521 results about "Graph computation" patented technology

The invention discloses a graph computation method and engine. The computation method includes the steps of (A) retrieving original relation data of a graph, and obtaining retrieval data corresponding to the vertexes and the edges of the graph, (B) selecting one or more vertexes of the graph to serve as start nodes for breadth-first or depth-first multi-step walking, obtaining walking topological graphs of multiple candidate final nodes, and calculating reach probability of the start nodes to the final nodes on the basis of a breadth-first or depth-first graph walking algorithm and according to the retrieval data corresponding to the vertexes and the edges participating in the walking path, (C) sequencing the calculated reach probability. The graph computation method can obtain results free of popular relations and strong relations. The graph computation engine can retrieve various graph data, supports various graph computation algorithms, supports topological graphs of self-defined graph computation, and supports real-time adding, deleting and modifying of the retrieval data, the topological graphs and shared library data.

The invention relates to the technical field of image processing, in particular to an unmanned aerial vehicle tracking method based on a twin neural network and an attention mechanism, which is applied to continuously tracking a visual single-target unmanned aerial vehicle. According to the method, weight redistribution of channel attention and space attention is realized by using two attention mechanisms, and the representation capability of the model on an unmanned aerial vehicle target appearance model is enhanced by using an attention model for template branches of a twin network; and search images are preprocessed in a multi-scale zooming mode, response graph calculations are separately carried out, inverse transformation of scale changes of the unmanned aerial vehicle in a picture issimulated in the mode, search factors capable of generating larger response values serve as scale inverse transformation of the unmanned aerial vehicle so as to correct the size of a frame used for marking a target, and the transverse-longitudinal proportion of the frame is not changed. According to the method, the tracking precision of 0.513 is obtained through testing (the average coincidence rate is used as a quantization precision standard), and compared with other leading-edge tracking methods, the method has the advantage that the performance is obviously improved.

The invention discloses a distributed graph calculation method based on a disk. The method adopts a distributed calculation model based on the disk, and partitions an original graph into P sub-graphs by a graph partitioning algorithm, one graph algorithm operation is finished through N-frequency iteration, one execution of each sub-graph is one task, and (P*N) pieces of tasks are contained; one task comprises the following steps: (1) loading and constructing the sub-graphs; (2) calculating the sub-graphs; and (3) storing a result, and sending relevant data to other sub-graphs. The method schedules tasks in a running water way, the tasks can be subjected to overlapping execution to hide the time delay of disk read, write and communication in a system execution process, the execution process causes the operation time of the whole system to be almost shortened to calculation time, system performance is greatly improved, and the system can still keep an extremely small system scale by facing to the graphs of different scales so as to greatly save the hardware cost of the system.

A method of building a database of beacon locations is disclosed. Mobile devices submit beacon sighting data to a server; the server using the beacon sighting data to build the database of beacon locations using force directed graph calculations. An iterative process calculates the optimal placement of beacons in a 2D or 3D topology of nodes and edges using force directed graph calculations.

The invention provides a hierarchical network structure search method, a device and a readable storage medium. The method comprises the steps of S1, constructing a super network; S2, acquiring image data and taking the image data as training data of network parameters and structure parameters of the super network; S3, generating a feature map, calculating a cross entropy loss function of the network parameters, and updating the network parameters of the super network; S4, generating a feature map and a KL divergence loss function, calculating a cross entropy loss function of the structure parameters to obtain a semi-implicit variational discard loss function, training and updating the structure parameters of the super network and obtaining a discard probability; s5, updating the basic unitby using the discard probability, and updating the annealing parameter; s3 to S5 are repeated, and the network parameters and the structure parameters are updated; s6, obtaining a final network. Themethod greatly reduces the search time and reduces the calculation complexity while guaranteeing the higher performance, guarantees the search stability and practicality, can be used in the fields ofimage target detection and classification, and improves the image processing speed.

The invention discloses a preprocessing method for deformation data of a panel of a reflector antenna. The method takes deformation data of the reflector antenna as a processing object, and selects different data preprocessing methods according to different data types in engineering of a sector panel sampling point three dimensional coordinate, a triangular unit nodal coordinate and a quadrilateral unit nodal coordinate; an area coordinate and a polar coordinate are introduced to simplify an integral domain of any unit; the efficiency of the numerical integration is improved by using the Gauss integral formula; the number of lobes of a directional diagram are evaluated by referring to the half power lobe width; the number of discrete points of each lobe are picked up to determine the discrete precision of the directional diagram; according to the discrete precision of the directional diagram, the numerical integration calculation is repeated, and a complete far field directional diagram of a deformed reflector antenna is drawn in the end. A simulation result shows that the invention can not only effectively reduce the data processing time, but also improve the calculating precision of the antenna directional diagram; thereby the preprocessing method can be applied to the preprocessing of the deformation data of the reflector antenna under a plurality of engineering situations.

The invention relates to the technical field of Spark big data processing, in particular to a closed-loop target customer identification method based on spark big data, comprising the following steps:S1, data acquisition; S2: data preprocessing; S3: multiple user identity association; S4: constructing a customer portrait model; S5: constructing a customer life cycle value (CLV) model: according to the existing research, the customer life cycle value (CLV) model of the system includes customer current value (CCV) and customer expected value (CFV); S6: constructing a Community Discovery Model:the model adopts a Fast Unfolding algorithm based on Spark GraphX parallel graph computation, and optimizes Modularity value continuously to mine the community where the customer is located; S7: Product recommendations based on random forests. The invention can subdivide the customers, which is beneficial for the enterprise to locate and identify the target customers, and utilizes the limited resources to develop the users and potential users with high value.

The invention discloses data storage and data computation methods and devices. The data storage method comprises the following steps of determining connection subgraphs included in to-be-stored graph data according to the incidence relation between data included in the to-be-stored graph data, wherein graph node data included in the different connection subgraphs cannot be overlapped; and storing the connection subgraphs included in the to-be-stored graph data into a server with the connection subgraphs as storage units. For the to-be-stored graph data, the connection subgraphs can be taken as segmentation units to segment the to-be-stored graph data into several connection subgraphs, and the connection subgraphs are taken as the storage units to store the different connection subgraphs obtained after segmentation into the server, so that when graph computation is carried out, graph nodes included in one connection subgraph only need to access the server stored with the connection subgraph once, and thus the number of server accesses is greatly reduced and the working efficiency of a system is effectively promoted.

The invention discloses a multi-frame image registration and fusion denoising method. The method comprises the steps of (1) selecting a reference frame and a to-be-registered frame; (2) registering the to-be-registered frame to the reference frame by utilizing a camera motion model, thereby obtaining a primary registration graph; (3) estimating a motion of a scene target and obtaining motion vector information by utilizing a characteristic graph, and performing registration on the primary registration graph to obtain a secondary registration graph; (4) calculating a consistency pixel graph forthe secondary registration graph; (5) performing pixel domain fusion and transform domain fusion on the secondary registration graph; and (6) performing adaptive weighted fusion on results after thepixel domain fusion and the transform domain fusion to obtain a final denoising result. A motion model of the scene target is estimated by utilizing the characteristic graph, so that the dense motionvector information can be obtained and the problem that a conventional optical flow algorithm is sensitive to noises is improved; and through transform domain fusion and adaptive fusion methods, imagedetails are kept while image noises are removed.

The invention discloses a botnet domain name family detection method. The method comprises the following steps: acquiring a suspicious domain name; constructing a domain name space-time association graph based on the association of each suspicious domain name among different dimensions, wherein in the domain name space-time association graph, each suspicious domain name is used as a node, an edgeis formed between two domain names with at least one association, and the association between the two domain names is used as an attribute value of the edge; and according to the judgment index of thecompactness of each node in graph calculation, determining to obtain domain names in compact connection in the domain name space-time association graph, and taking a set of the corresponding domain names as a botnet domain name family. According to the method, the relevance between different dimensions of the domain names is uniformly expressed in the form of the association graph, so that the method has higher detection capability. Meanwhile, the botnet domain name family can be detected more quickly, and the method has wider applicability. In addition, the invention further provides a botnet domain name family detection device, botnet domain name family detection equipment and a computer readable storage medium which have the technical advantages.

According to a computer identification and analysis method for a PIEM reinforcement diagram, a computer automation means is adopted, so that the workload of structural design personnel can be effectively reduced, the analysis and check speed of the drawing is increased, and the comprehensiveness and accuracy of standard check are improved. According to the invention, the whole process simulates the reading and analysis of the drawing by structural designers, converts the arrangement relationship of lines and characters in the PIEM reinforcement diagram into the organization logic relationshipbetween structural members and between the contents of each part, establishes a data model convenient for computer query and modification, and automatically analyzes and checks to obtain a result.

The invention discloses a graph data distributed processing system based on a CPU-GPU heterogeneous architecture, and the system comprises the steps: a graph summary generation method which is used for generating a summary graph of large-scale graph data and accelerating the convergence or operation of a graph algorithm; a runtime two-stage load balancing system which is used for balancing loads among the computing nodes and loads on a CPU and a GPU in each heterogeneous computing node; a message processing method improves the communication efficiency by compressing and combining messages; anddividing the large-scale graph data, and processing the graph data in a distributed manner on a plurality of computing nodes by adopting a BSP synchronization mode. The graph computing system based on the CPU-GPU heterogeneous architecture is realized, the efficiency and scale of processing graph data can be improved by the distributed processing system, and the system performance can be furtherimproved by utilizing the powerful computing capability of the GPU.

The invention belongs to the technical field of neural networks, and particularly relates to a crowd counting method based on multi-scale feature fusion. The method mainly comprises the following steps: extracting feature maps of three scales from a backbone network, sending the feature maps into a feature fusion sub-network, and calculating a density map by using the fused feature maps so as to predict the number of crowds in the image, wherein the feature fusion sub-network is designed into three convolution network branches, each branch is identical in structure, adopts an attention fusionnetwork and is divided into two paths, each path is composed of a convolution layer, a normalization layer and an activation function, and the two paths are identical in input and different in outputchannel number and are a single channel and an N channel respectively; a single-channel branch learns the feature weight of a multi-channel output branch, the feature weight is multiplied by the output of a multi-channel output feature map, finally, the feature maps of three large branches are superposed, the feature maps are sent to a decoding module together to output an image density map, and the integral value of the density map is the number of people in the image. According to the invention, the people counting precision is improved.

Techniques are disclosed relating to malware clustering based on function call graph similarity. In some embodiments, a computer system may access information corresponding to a plurality of malware samples and, based on the information, generate a function call graph for each of the malware samples. In some embodiments, generating the function call graph for a given malware sample includes identifying a plurality of function calls included in the information, assigning a label to each of the function calls, identifying relationships between the function calls, and generating the function call graph based on the relationships and the labels. Based on the function call graphs, the computer system may assign each of the plurality of malware samples into one of a plurality of clusters of related malware samples.

The invention provides a graph calculation method based on key value pair storage, which comprises the following steps that: a server loads an original graph data set and stores the original graph data set into a memory according to a key value pair mode; for a graph calculation task, a traversal index is added to key value pair storage. The server receives the graph calculation request sent by the client, analyzes the graph calculation request and sends the graph calculation request to the graph calculation engine for execution. The graph calculation engine accesses graph data through the traversal index, updates key vertexes belonging to keys in the local key value storage, and sends the updated key vertexes to a remote server; and update data sent by other servers is received, and thenthe local data is updated. The above steps are repeated until all calculations are completed, and a calculation result is returned to the client. According to the method, traversal indexes are used, the graph data traversal speed is increased, meanwhile, the distribution characteristics of the key value pairs are fully utilized for data propagation and updating, communication expenditure is reduced, and efficient graph calculation can be conducted in the storage mode of the key value pairs.

The invention provides an actual measurement task point search and task distribution method and a system. The method comprises the following steps: drawing a modular house type graph filled with different entity structures in different colors by utilizing BIM information; calculating the thicknesses of all wall pixels in the house type image, drawing matching templates, and obtaining a task centerpoint of each matching template; combining different entity structures in the house type image, generating a matching source image, matching the matching source image with the matching template, identifying different task points and obtaining task point information; according to the outline of each room, calculating to obtain all measurement stations in the house type; distributing tasks to the measurement robot, collecting three-dimensional point cloud data of all measurement sites, obtaining point cloud data of a complete house type, segmenting the point cloud data of the task points by utilizing ROI information of the task points, referring to design value information of the task points, and calculating to obtain a result of actual measurement. According to the method and the system, automatic identification and positioning of actual measurement task points in the house type are realized, and the measurement robot automatically realizes actual measurement according to the task points.

The invention discloses a node management method and device, equipment and a storage medium. The method comprises the following steps: acquiring node attributes of nodes in a system and a calling relationship among the nodes; based on a graph calculation technology, generating a node relation graph according to the node attributes of the nodes and the calling relation between the nodes; determining a calling sequence according to the calling relation between the nodes, and judging whether the node relation graph is a directed acyclic graph or not according to the calling sequence; if the noderelation graph is a directed acyclic graph, storing the node relation graph to a graph database; when a node call query request is received, obtaining a node identifier in the node call query request;and based on a shortest path algorithm, querying a calling chain of a node corresponding to the node identifier from the node relationship graph according to the node identifier to obtain a query result. According to the method, system nodes are managed by utilizing the knowledge graph, so that the query efficiency of the nodes is improved.

The invention relates to the technical field of distributed graph calculation, and provides a message transmission method and device in a distributed graph calculation process. The method comprises the steps that when a source vertex generates a message M1, the source vertex generates a to-be-sent message group according to a graph relationship where the corresponding source vertex is located; thesource node sends the to-be-sent message to a corresponding destination node according to the destination node identifier in each to-be-sent message group; after receiving the message sent by the source node, each destination node caches the message according to a message format; and each destination node traverses one or more vertexes stored in the destination node, determines the destination vertex belonging to the same graph relationship with the source vertex V1 in the message, and adjusts the corresponding destination vertex according to the content of the corresponding message M1. According to the invention, the number of messages which need to be transmitted across nodes and stored is irrelevant to the number of destination vertexes and is only positively relevant to the number ofdestination nodes, so that the communication cost of message transmission and the storage cost of message caching are greatly reduced.

The invention discloses an unsupervised oil tank target detection method based on a shape-guided significance model. The method comprises the steps of: inputting a remote-sensing image, calculating anedge response graph of the remote-sensing image, and carrying out clustering on all pixels in the remote-sensing image to form superpixels to obtain all the superpixels of the remote-sensing image; obtaining a plurality of clustering regions on the basis of all the superpixels and the edge response graph; utilizing the clustering regions to obtain round probability and a round probability graph;calculating a shape guidance-based significance graph according to all the superpixels and the round probability graph; obtaining a binary result graph through shape-guided significance graph; utilizing a binary result graph to mark oil tank regions in the remote-sensing image to obtain the target regions. Through the target detection method, oil tank targets in low-resolution remote-sensing images under different sizes and illumination conditions can be accurately detected, and the method has better robustness.

The invention relates to a syndrome early warning method and device based on a data model, a medium and equipment, and relates to the technical field of medical big data processing. The method comprises the steps: obtaining original medical data, and calculating the number of target patients with the same syndrome case in the original medical data through a syndrome monitoring model; when it is determined that the number of the target patients is larger than a first preset threshold value, extracting attribute information of the target patients with the same syndrome case, and generating a group set according to the attribute information of the target patients; according to the attribute information and a multi-dimensional space-time relation calculation model, calculating a space-time relation graph of the target patients included in the group set; and finally, calculating the number of the to-be-predicted objects and the infection probability of the to-be-predicted objects included in the time-space relation graph, and performing early warning on the syndrome according to the number of the target patients, the number of the to-be-predicted objects and the infection probability of the to-be-predicted objects. The method and device can increase the early warning efficiency.

The invention discloses a deep learning model training method for fine portrait picture segmentation. The method comprises: collecting and obtaining portrait pictures, and performing binaryzation to form a training data set; detecting a boundary by using a Canny edge algorithm; expanding the boundary by adopting an adaptive expansion operator to form an edge region, and obtaining a portrait edge/background region annotation graph; and inputting the original image into a deep learning model for training processing, calculating image gradient loss and segmentation cross entropy loss according tothe portrait edge/background region annotation graph in the training processing, and jointly performing training optimization on the deep learning model. According to the method, a rough manual annotation result is used, a segmentation result which is more accurate than manual annotation is trained in a self-supervision mode, the manual annotation cost is greatly saved, and the method can be applied to various scenes such as personnel monitoring, portrait analysis and portrait editing.