graph frequency domain feature vector to a data space to generate a sensor signal coupling curve, and identifying road surface damage according to the abnormal change of the curve, so that real-time monitoring of the road surface health state is realized. The application aims to improve the accuracy and real-time performance of the road surface health state monitoring.
Owner:HEBEI ZHUANYE CONSTRUCTION ENGINEERING CO LTD

Universal zero sample graph anomaly detection method and system and storage medium

The invention relates to the technical field of graph anomaly detection, in particular to a universal zero sample graph anomaly detection method and system and a storage medium. According to the method, a global information unification module is provided, dimensionalities of node features in different fields are aligned through singular value decomposition, spectral domain normalization is performed by using graph Fourier transform, and the spectral domain is inversely transformed into a node space, so that unified representation of the node features in different fields is obtained. A node neutralization difference scoring module is provided, a generation representation of a graph is obtained through a neutralization generation double-stage auto-encoder, and based on a scoring function of self-neighborhood offset, anomaly scoring is carried out on a similarity difference between a unified representation of node features in different fields and the generation representation of the graph. According to the method, excellent detection performance can be realized without depending on semantic information or a complex optimization module, semantic migration obstacles are effectively overcome, the robustness and generalization ability in different fields are remarkably improved, an anomaly score can be generated without any semantic tag, and effective anomaly detection is realized.
Owner:HARBIN ENG UNIV

Audio signal processing method and device, electronic equipment, chip and medium

The invention provides an audio signal processing method and device, electronic equipment, a chip and a medium, and relates to the technical field of signal processing. The audio signal processing method comprises the following steps: acquiring a stereo signal, wherein the stereo signal is a discrete audio signal; modeling the stereo signal as an audio map signal to determine a left channel map signal and a right channel map signal; respectively carrying out graph Fourier transform on the left channel graph signal and the right channel graph signal to obtain a left channel graph frequency signal and a right channel graph frequency signal of the stereo signal in a graph frequency domain; and determining a direct component, a first environment component and a second environment component of the stereo signal in a graph frequency domain according to the left channel graph frequency signal and the right channel graph frequency signal. Through the technical scheme provided by the invention, the problem that high-quality surround experience is difficult to obtain in related technologies is solved, and the user experience of the audio surround sound effect is improved.
Owner:BEIJING CO WHEELS TECH CO LTD

A weak signal detection method based on graph fourier transform

The application discloses a weak signal detection method based on graph Fourier transform, which comprises the following steps: firstly, calculating the power spectrum of a signal; then, respectively performing two kinds of graph transform on the power spectrum, one is undirected multiple loop graph transform, obtaining the self-loop weight value of each vertex, and defining the self-loop weight value as a graph signal; the other is simple undirected graph transform, obtaining the Laplace matrix of the graph; further, inverting the Laplace matrix and multiplying the inverted Laplace matrix with the graph signal to obtain the graph Fourier transform; finally, calculating the energy of the graph Fourier transform as the basis of signal existence detection, and completing the signal detection by comparing with a specific threshold. The self-loop information of the graph is defined as the graph signal, and the Fourier transform is performed on the graph signal, so that the self-loop information of the graph and the topological information represented by the graph Laplace matrix are effectively utilized, and the detection performance of the traditional graph domain spectrum sensing algorithm under low signal-to-noise ratio and fading channel is improved.
Owner:BEIJING TONGLI AILIN TECHNOLOGY CO LTD

Graph neural network machine learning model anomaly detection method based on graph Fourier transform enhancement

The invention discloses an anomaly detection method for a graph neural network machine learning model based on graph Fourier transform enhancement. The anomaly detection method comprises the following steps: 1) extracting and marking time-space sub-graphs; 2) feature enhancement based on graph Fourier transform; 3) combining structure information aggregation of GNN and GRU; 4) time-dependent modeling based on a gated attention circulation unit; and 5) abnormal edge classification and end-to-end training. According to the method, a dynamic edge anomaly detection task is reconstructed into structure change detection of a target edge space-time subgraph, on the basis, graph Fourier transform is combined to extract frequency domain features to enhance global structure perception, and a novel gating attention circulation unit is adopted to dynamically model a time dependency relationship. Therefore, efficient and robust recognition of structure and time anomalies in the dynamic graph is realized.
Owner:FUJIAN NORMAL UNIV