19results about How to "Semantic rich" patented technology

The invention discloses an event-driven emergency command and dispatch method, and relates to the field of emergency command and dispatch, and the method comprises the steps: carrying out the feature extraction of historical emergency event data through feature engineering; wherein the feature data set comprises an event type feature vector, an event frequency feature vector, a distribution density feature vector and a text feature vector; training an emergency event classification model based on a CART decision tree by using the feature data set; acquiring real-time emergency event data, and performing feature extraction on the real-time emergency event data to obtain a real-time event feature vector; using the obtained real-time event feature vector as an input, and using an emergency event classification model to identify an emergency event; and according to the identified emergency event, an emergency command and dispatch scheme is retrieved and generated from a preset emergency plan knowledge base by adopting a rule-based case-based reasoning method. In order to solve the problem of low individuation degree of the emergency command and dispatch scheme in the prior art, the individualized emergency command and dispatch scheme is retrieved and generated.

The invention provides a large model-based text generation and information recovery method and system. The method comprises the following steps of: converting complex situation information into simple and accurate text description by utilizing a text generation model; converting the text description of the situation information into a semantic vector by using a text encoder of a pre-trained CLIP model, and capturing semantic information of the text; encoding the noise image into a hidden space representation by using an encoder of a variational auto-encoder, and learning a compressed representation of the image; fusing the semantic vector generated by the pre-trained CLIP model with the image hidden space representation generated by the encoder of the variational auto-encoder, and optimizing the fused hidden space representation by using the trained de-noising Unet network; the optimized hidden space representation is decoded into a high quality situational information image using a decoder of a variational auto-encoder. By applying the technical scheme of the invention, the technical problem of incomplete received information caused by inevitable information loss in the data compression process in the prior art is solved.

The invention discloses a remote sensing image semantic change detection method based on high-resolution feature bidirectional fusion, which belongs to the field of remote sensing image semantic change detection, and comprises the following steps: performing multi-scale feature extraction in a high-resolution network, performing multi-scale feature integration by using a bidirectional deep and shallow feature aggregation module, and performing remote sensing image semantic change detection. The method comprises the following steps: acquiring difference characteristics through a high-resolution difference extraction module, refining by using an efficient channel attention module to obtain two semantic segmentation graphs and a binary change graph, and superposing the two semantic segmentation graphs on the binary change graph to finally generate a semantic change graph. Compared with other existing semantic change detection methods, the method provided by the invention is more excellent, especially has better effects in the aspects of identifying the contour of a change area and detecting a subtle change area of a complex scene, can reduce semantic errors caused by visual similarity, and has wider applicability.

The invention discloses a hierarchical entity neighbor multi-relation fusion knowledge graph completion method, which specifically comprises the following steps of: dividing a knowledge graph, taking a head entity in a knowledge graph triple as a central entity, taking a tail entity as a neighbor entity, and constructing an isomorphic sub-graph; wherein the knowledge graph triple further comprises a relationship; performing aggregation processing on the central entity and the neighbor entities to complete aggregation of the central entity and the neighbor entities; the relation is fused into the center entity, and fusion of the entity adjacency relation is completed; calculating an output matrix of multi-head self-attention through a multi-head self-attention mechanism according to the embedded vector of the fusion entity and the relation information; obtaining an output matrix of the encoder by using a feedforward neural network; a loss function is obtained through negative sampling and optimized, the minimum value of the loss function is calculated, and completion of the knowledge graph is completed.

The present application discloses a text data processing method and device thereof, the method comprising: converting the acquired text data into candidate phrases in a predetermined manner; determining whether a word segment in each candidate phrase belongs to at least one preset dictionary; if it belongs to at least one preset dictionary, determining the corresponding candidate phrase as a visual phrase. The present application makes the semantics expressed by the user more readable and semantically richer.

The invention provides a language model training method and device based on semantic analysis and a medium, and the method comprises the steps: carrying out the semantic analysis of a historical natural statement, so as to obtain a plurality of historical natural segmented words; performing sql language conversion on the historical natural segmented words to obtain historical initial sql language segmented words; if any historical natural statement meets the rewriting condition, determining the historical natural statement as a target historical natural statement; rewriting the historical initial sql language segmented words corresponding to the target historical natural statement to obtain historical rewritten sql language segmented words; and respectively inputting the plurality of historical initial sql language segmented words and the plurality of historical rewritten sql language segmented words corresponding to each target historical natural statement into an initial language model for training to obtain a target language model. By rewriting the plurality of historical initial sql language segmented words corresponding to the target historical natural statements meeting the rewriting conditions, the semantics of the obtained historical rewritten sql language segmented words are richer.

The invention provides a human-computer interaction activity-oriented text description model construction method. The method comprises the following steps of obtaining encrypted network traffic generated when a user interacts with intelligent equipment; acquiring auxiliary data synchronously acquired when the user interacts with the intelligent equipment, and inputting the auxiliary data into a pre-trained multi-modal large language model to obtain a corresponding language behavior text; performing timestamp alignment on the encrypted network traffic and a corresponding language behavior text to form a traffic-description pair data set; performing feature extraction on the encrypted network traffic in the traffic-description pair data set to generate a statistical feature sequence; inputting the statistical feature sequence into an end-to-end model until the model converges; in the trained model, the behavior description can be directly generated only by inputting the network traffic, and any auxiliary data acquisition is not needed.

The invention belongs to the technical field of network security, and discloses a traceability graph level advanced continuous threat sample automatic generation method, which comprises the following steps: generating an abstract attack graph structure by utilizing a flow-based graph generation model; determining a starting process node of each attack stage in the abstract attack graph structure; dividing the abstract attack graph structure into an initial intrusion sub-graph, a permission improvement sub-graph, a sensitive behavior sub-graph and a target achievement sub-graph based on each initial process node; identifying attack technologies in the initial intrusion sub-graph, the permission improvement sub-graph, the sensitive behavior sub-graph and the target achievement sub-graph, and acquiring entity names from the attack instantiation library based on the identified attack technologies; and filling corresponding entity names for the entities only having the entity types to obtain an automatically generated high-level continuous threat sample of the traceability graph level. A large number of available advanced persistent threat samples can be quickly generated.

The application discloses a heterogeneous industrial data to PDV format semantic enhancement conversion method and system; by acquiring a plurality of heterogeneous industrial data source files, converting the heterogeneous industrial data in each heterogeneous industrial data source file into a target data set in a preset format; based on a preset mapping rule set corresponding to the data format of the heterogeneous industrial data source file, converting the target data set into a primary semantic graph, and generating a target data packet associated with the primary semantic graph; through a semantic recognition model, performing feature analysis and semantic recognition on the primary semantic graph to obtain implicit semantic information, performing semantic enhancement processing on the primary semantic graph, generating a semantic enhancement triple, fusing the semantic enhancement triple with the primary semantic graph, and performing consistency checking to generate a target semantic graph; based on a quality evaluation result of the target semantic graph, converting the target semantic graph into a PDV standard semantic file, combining the target data packet, and generating a standard PDV model package.

The application provides a construction worker unsafe behavior intelligent identification method and early warning system, the method comprises the following steps: obtaining a to-be-detected image of a construction site; inputting the to-be-detected image into a construction worker unsafe behavior identification model to obtain an unsafe behavior identification result of a construction worker; wherein the unsafe behavior at least includes an unworn state, a wrong wearing state or a blocked state of personal protective equipment; and the construction worker unsafe behavior identification model is a neural network model for target detection. The detection accuracy is significantly improved, and the fine-grained identification capability is enhanced. Since a learnable position coding component is introduced into the backbone network, the model can explicitly model the spatial position relationship of the feature map, enhance the spatial position perception capability of the personnel (especially small targets and blocked targets) in the construction scene under complex background interference, solve the problem that the traditional convolutional network is not sensitive to absolute position information, and improve the feature extraction accuracy in a complex background.

The present application relates to the field of robot environment perception, and is used for solving the problem that the existing robot environment perception system lacks a unified and efficient hierarchical fusion architecture to systematically process multi-source heterogeneous data, in particular to a robot real environment perception early warning system for multi-sensor data fusion; the present application significantly improves the comprehensiveness and accuracy of environment perception through the cooperative work of laser radar, visual camera, millimeter wave radar and ultrasonic sensor; through the progressive processing flow of space-time registration, cross-checking, feature definition to deep fusion, the multi-source information is sorted by confidence and features are extracted, and the heterogeneous feature fusion module generates an environment dynamic semantic map rich in texture, structure, motion and semantic information by means of convolutional neural network, which provides accurate and comprehensive information basis for early warning decision, and at the same time, the dynamic hierarchical early warning mechanism provides reliable guarantee for robot safety and autonomous operation.

This invention discloses a language model-driven zero-shot object detection method and system. The method includes: training a supervised detection model using visible class data from a dataset; extracting semantic vectors from a large language model based on data class names to generate external knowledge; extracting visual features of visible class images using the supervised detection model, and training a generative adversarial network (GAN) based on pseudo-visual features synthesized from external knowledge; using the GAN to synthesize pseudo-visual features for invisible class data, training an invisible visual feature classifier, and obtaining an updated supervised detection model through parameter fusion, thereby achieving zero-shot object detection of image data. Through the technical solution of this invention, visual features with rich semantics and discriminative power can be generated, improving the understanding and expression of visual content, better aligning the visual space with the semantic space, and solving the problem of semantic confusion in zero-shot object detection.

The invention discloses a skeleton action recognition method and a detection system thereof. The detection system comprises a power supply module, a skeleton signal acquisition module, an image preprocessing module, a receiving thread module, a processor module and an action recognition module. A multi-granularity skeleton feature network is adopted in the skeleton action recognition module, the network is divided into a plurality of stages, and skeleton features of different granularities are processed in each stage. Through multi-scale feature fusion operation on skeleton joint point information, mapping coarse-grained joint point features to a fine-grained space by using feature aggregation, and fusing the coarse-grained joint point features with fine-grained skeleton features; meanwhile, the fine-grained features are compressed to a coarse-grained space by using convolution operation and average operation, and are fused with the coarse-grained features. Through the cross-granularity feature fusion mode, the model can fuse more semantic information while reserving spatial information, so that the precision and speed of skeleton action recognition are improved, and the processing capacity of a system in a complex scene is improved.

The present invention provides a building digital twin framework and modeling method based on semantic Web technology. The building digital twin framework includes physical space entities, virtual space models and associated databases, wherein the physical space entities provide building entity source data for the digital twin framework, the virtual space models are used to virtualize and abstract building resources, and the associated database provides data drive for framework operation based on the semantic function of the model. The associated database includes two relational databases for monitoring time series and structural analysis and three RDF models for monitoring, building information and structural analysis. By jointly querying data of RDF models in the same field and corresponding relational databases and establishing mapping relationships at the ontology level for RDF models in different fields, matching updates and real-time interactions of information flows of the digital twin framework are promoted, and by jointly querying and checking the consistency of measured responses and simulation results, the virtual-to-real mapping function of the model is verified at the data level.

The invention discloses a solid waste monitoring and data management system based on artificial intelligence, and relates to the technical field of water environment monitoring, the system is composed of an image acquisition module, a data preprocessing module, an encoder feature extraction module, an up-sampling fusion module, a detection head module and a data management module, and the system is based on an RW-YOLOv11 architecture. A C3K2Sc feature extraction unit is introduced into an encoder, and floating garbage edge and texture expression is enhanced through space attention and a dynamic channel reconstruction mechanism; a SurfCAU water surface content awareness enhanced up-sampling module is adopted in the neck network, detail compensation of low-resolution features is achieved, and the multi-scale feature fusion quality is improved; a SurfMSDFHead multi-branch structure is adopted in the detection head, multi-scale feature fusion is realized through dynamic weight, and the positioning precision of overlapped garbage is improved by using a Focaler-IOU interval weighting strategy. The system can realize real-time detection, classification and positioning of the floating garbage in the river channel, and has the advantages of high detection precision, good lightweight degree, strong adaptability to complex water surface scenes and the like.

The invention relates to the technical field of power grid fault classification diagnosis, in particular to a power grid fault diagnosis method and system based on text vectorization, and the method comprises the steps: carrying out the word segmentation processing and embedded conversion processing of a power grid alarm information text, obtaining an initial vector, extracting a keyword, and carrying out the vectorization processing of the keyword, and obtaining a label vector; encoding the initial vector for several times to generate a text vector, then splicing the text vector and a tag vector to obtain a spliced vector, carrying out feature extraction for several times to generate a high-dimensional feature vector, and carrying out full-connection forward processing on the high-dimensional feature vector to obtain a power grid fault diagnosis result. Deep semantic and logical relationships in the text can be mined, fault type samples with similar features are distinguished, and the accuracy of power grid fault diagnosis is improved.

The present invention discloses a method and system for rapidly extracting three-branch causal rules of traffic accidents, comprising the following steps: constructing a three-branch concept lattice derived from an object using an incremental construction algorithm; extracting three-branch decision association rules based on the three-branch concept lattice derived from the object; and rapidly extracting three-branch causal rules of traffic accidents. The method of the present invention is applied to mining traffic accident causal rules, thereby obtaining representative three-branch rules of traffic accident causation. After analysis, it is found that the three-branch decision association rules are more semantically richer and more delicately expressed than the two-branch decision association rules, and can express knowledge that the two-branch decision rules cannot express, and can mine richer and more comprehensive traffic accident mechanisms.

The invention provides a text information completion method and system, and the method comprises the steps: converting text information into sub-word vectors through a pre-training language model, reconstructing the sub-word vectors into word vectors, carrying out the semantic tag alignment, fusing the extracted word vector features with the corresponding semantic tags, and carrying out the prediction model training. Therefore, rich semantic and structured knowledge is provided for language expression, more accurate text representation is constructed, the accuracy of text information prediction is improved to a great extent, and the efficiency of data processing is improved.

The present application provides a method for constructing a social media disaster effective information detection model, which is used to extract the topic information of short texts through CRFTM, and then integrate the topic information to train the BERT model for disaster detection effectiveness, which not only helps to reduce the training cost, but also improves the detection accuracy of the model, and the detection performance of the model is significantly improved. The method includes: obtaining user-posted information on the target social media platform; extracting the topic information in the user-posted information through CRFTM. During the extraction process, CRFTM merges the short texts in the user-posted information into a pseudo-document with a longer text length, and then extracts the topic information from the pseudo-document through the combination of word vectors and word-related knowledge; using the topic information as training data, training the BERT model to identify effective disaster information in the input data, and effective disaster information refers to information that is helpful for responding to specific disaster events.