58 results about "Spatial graph" patented technology

The invention relates to a three-dimensional image matching method based on a convolutional neural network, which comprises the following steps: acquiring a plurality of three-dimensional matching image pairs and corresponding real disparity, and taking the three-dimensional matching image pairs and the corresponding real disparity as a data set; Constructing a convolutional neural network, and selecting a linear correction unit (RELU) function for activation; Training the convolutional neural network by adopting a back propagation algorithm, and determining a network error function and a learning rate; Through the calculation of the convolutional neural network, the network outputs a matching cost space diagram of the left and right image blocks; And performing matching cost aggregation,disparity selection and disparity refinement on the cost space graph, and selecting a pixel point with the minimum cost as a matching point to obtain a final disparity map. The image can be directly used as network input, the overall disparity map obtained through the convolutional neural network matching algorithm is relatively smooth, a relatively good matching effect can be achieved in a non-texture region and a depth value abrupt change region, and relatively good robustness is still achieved for an image pair with illumination change and incomplete correction.

The invention discloses a graph learning-based graph convolutional neural network traffic prediction method and system, and the method comprises the steps: obtaining a more accurate new adjacent matrix through the learning of a graph learning module, capturing a plurality of spatial relationships through the graph learning module, and improving the traffic prediction precision; meanwhile, the space-time convolution block comprises two space-time convolution layers and a space graph convolution layer, and the space-time convolution layers are obtained by combining expansion convolution and a gating mechanism, so that long-time dependence can be effectively captured, and training time and space resources are saved.

The invention discloses an area-based remote sensing image registration method, and mainly solves the problems that registration of multispectral images with high image gray scale difference cannot be performed by existing algorithms. The technical scheme is that 1. a Gaussian scale space is established for a reference image and an image under registration respectively; 2. a sketch graph of the reference image and the image under registration is generated respectively; 3. the Gaussian scale space graphs and the corresponding sketch graphs of the reference image and the image under registration generate respective structural areas and non-structural areas; 4. the features points of the structural areas and the non-structural areas are detected; 5. the features points of the two structural areas of the reference image and the image under registration are added; 6. descriptors of the added feature points are obtained by using an SIFT algorithm, and registration parameters are obtained by utilizing the feature points descriptors; and 7. affine transformation is performed on the image under registration by utilizing the registration parameters so that a registration result is obtained. Registration of multi-source and multispectral remote sensing images with high gray scale difference can be realized.

The invention discloses a method for quickly constructing a spatial topological relation of a vector earth surface coverage pattern spot and a storage medium. The method comprises the following steps:constructing an edge topological relation network of a task graph, extracting the construction and edge relation of the topological relation in a task area, and carrying out edge connection and storage on the topological relation between the task areas. The data structure based on a hash table structure is used for constructing the topological relation between the chart and the task area, and thekey code value is mapped to one position in the chart to access the record, so that the searching speed is increased. Thousands of sub-tasks are constructed according to an index structure, namely atask partition graph, of the spatial graphic data, parallel driving can be conducted through multiple machines and multiple processors, accelerated processing is achieved, and task time consumption isreduced.

The present invention overcomes the limitations of the prior art by exploiting properties of the projection of a three-dimensional building structure (such as a roof) onto the ground. This projection is a two-dimensional spatial graph, which can be constructed for example by a user or by an image recognition algorithm. The spatial graph is processed to recreate a three-dimensional model of the building structure.

The invention discloses an operation assisting system and method based on medical image three-dimensional reconstruction. The system comprises a host, a display unit and a man-machine interaction unit. The host comprises an image data processing module, an image data setting module and an analog image module. The image data processing module performs three-dimensional model reconstruction on imported DICOM data to generate a first image; the image data setting module is used for differentially setting focuses and tissues in the first image to obtain a target image; the simulation image moduleloads the target image so as to carry out simulation space graphic display; the man-machine interaction unit performs interactive operation on the simulation space graph to realize a simulation operation; and the display unit displays the simulation operation process in a naked eye 3D form. The method has the following beneficial effects: the reconstructed focus and each tissue are differently arranged, so that the corresponding part is highlighted, the sense of space is enhanced, and meanwhile, the defect of dizziness of AR equipment is also avoided in a naked eye 3D form.

The invention provides an intelligent molecular design method based on an auto-encoder and third-order graph convolution. The method comprises the following steps: 1, enabling an encoder to express amedicine molecule in a molecular graph form, and enabling each molecular graph to be decomposed into a corresponding third-order substructure; 2, performing graph convolution on the third-order substructure by the encoder, and outputting an implicit vector of a molecule to acquire an implicit vector space; 3, training a decoder; and 4, enabling the decoder to find an implicit vector similar to themedicine molecule in the implicit vector space, and decoding the implicit vector to obtain a new medicine molecule. The method disclosed by the invention has the advantages that the three-order substructure of a molecule is subjected to spatial graph convolution by combining an auto-encoder and a graph convolution network method to find out an atom corresponding substructure, namely a functionalgroup, with high contribution degree to the molecule.

The invention discloses a controller local area network protocol verification method based on state space search. The method comprises the steps of system modeling, model pre-processing and model verification. In the modeling process, an abstract time state automat model is constructed, a state transition diagram prevents, through the utilization of the idea of dynamic planning, a state spatial graph from being generated in a redundant state. In the verification of a controller load area network software model, the state search algorithm based on decomposition is used for further optimizing the state search space of the model, the verification process of the whole system is made to be further optimized in time and space complexity, and data processed through the algorithm are output and converted into a language form which can be understood by inspection personnel easily. The controller local area network protocol verification method can be used for effectively verifying whether relevant network protocols meet requirements, and relieves the problem of state space explosion in the verification process. The space-time complexity of the system in the verification process is low.

The invention discloses a deep learning network based on a group convolution feature topological space. The deep learning network comprises a convolution feature extraction layer, a group convolutiontopological layer and a deep feature recognition layer. The convolution feature extraction layer is used for extracting multi-channel CNN convolution features of the sample data and taking an extraction result as input of the group convolution topology layer; the group convolution topology layer is used for combining the extracted multi-channel CNN convolution features, forming group convolution according to group classification by using channel indexes, constructing a graph topological space, regarding each group convolution feature as a graph topological space node, automatically / manually constructing a graph topological space node connection rule, generating a Laplace matrix L, and taking the Laplace matrix L as input of a depth feature recognition layer; and the depth feature recognition layer is used for outputting group convolution feature topological space diagram features corresponding to the sample data according to the input Laplace matrix L. According to the method, graph topology space rules of CNN features under different channels can be given, so that the traditional CNN training and convergence speed is increased.

The invention provides a pedestrian track prediction method and device, equipment and a storage medium. A space-time diagram is constructed, the space-time diagram comprises space diagrams corresponding to a current video frame at multiple moments, and the space diagrams comprise target nodes and connection edges between the target nodes. According to node attributes of the target nodes and connection edge attributes of the connection edges, final space attribute values of target objects are determined, the node attributes comprise scene features and track features of the target objects, and the connection edge attributes comprise interaction strength between the two target objects. The time dependency relationship of the target objects is determined according to the final spatial attribute values at the multiple moments. The track of each target object in the video frame with the preset time length is predicted according to the time dependency relationship. Scene features and track features are used as node attributes, interaction strength optimization is carried out according to the node attributes and the connection edge attributes, a final space attribute value is obtained, and the accuracy of a pedestrian track prediction result in a complex scene is improved.

The invention discloses a drawing system and an application thereof. The drawing system comprises positioning modules of which each is provided with a wireless transmitting module, a wireless receiving module, a ranging module, a GPS module, a radar module, state indicator lamps, a communication module, a coding module, and a height gauge. According to the drawing system, a lot of positioning modules distributed in positions to be measured are interconnected, the positions are connected to form multiple areas, the multiple areas overlap one another, and the general spatial graph of the positions to be measured can be obtained. The advantage is that no matter whether the drawing system is used for drawing the space inside a building or is used for drawing an unknown area, the only thing needed is to scatter a lot of positioning modules in an environment needing drawing, and the operation is very convenient.

The invention discloses a method for measuring volume. The method comprises the following steps: S1, acquiring an image of a to-be-measured object through an augmented reality module; S2, matching theimage of the to-be-measured object with an image recognition library to obtain coordinates of fixed points, the fixed points being capable of forming a space graph; and S3, calculating the volume ofthe to-be-measured object according to the coordinates of the fixed points. According to the method and a device for measuring volume in the invention, the image of the to-be-measured object is acquired through the augmented reality module, the vertex coordinates are acquired through image recognition, so that the volume of the to-be-measured object is calculated, only the to-be-measured object needs to be photographed in the whole process, manual measurement is not needed, the operation of volume measurement is effectively simplified, and the volume of the object can be rapidly measured.

The invention discloses a behavior recognition method and device, and relates to the technical field of computers. One specific embodiment of the behavior recognition method comprises the steps of dividing each frame of skeleton sequence data of a to-be-recognized object in a preset time range into at least one hop of sub-graph data; determining behavior characteristics of the object according tothe divided multi-hop sub-graph data; determining a time feature corresponding to the behavior feature according to each frame of skeleton sequence data of the to-be-recognized object in a preset timerange; According to the behavior characteristics of the object and the time characteristics corresponding to the behavior characteristics, recognizing to obtain a behavior classification result of the object. According to the embodiment, human body part information of different scales can be extracted, the diversity of spatial domain information can be improved, and the receptive field of the spatial graph convolutional network can be expanded.

The invention discloses a multi-information integrated power grid space-time multi-dimensional layer model construction method. The method comprises the following steps: analyzing power grid data; dividing the power grid data into three categories, namely a space layer, a physical layer and a data layer, and setting the three categories as power grid layers; creating a node and a relationship in each graph layer; creating a relationship among different graph layer nodes; storing a graph database and a relational database according to the data characteristics; updating the layer data along withthe time axis; carrying out layer calling and layered display in the power grid space-time multi-dimensional layer model according to requirements. According to the method, a power grid space-time multi-dimensional layer model construction process is explained, layers and internal relations are clearly divided, changes of the layers along with time are proposed, a layer database is introduced, and a system solution is provided for fusing data, geographic information and user information of all professional ports and achieving one graph of a power grid.

The invention relates to the technical field of video information, in particular to a video highlight detection method and device based on a graph neural network. In order to solve the problem of lowdetection precision of video highlights in the prior art, the invention provides the method, which comprises the steps of obtaining image feature information of each frame of image in a to-be-detectedvideo through a preset image feature extraction model based on the to-be-detected video obtained in advance; constructing a space diagram corresponding to each frame of image based on the image feature information of each frame of image; according to the space diagram corresponding to each frame of image, obtaining semantic features of the object in each frame of image through a preset semantic feature extraction model, and constructing a time sequence diagram corresponding to each frame of image according to the semantic features of the object in each frame of image; and according to the time sequence diagram corresponding to each frame of image, obtaining a user interest score of each frame of image in the to-be-detected video through a preset video clip detection model. The detection accuracy of the video highlight is improved.

The embodiment of the invention provides an image target identification method and device. The method comprises the following steps that: obtaining an image to be identified; inputting the image to beidentified into a preset three-dimensional classification model; extracting the three-dimensional image characteristics of the image to be identified by the three-dimensional classification model, according to the three-dimensional image characteristics, generating a three-dimensional probability spatial graph of the image to be identified, identifying a target object in the image to be identified according to a pre-stored relationship between each target object and the three-dimensional probability spatial graph, and outputting the target object as an identification result; and obtaining theidentification result output by the three-dimensional classification model. By use of the embodiment of the invention, image target identification accuracy can be improved.

The invention provides a multi-dimensional enhanced epileptic seizure prediction system based on a graph convolutional network, the scheme takes a multi-channel spatial relationship as a breakthrough opening, the contribution degree of the multi-channel spatial relationship to epileptic seizure prediction is explored from three dimensions of frequency + space + time. The model comprises three parts, respectively an information reconstruction space, a graph encoder and a space-time predictor. The information reconstruction space and the graph encoder mentioned by the model allow feature enhancement and feature extraction of richer epilepsy electroencephalogram signals, and especially explores the correlation among electroencephalogram channels, so that feature representation is enhanced, and the epilepsy electroencephalogram seizure prediction accuracy is improved; meanwhile, the core structure of the space-time predictor in the scheme adopts a gating circulation unit which is used for exploring the law of epilepsy electroencephalogram signals on the time level, and the purpose of improving the model operation efficiency is achieved by reducing the network parameter scale.

The invention discloses a semantic analysis method for graph migration, which comprises the following steps of: extracting a scene graph, and generating a scene graph model by utilizing single-frame picture text description; extracting a space diagram, and generating a space diagram model by utilizing the single-frame picture; model training prediction: training a neural network through an input graph and an output text, and predicting a test set; wherein the model training prediction is trained through a graph convolution network and a long-term and short-term memory network. The method can adapt to changes of objects in the video in space and behaviors, and correct description can be made.

The invention discloses a controller local area network protocol verification method based on state space search. The method comprises the steps of system modeling, model pre-processing and model verification. In the modeling process, an abstract time state automat model is constructed, a state transition diagram prevents, through the utilization of the idea of dynamic planning, a state spatial graph from being generated in a redundant state. In the verification of a controller load area network software model, the state search algorithm based on decomposition is used for further optimizing the state search space of the model, the verification process of the whole system is made to be further optimized in time and space complexity, and data processed through the algorithm are output and converted into a language form which can be understood by inspection personnel easily. The controller local area network protocol verification method can be used for effectively verifying whether relevant network protocols meet requirements, and relieves the problem of state space explosion in the verification process. The space-time complexity of the system in the verification process is low.

The invention relates to a behavior recognition method and system based on a human skeleton. The method comprises the following steps: acquiring an image sequence which is continuous in time; using an attitude estimation method to obtain a skeleton sequence; the skeleton sequence is sent to a first model and a second model, the first model is a time convolutional network and is used for extracting time features of the skeleton sequence, and the second model comprises a view adaptive network and a space graph convolutional network and is used for extracting space features of the skeleton sequence; and fusing the time features and the space features, and outputting human behavior actions. Compared with the prior art, the method has the advantages that the image information is converted into the human skeleton sequence information, the spatial features and the time features are extracted respectively, the visual angle self-adaptive adjustment is performed in the spatial feature extraction, and the connection relationship between the joint points is added, so that the recognition accuracy and robustness are better.

The invention discloses a spatio-temporal diagram neural network gas concentration prediction method based on spatio-temporal data, and the method specifically comprises the following steps: 1, collecting gas concentration data in a mine through gas monitoring points arranged in the mine, and taking the gas concentration data as a gas data concentration data set; 2, generating a space diagram structure of the mine monitoring points; 3, processing the gas concentration data collected by the mine monitoring points as a time sequence to obtain a gas concentration time sequence, and establishing a training sample set of a space-time diagram neural network according to the obtained gas concentration time sequence; 4, using the sample training sample set of the space-time diagram neural network to construct a space-time diagram neural network gas concentration prediction model; and 5, outputting a gas concentration prediction result. According to the method, the problem of low prediction precision caused by the fact that a traditional neural network gas concentration prediction model does not consider the space-time characteristics of actually measured gas data is solved.

The invention discloses a visualization method and device based on a space graph convolution energy and material supply chain, and the method comprises the steps: carrying out the mining of energy and material supply chain data, and constructing demand and productivity data display data; determining associated data of the demand and productivity data of the energy and material supply chain; generating a voice interaction mode and a data transmission path based on the interaction data of the energy and material supply chain; and according to the productivity data display data, the associated data, the voice interaction mode and the data transmission path, generating visual information so as to display the data of the energy and material supply chain based on the visual information. According to the invention, the visual display of the data is realized, the efficiency and accuracy of data acquisition are improved, and the operation efficiency of the material supply chain is ensured.

The invention provides an image weak supervision target detection method based on class-agnostic foreground mining. The method comprises the steps of generating a foreground attention map through a CNN based on an image to be subjected to target detection; calculating the foreground relative confidence FRC of each candidate frame based on the foreground attention map, and screening out foreground candidate frames according to the FRC of each candidate frame; constructing an instance space graph based on the foreground candidate frames, constructing a tag semantic graph based on tags of the data set, performing graph matching on the instance space graph and the tag semantic graph, and classifying each foreground candidate frame according to a graph matching result; and generating a pseudo supervision frame according to the classification result of each foreground candidate frame, merging the pseudo supervision frame and a spatial neighbor frame of the pseudo supervision frame to obtain a pseudo instance label, and taking the pseudo instance label as a target detection result of the image to be subjected to target detection. According to the method, positioning and classification tasks are separated, so that bidirectional improvement of positioning and classification performance is realized, and weak supervision target detection performance of the image is effectively improved.

The invention relates to the technical field of computer vision, in particular to a weak supervision time sequence action detection method based on space-time correlation learning, which comprises the following steps: S1, extracting features from video frames through an I3D network; s2, constructing a dynamic space graph network structure for the video to obtain video space features; s3, constructing a one-dimensional time sequence convolutional network to obtain video time sequence features; s4, fusing the time sequence features and the spatial features; s5, using action-background attention mechanisms, namely action attention and background attention, which are respectively used for pooling original video features; s6, predicting a class activation sequence of space-time correlation of actions and backgrounds in the video, predicting an action activation sequence or a background activation sequence in the video, and respectively obtaining three classification losses; s7, calculating a total loss function; and S8, using the trained model for action detection. According to the method, the problem that the action example is incomplete and inaccurate in the existing weak supervision time sequence action detection method is solved.

The invention relates to the field of the geographic information system, and specifically relates to an optimization method and system of spatial graphic aggregation. The method comprises the following steps: acquiring a grouping condition of users, and establishing an input cache, receiving input spatial graphs, outward caching each input graph with a certain distance (according to an input condition) to form a new graph and placing in the input cache, checking the graphs in the input cache, taking out the intersected graphs to combine, and inwardly caching the combined graph with a certain distance (according to the input condition) and then outputting. The system comprises an input preprocessing module for establishing the input cache and pre-inputting graphic data into the input cache,and a spatial graphic intersection and joint analysis module connected with the input preprocessing module and used for performing outward caching operation on the graphic data and performing inwardcaching operation on the mutually overlapped graphic data after the outward caching operation. Through the technical scheme provided by the invention, the defects of being low in efficiency and high in cost in the prior art can be effectively overcome.

A system and method for completing trajectory prediction from agent-augmented environments that include receiving image data associated with surrounding environment of an ego agent and processing an agent-augmented static representation of the surrounding environment of the ego agent based on the image data. The system and method also include processing a set of spatial graphs that correspond to an observation time horizon based on the agent-augmented static representation. The system and method further include predicting future trajectories of agents that are located within the surrounding environment of the ego agent based on the spatial graphs.

The invention discloses a multilayer network clustering method based on semi-supervision, relates to the technical field of artificial intelligence and complex networks, and not only takes obtained consensus prior information as a preprocessing means to enable a low-dimensional representation matrix H (i) of each layer obtained through symmetric non-negative matrix factorization to be more excellent. Moreover, the obtained consensus prior information is coded into a consensus subspace graph regularization item, and the consensus low-dimensional subspace H is optimized during the overall non-negative matrix factorization, so the method can make full use of the complementary topological structure information of each network layer, and can also make full use of the obtained consensus prior information; and the method is especially suitable for a multi-layer network with a large amount of noise and a sparse structure. The method is applied to social networks, protein networks and other multi-layer networks, cluster structures of various types of multi-layer networks are successfully recognized, and the method has great significance in understanding some social interaction behaviors among people, recognizing crowds with specific social attributes and improving social cooperation efficiency.