51 results about "Residual graph" patented technology

The invention discloses a cross-border e-commerce abnormal transaction behavior risk prediction method based on a graph neural network, and the method comprises the steps: synchronously collecting multi-source data of a cross-border e-commerce platform, and writing the multi-source data into a data buffer region according to a unified event timestamp; mapping the multi-source data into a heterogeneous dynamic transaction graph; in the current time window, performing structure comparison on the historical transaction graph and the heterogeneous dynamic transaction graph to generate a structure difference sub-graph, and performing normalization processing to obtain a residual graph; splicing the heterogeneous dynamic transaction graph and the residual graph, inputting the spliced graph into a time sequence graph neural network, obtaining a risk embedding vector of a target transaction sub-graph, and calculating an abnormal score through a mapping function; and comparing the abnormal score with a first threshold value, outputting a risk level of the corresponding transaction and triggering early warning. The cross-border e-commerce abnormal transaction behavior risk can be quickly and accurately identified and predicted, the active identification and real-time early warning of the abnormal transaction risk can be realized, and the real-time performance and accuracy of the risk management of the cross-border e-commerce platform can be greatly improved.

The invention discloses a new energy charging SaaS operation management method based on artificial intelligence. The method comprises the steps of collecting new energy charging facility data and generating a structured data set; constructing a charging state map to form a map structure sequence; inputting an improved space-time residual image convolutional network, and outputting an embedded vector; a price strategy is generated by combining a multi-source information input condition variational auto-encoder; inputting the equipment state embedded vector and the operation sequence into an improved Transform model, identifying the health state of the equipment and generating a maintenance suggestion; and packaging a final result into a service module, deploying the service module to an SaaS platform, and providing multi-end calling. The new energy charging SaaS operation management method is realized by fusing an improved space-time residual image convolutional network and introducing a Transform structure of a gating fusion mechanism.

The invention relates to an autism classification method based on a multi-scale residual image neural network, and aims to cope with the challenge of crowd autism classification in multi-modal medical data. The method comprises the following steps: firstly, providing a new function connection feature construction method, extracting second-order function connection features by using tangent Pearson embedding to capture a high-order interaction relationship between brain intervals, and then adopting a maximum independent domain to adaptively minimize statistical dependence between the features and acquisition sites, and combining F-score to screen the features with the most discriminative ability, so as to obtain the feature with the most discriminative ability. And redundancy is effectively removed. Secondly, a multi-modal edge weight calculation method fusing imaging information and non-imaging information is provided, so that noise interference is effectively suppressed while a key discriminant relation is reserved. And finally, expanding a node receptive field layer by layer by stacking multiple layers of Chebyshev convolutions with residual errors on the subject graph so as to capture multi-level relation characteristics, and performing weighted modeling and adaptive fusion on convolution output of each layer by using a multi-head self-attention mechanism, so that effective integration of multi-scale information is realized, and accurate classification of autism is realized. The method has excellent performance in the aspect of autism classification, and an innovative, feasible and effective solution is provided for solving the autism classification task in the multi-modal medical data.

The invention discloses a geometric topology fusion-based part assembly method, which comprises the following steps of: acquiring boundary representation models of at least two to-be-assembled CAD parts, and converting the boundary representation model of each CAD part into a structured heterogeneous geometric topology graph; performing feature coding on each node in the heterogeneous geometric topological graph to form node feature representation; inputting the node feature representation into a graph attention reasoning module, and obtaining a node embedding representation representing a geometrical relationship and a topological relationship in the CAD part through multi-layer residual graph structure information propagation and feature aggregation; based on the node embedding representations of the different parts, calculating association scores of node pairs among the different parts; and performing supervised training on the feature coding module and the graph attention reasoning module by using the labeled assembly constraint sample data. According to the method, fine geometric features can be extracted from a boundary representation B-Rep model, and a complex topological dependency relationship is inferred, so that robust assembly constraint inference is realized.

The invention discloses a training method and a fault diagnosis method of a solid oxide fuel cell system multi-fault diagnosis model based on a graph neural network, and the method comprises the steps: obtaining multi-source time sequence operation data of an SOFC system in different operation states, and obtaining a data set; the operation state comprises a normal state, a single fault state and a composite fault state, and the data set is marked with an operation state label; preprocessing the sample data in the data set, generating a node feature matrix and a corresponding adjacent matrix according to each piece of sample data, and forming a graph structure data set; constructing a graph neural network model which sequentially comprises a multi-head graph attention layer, a residual graph convolution layer and a global pooling classification layer, then taking graph structure data in the graph structure data set as input, taking a corresponding operation state as output, and training the graph neural network model to obtain a trained multi-fault diagnosis model; the method can be used for multi-fault decoupling diagnosis of the SOFC system, and diagnosis accuracy and practicability are improved.

The application discloses an electroencephalogram epilepsy detection method and system based on a node adaptive graph neural network, relates to a computer system based on a biological model, and is proposed in view of problems such as fixed graph structure in the prior art. The method comprises the following steps: an electroencephalogram signal acquisition and preprocessing step; mixed EEG graph construction; adaptive residual graph optimization; node-specific diffusion convolution; time sequence feature modeling; and classification output. The system comprises a data acquisition module, a mixed graph construction module, an adaptive graph construction module, a node-specific convolution module, a time sequence modeling module, and a classification output module; when the system is running, the method steps are executed, so that the electroencephalogram epilepsy detection function based on the node adaptive graph neural network is realized. Technical advantages include: (1) graph structure self-learning; (2) brain region personalized modeling, and strengthened regional feature expression; and (3) combined space-time dependence modeling, and complete description of the epilepsy dynamic process.

The embodiment of the invention provides an unmanned aerial vehicle group network self-healing method and device based on graph learning, and relates to the field of unmanned system communication and control. The method is used for solving the problems that an existing unmanned aerial vehicle group self-healing method is large in power adjustment limitation, a moving strategy has defects, and a graph learning method faces topological imbalance and is insufficient in response and generalization ability. According to the method, through a hierarchical residual graph topology reconstruction strategy, the problems of message passing unevaluation and excessive node aggregation occurring in the unmanned aerial vehicle relocation process are effectively relieved, it is ensured that the recovered network topology structure is more balanced and stable, and the overall elasticity and survivability of the network are improved; by introducing task semantic embedding, dynamic diffusion intensity and implicit multi-branch adaptive aggregation, the model can dynamically perceive and adaptively respond to damage scenes distributed by different nodes, and the response efficiency and cross-scene applicability of a self-healing algorithm are improved. By designing a differentiable joint loss function, reasonable constraint and stable gradient guidance are provided for optimization of a graph learning algorithm.

The application relates to the technical field of semiconductor manufacturing, and discloses a wafer defect detection method, a wafer defect detection device and a medium. The method comprises the following steps: performing image processing on a target wafer image of a wafer object to be detected to obtain a plurality of local spectrum regions; performing complex filtering processing on the local spectrum regions in a plurality of scales and a plurality of directions to obtain response information of each local spectrum region under different scales and different directions; determining a phase consistency residual graph, a frequency band leakage residual graph and a direction offset residual graph based on phase responses and energy responses in the response information, a pre-constructed spectrum phase reference field, a main frequency band and a main direction; performing multi-scale residual back projection processing on the target wafer image to obtain a multi-scale residual back projection graph; generating a defect saliency graph based on any one or several of the phase consistency residual graph, the frequency band leakage residual graph, the direction offset residual graph and the multi-scale residual back projection graph; and performing defect detection based on the defect saliency graph to obtain a target detection result.

The application discloses an annual cloud coverage prediction method and device based on a graph neural network and monthly features, and belongs to the technical field of remote sensing information intelligent processing, atmospheric science and deep learning. The method comprises the following steps: acquiring historical cloud coverage data of a geographical sampling point, and constructing a 12-dimensional monthly feature vector as an input feature for each sampling point; constructing an adjacency matrix representing the topological relationship of a space according to the geographical coordinates of the sampling point and based on a spherical distance and a K nearest neighbor algorithm; inputting the input feature and the adjacency matrix into a pre-trained spatio-temporal graph neural network model, and outputting a daily cloud coverage grade prediction result of a target year; and the model is composed of a cascaded residual graph convolution network module, a bidirectional long short-term memory network module, a multi-head attention module and a plurality of independent monthly classifiers. The application realizes annual scale cloud coverage prediction only by relying on historical data, and has significant precision advantages and generalization capabilities in complex terrain areas and scenes with insufficient historical data.

The application provides a direction of arrival determination method, device, electronic equipment and storage medium, wherein the direction of arrival determination method comprises: obtaining a vector covariance matrix of a snapshot data of a to-be-tested signal based on the snapshot data; obtaining a noise residual graph based on the vector covariance matrix; obtaining a de-noised covariance feature based on the noise residual graph and the vector covariance matrix; and determining the direction of arrival of the to-be-tested signal based on the de-noised covariance feature. The application can improve the accuracy of direction of arrival estimation based on a noise residual graph.

The invention discloses an image compression method oriented to key point detection optimization, and aims to solve the problem that downstream task performance cannot be ensured at a low code rate in traditional compression. The method comprises the following steps: generating a track main code stream and a skeleton diagram from an original image frame through attitude token extraction and track parameterization; applying a reversible skeleton to render the track main code stream and the skeleton graph so as to synthesize a basic reconstruction image and a skeleton rendering state; calculating a residual image between the original image frame and the basic reconstruction image; coding the residual image based on task hard constraints derived from a skeleton rendering state and a key point similarity threshold, and generating an auxiliary residual code stream and a coding sequence strategy; and outputting the track main code stream and the auxiliary residual code stream. According to the method, the motion structure and the detail residual error are separated, and the task hard constraint is introduced to carry out code rate allocation, so that the performance of the key point detection task can reach the preset threshold value under the condition of low bit rate.

Embodiments of the present application provide an image tampering detection method and device. The method comprises: performing noise extraction processing on a target image to obtain a noise residual graph. The noise residual graph is divided into N reference subgraphs. For any reference subgraph, the tampering score maximum subgraph is determined by a tampering score model; the tampering score maximum subgraph is one of the reference subgraph and the changed subgraph of the reference subgraph; the changed subgraph is a subgraph obtained by expanding or reducing the area of the reference subgraph in the noise residual graph. The tampering region of the target image is determined by the N tampering score maximum subgraphs corresponding to the N reference subgraphs. The area of the changed subgraph is continuously changed adaptively by the score of the tampering score model. If there is a tampering trace in the subgraphs obtained in this way, the relative proportion of small size tampering traces in the subgraph is the highest, which can effectively avoid the omission of small size tampering traces. In this way, the recognition rate of small size tampering traces is improved.

The application discloses a kind of multi-task cell analysis method and system based on residual graph neural network, comprising: normalizing single-cell transcriptome data, selecting the top 2000 genes with the highest transcription level from the normalized data, obtaining new single-cell transcriptome data;According to the new single-cell transcriptome data, construct and train denoising auto-encoder, reduce the dimension of original single-cell transcriptome data, obtain the feature representation after dimension reduction;Using the feature representation after dimension reduction constructs adjacency matrix, constructs residual graph neural network model;Connecting the graph neural network model with the denoising auto-encoder, construct double self-supervised model and train;According to the double self-supervised model, output the clustering result, interpolation result and low-dimensional representation of single-cell transcriptome data.The method and system provided by the application greatly improve the feature discrimination of network extraction, improve the performance of each single-cell analysis task.

The invention relates to the technical field of hybrid electric vehicles, in particular to a hybrid electric vehicle type generator power supply system and method, and the system comprises a multi-source data processing module, a driving condition prediction module, an efficiency solving module and a database. The multi-source data processing module collects working condition, energy state and environment data, and obtains clean data through a variational mode decomposition and wavelet threshold denoising combined algorithm; the driving condition prediction module constructs a dynamic graph based on a residual image neural network, integrates adjacent vehicle motion and gradient influence prediction working conditions, constructs a first optimization function in combination with hybrid electric vehicle performance requirements, and generates target power instructions of an engine and a generator set through an optimization algorithm. And the efficiency solving module is used for solving a second optimization function taking the maximum total efficiency of the system as a target and outputting the optimal torque and the rotating speed set value of the engine in real time. The efficiency and working condition adaptability of the power generation system can be effectively improved, the fuel economy is optimized, and the service life of the battery is prolonged.

The application discloses a kind of based on graph neural network's solid oxide fuel cell system multi-fault diagnosis model training method and fault diagnosis method, obtains the multi-source time series operation data of SOFC system under different operating conditions, obtains data set;Operating condition includes normal state, single fault state and compound fault state, and data set is labeled with operating condition label;Sample data in data set is preprocessed, node feature matrix and corresponding adjacency matrix are generated according to each sample data, and graph structure data set is formed;Graph neural network model including multi-head graph attention layer, residual graph convolution layer and global pooling classification layer in turn is constructed, then the graph structure data in graph structure data set is as input, corresponding operating condition is as output, graph neural network model is trained, i.e. the multi-fault diagnosis model trained is obtained, and can be used for the multi-fault decoupling diagnosis of SOFC system, improves diagnostic accuracy and practicality.

This application relates to a spatial geographic data processing method and apparatus. The method includes: acquiring spatial geographic data corresponding to dynamic geographic entities, the spatial geographic data including location change sequences, attribute label sequences, and external event records; performing cross-domain nested encoding processing on the spatial geographic data to obtain a multi-domain entity relationship encoding set with multi-dimensional unstructured temporal features and heterogeneous semantic relationship dimensions; in the multi-domain entity relationship encoding set, performing residual graph generation processing on the temporal nodes of geographic entities exhibiting label drift and state jumps to obtain a set of residual graphs spanning temporal periods; and performing semantic transition compression processing on unstructured deterministic regions based on the residual graph set to obtain a semantic chain set, which is used to characterize the spatiotemporal semantic reconstruction expression model of dynamic geographic entities in a multi-stage evolution process. This method can compress and model the state mutation patterns of dynamic geographic entities in unstructured deterministic regions.

The invention discloses a disaster recovery redundant link optimization method and device for a smart power grid dispatching communication network, and the method comprises the steps: firstly obtaining the topology and link attributes of the dispatching communication network, mapping the time delay, rental or bandwidth and other indexes into weights, calculating a shortest main path from a source dispatching center to a target substation on an original graph, and carrying out the calculation of the shortest main path from the source dispatching center to the target substation; constructing a split residual image based on the path, and reserving a cost bit of a shared link in the split residual image; a candidate standby path with the shared link number not exceeding a threshold value is obtained through a limited shortest path algorithm, a final redundant path pair is generated in combination with the main path, the total weight of the final redundant path pair is calculated according to a union set, and the weight of the shared link is metered only once; and finally, issuing the obtained main and standby paths to electric power communication network equipment to realize rapid switching when the link fails. According to the method, the construction and operation and maintenance cost is reduced while the reliability is ensured, the resource utilization rate is improved, and good deployment feasibility and near-linear time complexity performance are achieved.

This invention discloses a real-time early warning method and system for road maintenance facilities, belonging to the field of intelligent transportation technology. The method includes: constructing a facility graph model based on dual rules of physical proximity and functional dependency; introducing a dynamic weight correction mechanism to generate a dynamic weighted adjacency matrix; real-time acquisition of multi-source sensor time-series data, which, after preprocessing, is input into a time-series graph neural network; extracting spatiotemporal features of nodes through spatial feature aggregation and multi-scale temporal evolution modeling; mapping the final hidden state to a future anomaly probability sequence; dynamically adjusting the early warning threshold based on the number of historical false alarms, and comparing it with the predicted probability to trigger tiered early warnings. This invention solves the problems of difficult identification of clustered anomalies and delayed early warning response in complex environments by quantifying the implicit functional coupling relationships between facilities and fusing residual graph convolution and multi-scale dilated convolution into a TGNN architecture, achieving high-precision, adaptive proactive early warning of road facility anomalies.

The application discloses a method for evaluating the quality of UHD images and videos based on graph convolution and space-time correlation, which comprises the following steps: dividing an input ultra-high-definition image according to a preset grid, and obtaining node features with uniform dimensions through linear mapping. Based on the spatial distance between the normalized central coordinates, the node features and a normalized adjacency matrix are input into a multi-layer residual graph convolution network for feature propagation to obtain node representation. The node representation is subjected to attention weighted pooling to obtain a standardized quality prediction value. The quality prediction value is subjected to learnable affine calibration, and is subjected to inverse standardization according to statistical parameters of quality scores of a training set to obtain a final image quality score. The method can adapt to the high-dimensional input characteristics of ultra-high-definition images, and can combine the time sequence correlation between video frames to realize a high-precision and high-time-efficiency method for evaluating the quality of ultra-high-definition images and videos without reference.

The invention relates to an infantile autism classification method based on a graph attention mechanism and an adaptive multi-modal graph convolutional network, which utilizes image data and non-image data to explore early diagnosis of infantile autism spectrum disorders, and identifies biomarkers related to infantile autism. The method comprises the following steps: firstly, constructing a self-adaptive crowd graph which takes image features as nodes, dynamically optimizing an edge weight based on the correlation between non-image features by utilizing a pairwise correlation encoder, and then, carrying out sparse processing on a graph structure by adopting an edge random discarding strategy; and then refining the multi-modal features layer by layer by using a residual image convolution network, introducing an image attention mechanism to carry out weighted fusion on the complementary multi-modal features, and finally outputting a final classification result through a multi-layer perceptron. The method has excellent performance in the aspect of autism diagnosis and classification, and meanwhile, biomarkers related to autism classification can be recognized.

The application discloses a single-cell RNA sequence data clustering method based on a robust residual graph convolutional network, and comprises the following steps: determining a data set: selecting several public single-cell RNA sequence data sets; single-cell RNA sequence data preprocessing: performing cell and gene screening operations on the single-cell RNA sequence data set; network construction: constructing a single-cell RNA sequence data clustering network based on a robust residual graph convolutional network; network training: inputting the single-cell RNA sequence data into the constructed network for network training, and evaluating the current clustering performance by four clustering evaluation indexes after the training is completed; and cell clustering: inputting the single-cell RNA sequence data into the trained clustering network to obtain a clustering result. The application can make high-accuracy clustering on single-cell RNA sequence data in view of the noise problem existing in the single-cell RNA sequence data, and can be used for single-cell type annotation and recognition.

The invention relates to the technical field of resource management, and provides a supply chain risk quantification method based on large language model weak supervised learning. The method comprises the following steps: obtaining a multi-time-period operation disclosure text of a target enterprise and cleaning clauses to form a structured statement set; screening a potential risk statement subset based on the supply chain risk keyword library; calling a large language model to carry out weak supervision labeling to generate a risk category label; training a small risk sentiment classification model by using a label, reasoning a full amount of statements, and outputting a probability value of each risk category; aggregating the probability according to a time period to obtain a semantic distribution vector to represent a risk semantic state; and constructing a semantic residual image for the adjacent periodic vector difference values, and calculating a supply chain risk trend index according to the semantic residual image to realize risk evolution trend quantification.

The embodiment of the application discloses a hyperspectral anomaly detection method and system based on global and local feature learning and a medium, the method comprising: extracting latent layer features of hyperspectral data by using two convolution layers; extracting local features by using two residual blocks and a convolution operation after the latent layer features pass through a convolution layer; obtaining down-sampling features by performing global average pooling on the latent layer features, obtaining graph convolution module output features according to global information of the down-sampling features, and obtaining global features based on the graph convolution module output features; obtaining low-rank representation features based on a similarity matrix between the latent layer features and a memory matrix; obtaining fusion features according to the local features, the global features and the low-rank representation features; obtaining reconstructed hyperspectral data by performing a convolution and batch normalization on the fusion features after the fusion features pass through a convolution layer; determining a residual image according to a difference between the hyperspectral data and the reconstructed hyperspectral data, performing anomaly detection on the image to obtain a final two-dimensional detection image, and improving anomaly detection accuracy.

The invention relates to the technical field of graph neural networks, in particular to a graph similarity calculation method based on adaptive information fusion, which effectively enhances multi-layer feature representation and captures key information of a complex graph structure by constructing a residual graph neural network R-GIN; designing a multi-pooling attention network MPA based on multiple pooling and an attention mechanism, accurately capturing node features, and minimizing information loss; then adaptively adjusting the attention weight by constructing a node-graph adaptive information fusion module and a graph-level adaptive information fusion module, and improving the accuracy and robustness of similarity calculation; and finally, splicing the node-graph adaptive fusion score and the graph level adaptive fusion score to obtain a final similarity score. Through verification of a real data set, the method overcomes limitations in the aspects of information extraction, feature fusion and similarity calculation, and has innovation and practical values in graph similarity calculation.

The invention belongs to the technical field of radar radiation source signal intra-pulse analysis, and particularly relates to an intra-pulse modulation recognition method based on a time-frequency fusion graph and a radar graph convolutional network, and the method comprises the steps: constructing a new physically interpretable graph structured representation mode of a radar signal, and comprehensively utilizing the time-frequency information of the signal to form a time-frequency fusion graph; according to the method, the deep learning-based IPMR algorithm is adopted, and a special hierarchical residual image convolution network-radar map convolution network is specifically designed, so that the method can be directly generalized to a complex multi-path time-varying channel scene, and additional retraining is not needed, and therefore, a fundamental limitation of many existing deep learning-based IPMR methods can be broken through; according to the method, training can be carried out only under the additive white Gaussian noise channel, then time-varying multipath channels with different characteristics are directly adapted, priori knowledge of the channel and retraining on target domain data are not needed, and the method has the outstanding generalization advantage.

The invention relates to the technical field of integrated chip manufacturing, in particular to a lithium niobate waveguide defect rapid discrimination method based on depth map reconstruction, and the method specifically comprises the steps: collecting a transverse equivalent response sequence; constructing a transverse coupling residual error and smoothing the transverse coupling residual error to obtain a coupling residual error image; constructing a depth suppression coefficient matrix, and performing coupling suppression on the transverse equivalent response sequence to obtain a final decoupling depth matrix; generating a primary three-dimensional depth map, extracting a depth peak map, and constructing a binary mask to form a defect candidate mask map; extracting a defect coordinate set corresponding to each waveguide, and performing depth reverse checking on each extracted candidate point to obtain a defect depth value; and calculating defect strength to obtain a final defect judgment result set. The problems that in the prior art, lithium niobate waveguides are low in detection sensitivity and high in misjudgment rate, and it is difficult to rapidly and accurately discriminate defects of all independent waveguides in mass production of lithium niobate chips are solved.

The application relates to the technical field of semiconductor manufacturing, and discloses a wafer defect detection method, a wafer defect detection device and a medium. The method comprises the following steps: performing image processing on a target wafer image of a wafer object to be detected to obtain a plurality of local spectrum regions; performing complex filtering processing on the local spectrum regions in a plurality of scales and a plurality of directions to obtain response information of each local spectrum region under different scales and different directions; determining a phase consistency residual graph, a frequency band leakage residual graph and a direction offset residual graph based on phase responses and energy responses in the response information, a pre-constructed spectrum phase reference field, a main frequency band and a main direction; performing multi-scale residual back projection processing on the target wafer image to obtain a multi-scale residual back projection graph; generating a defect saliency graph based on any one or several of the phase consistency residual graph, the frequency band leakage residual graph, the direction offset residual graph and the multi-scale residual back projection graph; and performing defect detection based on the defect saliency graph to obtain a target detection result.

The application discloses an annual cloud coverage prediction method and device based on a graph neural network and monthly features, and belongs to the technical field of remote sensing information intelligent processing, atmospheric science and deep learning. The method comprises the following steps: acquiring historical cloud coverage data of a geographical sampling point, and constructing a 12-dimensional monthly feature vector as an input feature for each sampling point; constructing an adjacency matrix representing the topological relationship of a space according to the geographical coordinates of the sampling point and based on a spherical distance and a K nearest neighbor algorithm; inputting the input feature and the adjacency matrix into a pre-trained spatio-temporal graph neural network model, and outputting a daily cloud coverage grade prediction result of a target year; and the model is composed of a cascaded residual graph convolution network module, a bidirectional long short-term memory network module, a multi-head attention module and a plurality of independent monthly classifiers. The application realizes annual scale cloud coverage prediction only by relying on historical data, and has significant precision advantages and generalization capabilities in complex terrain areas and scenes with insufficient historical data.