202 results about "Graph algorithms" patented technology

A system and method for identifying a threatening network is provided. The system comprises a network movement before / after algorithm that provides a graphical plot of changes in networks' communications activity from before to after a key event occurs, so that an analyst is able to identify anomalous behavior; a network progression algorithm that provides a graphical plot to analyze behavior in small increments of time without specification or emphasis upon a particular event, so that the analyst is able to see a trend in behavioral changes; a statistical network anomaly ranking algorithm that provides as output a ranked list of the networks; and an anomaly trend graphs algorithm that analyzes and visualizes the networks' anomaly scores over time, so that the analyst is able to see which networks are consistently suspicious, which networks accumulate more suspiciousness in response to an event, and which networks are trending toward more suspiciousness.

The invention provides a natural background video sectional drawing method. The method comprises the steps that 1) the method automatically selects a key frame, specifically the method automatically selects the key frame in a video by calculating inter-frame difference; 2) a key frame Trimap map is generated, specifically a user us allowed to provide useful information through a small amount of manual interaction and a GrabCut segmentation algorithm and a Gaussian blur method are used to semi-automatically generate the Trimap map on the key frame; 3) a middle frame Trimap map is generated, specifically bidirectional optical flow propagation is used to generate the Trimap map of each middle frame, and color, gradient and other information are used to correct errors generated by optical flow propagation; and 4) according to the Trimap maps of all video frames and a Bayesian sectional drawing algorithm with added three-dimensional space-time optimization, the alpha mask map of each frame of the video is calculated, and the alpha mask map and a new background are synthesized to acquire a new synthesized video. According to the invention, the method has the advantages of fast speed and less user interaction, and the finally acquired new synthesized video has the advantages of good time coherence, high reality and the like.

The invention discloses a synchronous positioning and composition algorithm based on point cloud segmentation matching closed-loop correction, and belongs to the technical field of robot autonomous navigation and computer graphics. According to the algorithm, inter-frame matching is carried out on feature points extracted from three-dimensional point cloud to obtain relative pose transformation ofthe robot, meanwhile, the obtained pose is stored at the rear end in a graph form, and then the point cloud is recorded based on the pose to form a map; the point cloud is fragemented and stored by using a point cloud segmentation and description algorithm, and the point cloud fragments are matched by using a random forest algorithm to form a closed-loop constraint condition; and finally, the historical poses and the map are corrected through a graph optimization algorithm to realize synchronous positioning and composition of the robot. According to the method, while the local positioning precision is ensured, the historical poses and maps are stored and corrected, accumulated errors in an outdoor long-distance environment are effectively reduced, and then synchronous positioning and composition of the robot with good global consistency are achieved.

The invention discloses an SLAM-based visual perception mapping algorithm and a mobile robot. The algorithm comprises the steps of obtaining image data of a to-be-mapped area environment and rotatingspeed data of a driving wheel; performing feature extraction on a color image in the image data; after the rotating speed data is calculated and processed according to a kinematics equation, fusing and optimizing the rotating speed data and a depth image in the image data by using a method of minimizing an encoder and a re-projection error to obtain a current frame pose of the robot; screening current frame image data acquired when a robot walks in a to-be-mapped area to obtain a key frame, identifying and classifying objects in an environment by utilizing a color image of the key frame, deleting the classified dynamic objects from the key frame, carrying out sparse processing on static objects, and carrying out local loopback for correction; taking pixel points of the color image and thecorresponding depth image in the key frame as map points, storing the map points through an octree, and carrying out mapping.

A system and method of determining a common neighborhood of users sharing a common activity from a plurality of users is provided. The system and method may be used to predict, for a user in the common neighborhood of users, a potential activity from the activities of at least one other user in the common neighborhood of users.

A method and a system for analyzing dynamic graphs are disclosed. In accordance with such method and system, computations are performed at a plurality of graph vertices every time a change in the graph occurs. In order to minimize the computational load of each computation iteration, previous computation results are reused when the inputs for a computation at a given vertex are unchanged from previous computations. This approach enables real-time data mining from large dynamic graphs, without requiring users to devise their own incremental graph algorithms.

The invention provides a visualized alarm application display method, system and equipment and a storage medium. The method comprises the following steps of: obtaining information of all the applications and respectively storing each application as a node; traversing the information of all the applications and constructing a network data model of the applications, wherein the network data model ofthe applications comprises the nodes and a calling relationship among the nodes; obtaining information of all the alarm applications and adding nodes corresponding to the alarm applications into an alarm network diagram; obtaining a calling relationship among the alarm applications and calculating calling paths between adjacent applications in the alarm network diagram by adoption of a diagram algorithm; connecting the nodes in the alarm network diagram according to the calling paths by adoption of connection lines; and displaying the alarm network diagram. According to the method, data is changed from traditional relationship type storage to network structure storage, the calling relationships among the applications are directly stored, and through the diagram algorithm, the calling paths between adjacent applications can be conveniently found and displayed, so that the repeated query during the relationship search is avoided.

The invention discloses an intelligent suspicious transaction monitoring method based on a semi-supervised graph neural network. The method comprises the steps of collecting and storing original transaction flow; constructing a fund transaction network based on a transaction relationship at the account level; dividing accounts in the fund transaction network into different transaction communities;performing risk assessment and screening on the transaction community to generate a high-risk-density fund transaction network; deriving individual transaction characteristics of the account; and inputting the individual transaction characteristics of the high-risk-density fund transaction network and the account into a semi-supervised graph neural network, outputting the fund transaction risk probability of the account by the semi-supervised graph neural network, and judging the account of which the fund transaction risk probability is higher than a first threshold value as a high-money laundering risk account. The method has the advantages that the abnormal risk of an individual account can be judged, an advanced semi-supervised classification model is constructed through deep data mining and graph algorithm mining, and a traditional risk control means can be remarkably improved.

The invention discloses a WiFi seamless switching method based on a neighbor graph algorithm. The method comprises the steps of building a neighbor graph, selecting a target AP through two-stage decision making, and switching to the target AP, wherein information in the neighbor graph mainly comprises an AP mark, the AP position, a working channel, BSSID, a load, the maximum flow and the Qos level, the first-stage decision making greatly shortens time needed by scanning, and the second-stage decision making is used for selecting the target AP; after the AP is selected, adopting a resource restoring mode for rapid switching. Due to the fact that existing information in the neighbor graph is utilized in the scanning process, the number of scanning channels is decreased, the time for scanning each channel is shortened, and the scanning speed is greatly increased; meanwhile, due to the fact that the number of detection response frames is decreased, the bandwidth needed by scanning is decreased, and the method has a great help to improve the network performance; in addition, the method is tightly combined with an existing 802.11r protocol, and the method is easy to deploy and implement.

The invention provides a method and a system for retrieving a human face on the basis of a Haar classifier and ORB characteristics. In the invention, strong classifiers are obtained by training classifiers, the Haar classifier is obtained after screening and cascading all the strong classifiers, and then existence of the human face in an image to be retrieved is detected by the Haar classifier. Accuracy of the classifier is enhanced by a screening type cascading mode, and speed of an algorithm is accelerated by using an integration graphic algorithm; and then the ORB characteristics are extracted from the detected human face region image and clustered and matched, and characteristic points of the ORB characteristics are detected via direct comparison of gray values of central points and surrounding field pixel points. Existence of the human face in the image to be retrieved can be rapidly detected and accurate matching can be performed so that the method can be widely applied to rapid retrieving of the target human face in a large amount of video or image databases, and has high market application value in practical applications like rapid searching of crime target tracks for a public security monitoring system.

The invention discloses a commodity information pushing method based on link forecasting. The commodity information pushing method includes the steps of determining all associated users of a user to be recommended when it is determined that a selected online social network has the multi-association attribute, building a two-part graph from the user to be recommended and the associated users to commodities, judging whether a mapping node of the user to be recommended exists in the two-part graph or not, if it is determined that the mapping node exists, determining commodity information pushed to the user to be recommended based on a two-part graph algorithm, reconstructing a two-part graph from the user to be recommended and the associated users to the commodities according to selection information input by the user to be recommended, determining commodity information pushed to the user to be recommended based on the updated two-part graph, and then repeating the treating process according to selection information input by the user to be recommended. The invention further discloses a commodity information pushing device based on link forecasting.

The invention discloses a substation anti-misoperation logic intelligent generating method based on topology analysis, and aims at avoiding configuring locking logic for preventing misoperation and self-adapting, reducing building and maintenance workloads, and reducing use complexity, based on ensuring the reliability and the rigor of a Five Prevention system. The substation anti-misoperation logic intelligent generating method based on topology analysis comprises the following steps: analyzing and exporting primary equipment of the substation and connection relation thereof included in an SSD file, mapping the primary equipment of whole substation and connection relations thereof into an undirected graph according to the topological relation, analyzing and summarizing the operation rules of various kinds of primary equipment and abstracting to be a language which can be used by a computer, combining the undirected graph with the language which can be used by the computer to generate equipment locking logic by using nomography from the aspect of mathematics, and thus achieving the purpose of preventing misoperation. According to the substation anti-misoperation logic intelligent generating method based on topology analysis, innovation is provided for the inference engine of the traditional Five Prevention system from the aspect of implementation model, the technological level of intelligent substation for preventing misoperation is improved, and the construction and maintenance efficiencies are greatly improved.

The invention relates to a welded line defect detection method which comprises the following steps of: judging a pixel proportion and judging an image shape, wherein the pixel proportion judgment comprises the following steps: determining the position of a welded line, acquiring an image of the welded line, locking a rectangular frame at the position of the welded line on the image, calculating the gray scale value of each pixel point on the image in the rectangular frame, dividing the gray scale values into low gray scale values from 0 to 128 and high gray scale values from 128 to 255, when the proportion of the low gray scale values of the pixel points of the image is more than 85 percent, determining that the welded line has defectes; and when the proportion is less than 85 percent, judging a shape of the image; and the image shape judgment comprises the following steps: performing binary conversion on the acquired image of the welded line, performing edge extraction on the converted image, extracting the shape of the welded line in the image, judging the shape of the welded line through an eight-communicated graph algorithm, and if the shape of the welded line is irregular, determining that the welded line is defected. By the welded line defect detection method, the detection work finished by a worker can be finished by a computer, so that the labor cost is reduced; and the stability and the high efficiency are realized.

The invention discloses a node abnormity detection method and device based on a graph algorithm and a storage device, and the method comprises the steps: obtaining the attribute features of each nodeof a network cluster in a predetermined time period, building the connection of edges through the similarity of the attribute features, and connecting the nodes to form an undirected graph; calculating the attribute features by using a feature relationship operator to obtain feature vectors of the attribute edges; calculating different metrics of each node to obtain a group of feature vectors of each node; utilizing a predetermined training algorithm to train the feature vectors of the nodes to obtain a group of feature representations of the nodes; and calculating a reconstruction error by using a predetermined self-encoding model to obtain an abnormal offset value of one group of feature vectors of each node, and judging whether each node is abnormal according to the abnormal offset value. Through the mode, the nodes with abnormal behaviors can be rapidly and efficiently detected.

The invention belongs to the field of three-dimensional reconstruction technology, and relates to a multi-view stereo matching three-dimensional reconstruction method of a disordered image. According to the method of the invention, at a view selection algorithm part, 1), for aiming at a global view selection process, firstly a decision condition is introduced; and furthermore the number of views in a candidate view set is fixed adaptively instead of the view set with fixed number of views. No scoring is performed on the views which do not satisfy a condition in an algorithm. Along with continuous expansion of an application range, the number of views in an experiment data set gradually increases, and reduction of view selection time is very important. 2), For a local view algorithm, the method improves a weight influence factor of a polar plane included angle between a neighbor view and a reference view, so that the selected neighbor view is more accurate and more representative. 3), In a view matching area increasing process, the invention provides an adaptive shape space plane sheet algorithm, an adaptive shape can adaptively adapt boundary information, thereby better matching a result.

The invention relates to the data mining, in particular to a method for constructing a data association network. The method for constructing the data association network of the invention adds a graphalgorithm and the text similarity mapping, and constructs the data association relation, and comprises the following steps of collecting and pre-treating data; constructing a data association algorithm; constructing a table-between association data model based on the data flows of the above algorithm. Compared with the prior art, the invention improves the scientific quantitative analysis method,adopts the simple text mapping and the graph algorithm to find part of the correlation relationship, and by introducing the text similarity as a quantitative index, improves the accuracy of the matching judgment greatly.

The APT detection correlation analysis method based on the graph algorithm provided by the invention comprises the following steps: acquiring terminal user behavior data and kernel-level data generated by a detection system to obtain original audit log data; storing the original audit log data into a database with a graph algorithm; performing TTP rule matching on the original audit log data according to an ATT & CK knowledge base model to obtain an alarm event; and evaluating the strength of the dependency relationship between the alarm events by utilizing the path correlation degree, constructing an APT attack scene graph, and promoting the original audit log data to an APT attack step. According to the method, the APT activity is detected with high accuracy and low false alarm rate by analyzing the correlation between suspicious information flows of an attacker at an attack stage, so that the ongoing attack activity can be summarized and backtracked effectively in real time, the real-time network response activity is helped to be carried out, and the visualization of an attack scene is realized.

The invention provides a Chinese news brief generating system. The system comprises a pre-processing module, a marking module and a postprocessing module, which are connected in sequence, wherein the pre-processing module is used for receiving documents to be processed; the marking module is used for acquiring sentences output by the pre-processing module and marking the importance; the postprocessing module is used for determining the news brief according to the marking results. Based on an unsupervised graph algorithm and assisted by simple domain rules, the system can rapidly read the Chinese-oriented news and automatically generate the news brief.

The invention provides a method for analyzing the evolution of associated networks in forums, 1. Reasonable time division and segment extraction; 2. Measurement parameters for measuring community evolution; 3. Longest common subsequence algorithm for solving public nodes in groups; 4. , Discovery of cut points in community network graph based on DFS algorithm. The present invention is applicable to different applications, networks of different scales, and networks of different types. It can be used as long as it can be converted into an undirected graph and follows the same principles. Not only can it be used for the visualization of community network discovery, but also it can analyze the changes of the extracted network over time. Through the analysis of multiple posts, it is found that the evolution process of different posts is also different. For posts with unattractive and boring content, they only have a certain number of replies in the initial stage of posting, and the number of generated communities is relatively small. Important There are also relatively few active nodes. And for those attractive posts, it will last for a long time and the number of communities is large.

The invention discloses an ORB algorithm-based cognitive map construction method. The method comprises the following steps: a robot movement platform detects an external environment, and a visual sensor acquires a visual image and extracts the ORB features of the visual image to construct a local scene cell model; an excitation connection between every local scene cell and a posture cell is established, and a pose cell model controls the stability of an internal pose network through a series of internal activities; path integration of internal mileage information is carried out to estimate thecurrent position of a robot in real time; and a mapping algorithm uses the spatial information characterization of the pose cell model to construct the cognitive map of the environment in real time.The ORB algorithm-based cognitive map construction method can solve a defect that original cognitive map models cannot work indoors, and has high robustness, biological inspiration and practicality.

The invention relates to a three-dimensional laser mapping method and a system, and the method comprises the steps: obtaining data through a sensor placed on a robot, and filtering out unqualified data; calculating the pose of the robot by using a coarse-to-fine matching algorithm; and constructing a new image according to the pose of the robot, and then modifying a map according to a loop closingdetection result. According to the method, the scene map with high consistency can be generated, the algorithm is simple, and the accuracy is high.

The invention provides a semantic map construction method based on a depth camera and a sweeping robot, and is applied to the technical field of robots. The method comprises the following steps: obtaining images of the sweeping robot at multiple positions in the located environment space through the depth camera, wherein the image comprises an RGB image and a depth image; constructing a three-dimensional map of the environment space through a simultaneous localization and mapping (SLAM) algorithm based on each depth map; carrying out semantic recognition on each RGB image through a pre-trainedneural network recognition model; obtaining semantic information of each obstacle in the environment space; fusing the three-dimensional map and the obtained semantic information of each obstacle toobtain a three-dimensional semantic map, that is, the three-dimensional semantic map of the environment space constructed by the application, compared with a two-dimensional map of the environment space constructed according to a laser radar, the richness of information contained in the constructed map of the environment space and the accuracy of the constructed map are improved, and the effectiveworking space range of the sweeping robot is expanded.

The invention discloses a code graph editing and generating method and system. The method specifically comprises the following steps that: obtaining the code graph design requirement information of a user, and designing a code graph design style according to the code graph design requirement; selecting a data source, and reading all code values of the data source; and according to the code graph design style, generating code graph algorithms and configuring and generating the code graph for all code values according to the code graph design style. The source and the amount of the code values are strictly restricted and can not be tampered in a generation process, and the generated code graph is guaranteed to be legal and unique to realize the accuracy of one object with one code; and in addition, according to the design requirement information, the file format and the file name naming rule of a code graph file are configured, and different code graphs can be generated according to different coding positions, different coding types, different code graph formats and different naming rules for a printing press to carry out jet printing.

The invention provides a mobile robot mine scene reconstruction method and system based on SLAM. The method comprises the following steps: acquiring laser point cloud data and visual point cloud data measured by a mobile robot and calibrated synchronously; fusing the obtained laser point cloud data and visual point cloud data; carrying out point cloud motion distortion removal processing and point cloud filtering processing on the fused point cloud data; adding IMU pre-integration data, point cloud key frame data and GNSS data into a constraint sub-graph by combining the processed point cloud data and adopting a multi-constraint factor graph algorithm based on graph optimization, and performing loopback detection to obtain a reconstructed three-dimensional map. According to the invention, three-dimensional reconstruction in the mine scene can be effectively realized, the point cloud map with colors is finally obtained, and the precision of mine scene reconstruction is improved.

The invention provides an interactive matting method based on a recurrent neural network, and belongs to the computer technical field. The invention designs an interactive matting method based on a circulating neural network for solving the problem that the workload of the existing digital matting algorithm is large and the input quality of the user is highly dependent. The designed framework reduces the non-value judgement result by artificially detecting the information area in the cycle process, and can obtain the matting result of the mainstream matting method while reducing the workload.The invention has the following beneficial effects: (1) an interactive model, (2) a long-term and short-term memory network, (3) a reinforcement learning method and (4) a three-dimensional scene rendering.

A system element view-based visual modeling method for constructing system view. After determining basic constituents of the system element view, the present invention constructs system views through a hierarchy view, interface views, algorithm views, process views, and transfer views as step features and thereby provides a specification for visual system modeling in modeling in various fields; such specification has advantages including generality and convenience; system views constructed through the present invention are executable, have clear structures, adjustable hierarchies, and controllable granularities; this modeling method supports both top-down analysis and bottom-up integration for modeling in various systems including software systems and information systems. The modeling method is simple and completely visual, thereby even those not skilled in any modeling language nor computer programming language can easily and independently construct the system view, eliminating the tedious and unnecessary communication with and dependence on professional modelers and application developers, greatly reducing modeling time.

The invention discloses an aluminum profile defect detection method. The method comprises the following steps: detecting the surface of an aluminum profile by adopting an aluminum profile surface defect detection model; if flaws are detected, performing category judgment and regression positioning on the flaws, wherein the aluminum profile surface defect detection model takes Resnet-101 as a main network; the aluminum profile surface flaw detection model adopts a feature pyramid network and a deformation convolution algorithm and is used for performing targeted detection on irregular flaws on the surface of an aluminum profile; according to the aluminum profile surface defect detection model, an ROI Align feature map extraction algorithm and an improved candidate box generation network loss function are adopted to position tiny defects. A feature pyramid network model is improved through a deformation convolution algorithm, so that the detection model has higher matching capability for irregular flaw features; and by improving a candidate box generation network loss function, the regression positioning of the detection model on the tiny flaws is more accurate.

The invention relates to a coded disc and laser radar fused odometer method and a mapping method. The odometer method comprises the following steps: acquiring wheel odometer information through a wheel encoder and a steering wheel angle sensor; obtaining point cloud data through a laser radar, and performing format conversion on the point cloud data to obtain a laser scanning sequence; carrying out feature point extraction on the laser scanning sequence, wherein feature points comprise plane points and edge points; and taking the wheel odometer information as an initial optimization value, realizing pose optimization based on the feature points, obtaining a laser odometer result, taking the laser odometer result as a final odometer result, optimizing the pose based on a factor graph, a fused loop factor and an optional GPS (Global Positioning System) factor, and generating a global map on line through incremental smoothing and mapping algorithms. Compared with the prior art, the method has the advantages of low drift, low delay and the like.

The invention relates to a graph big data streaming division method based on graph algorithm load. The graph big data streaming division method comprises the following steps: a data loading and initializing step: loading graph big data and initializing sub-graphs; a measurement selection and calculation step: calculating an expected measurement value of each sub-graph according to a target of graph big data parallel processing; a heuristic rule selection step: selecting a heuristic rule from a pre-established heuristic rule set; and a sub-graph division step: calculating preset positions of vertexes of the graph big data through a graph big data processing system, dividing the sub-graphs, and calculating the preset positions based on expected measurement values and heuristic rules of the sub-graphs. Compared with the prior art, the method has the advantages that a non-equilibrium division mode is adopted, the division metric measure is selected according to a parallel computer system structure, graph algorithm execution behavior characteristics and the like, the heuristic rule is selected according to different division targets, and diversified graph division requirements and targets in various application fields can be met.