50 results about "Semantic filtering" patented technology

The invention provides a vehicle re-identification data set generation method and device based on a diffusion model, and belongs to the technical field of vehicle re-identification, and the method comprises the steps: building a cue word template based on vehicle attributes, and carrying out the iterative optimization; taking a real vehicle re-identification unmarked image as input, predicting an image-text pair of an unmarked vehicle by utilizing a pre-trained multi-modal image description model, and performing fine adjustment on a diffusion model; and based on the optimized cue word template and the fine-tuned diffusion model, performing optimization in combination with target detection and semantic filtering, constructing a vehicle re-identification data set generation model, and generating a vehicle re-identification data set. According to the method, the effect better than that of an existing virtual data set generation method can be achieved on a downstream vehicle re-identification data set, the number of real data needed by the generation method is only 1% of that needed by an existing method, the real data does not need to be labeled, privacy protection is facilitated, and data labeling expenditure is reduced.

The invention provides a meteorological-based online question and answer service agent construction method and device, and relates to the field of agents, and the method comprises the steps: constructing a multi-dimensional knowledge base based on meteorological and industry knowledge data, and carrying out the real-time retrieval of a retrieval enhancement generation model RAG, and dynamically updating the knowledge data stored in the multi-dimensional knowledge base; constructing a standardized tool library based on weather and industry tools; obtaining user dialogue data, identifying a user demand real intention of a current problem in the user dialogue data, performing semantic screening on tools in the standardized tool library by utilizing an Embedding model according to the user demand real intention, and determining a target tool matched with the user demand real intention in the standardized tool library according to a semantic screening result; and based on the knowledge data, utilizing a TODOLIST planning mechanism to perform dynamic answer path planning for a current question in the user dialogue data through a to-be-called target tool to obtain an answer planning path.

The invention discloses an SLAM dynamic point semantic filtering method based on DDMA-SAM, and belongs to the technical field of synchronous positioning and mapping. A decoupling distillation mechanism is introduced, an image encoder in an original SAM model is subjected to lightweight optimization, a DDMA-SAM semantic network integrating a multi-scale aggregation detection module and an efficient mask decoding module is constructed, and the SLAM dynamic point semantic filtering method based on DDMA-SAM is obtained. And the segmentation performance is improved while the model parameters are greatly compressed. Based on the semantic network, providing a semantic prior and geometric consistency combined-driven double filtering strategy; based on a semantic mask and a confidence threshold, carrying out preliminary dynamic point identification; in combination with the epipolar geometric constraint and the triangulation reprojection error of random sampling consistency estimation, fine elimination of dynamic feature points is realized, and only static points are reserved to participate in camera pose estimation. According to the method, the real-time performance of the system is kept, and meanwhile, the mapping quality and the track stability in a dynamic scene are effectively improved.

The invention belongs to the field of artificial intelligence and knowledge engineering, and discloses a dynamic knowledge distillation method and system based on a knowledge graph, which are used for extracting and reconstructing high-value knowledge from structured or unstructured documents. Aiming at the existing problems of static processing, semantic segmentation, recombination stiffness, cross-document association missing and the like, the method realizes dynamic distillation through the following steps: generating separators in real time based on a domain knowledge graph core entity set, breaking through a fixed partitioning rule, and adapting to multi-type document structures; a concept is utilized to trigger a correlation decision model to filter non-correlation texts, and semantic fragmentization is reduced; loading an adaptive prompt template according to the domain attribute of the knowledge graph, and fitting the characteristics of the fields of medical treatment, education and the like; knowledge is dynamically reconstructed through a logic framework defined by a knowledge graph, and cross-document knowledge point semantic association (such as policy document complementary clause linkage) is achieved. According to the system, through dynamic blocking, semantic filtering, field adaptation and logic recombination, the limitation of a traditional technology is solved, the flexibility and field adaptability of knowledge purification are remarkably improved, and the system is suitable for high-value knowledge mining and structured output of massive documents.

The invention discloses a report generation method and device, equipment, a medium and a product, and relates to the technical field of artificial intelligence, and the method comprises the steps: responding to a received user input instruction, and assisting a user interface module to generate a target keyword and a query short sentence according to the user input instruction through a first agent role; performing document screening and sorting on the basis of the target keyword and the query short sentence to obtain a finely sorted document; performing semantic filtering on the finely arranged document through a semantic filtering module, and calling a second agent role to generate a final document; and calling a third agent role through the semantic processing module to perform semantic processing on the final document, and / or processing the final document through the tool calling module to generate a target report, thereby avoiding the dependence of the report generation process on manpower, realizing automatic report generation, effectively reducing the labor cost, and improving the report generation efficiency. The time for generating the report is saved, so that the efficiency of generating the report is improved, and the accuracy of generating the report can be improved.

The invention discloses an intelligent control system and method for printing equipment. The system comprises an acquisition unit used for acquiring a user voice instruction; the storage unit is used for storing preset corpus data; the recognition unit is used for performing semantic recognition on the user instruction, generating a text instruction and performing semantic filtering to obtain effective instruction text data; and the control unit is used for responding to an instruction of controlling the printing equipment by user voice and generating a control instruction for the printing equipment. According to the method, the user instruction is filtered and then the execution instruction is generated, so that the printing equipment instruction is generated after the validity of the instruction is ensured, unnecessary starting of the printing equipment caused by misrecognition can be prevented, and the accuracy of the voice instruction can be improved to a certain extent.

The invention discloses a task-aware robot availability prediction method based on retrieval enhancement and cross-image alignment. The method comprises the following steps: constructing an accessibility memory bank containing object appearance characterization and interaction prior; performing semantic filtering on the memory bank based on the task tag, and retrieving a reference sample according to appearance similarity; a prediction process is divided into two parts, namely static contact point positioning and dynamic action direction prediction, contact points are correspondingly positioned by querying dense features of an image and a reference sample, and an action direction is regressed by a retrieval-enhanced cross-image alignment model; the model performs action vector modulation on the reference features, and performs gating cross-attention fusion with the query image features, and the fusion weight is jointly determined by the similarity and the learnable weight; and finally, the two-dimensional contact point and the action direction are lifted into a three-dimensional pose and a three-dimensional trajectory, and executable grabbing and post-contact actions are generated. According to the method, cross-category robust generalization under the condition of few samples can be realized, and the operation success rate of the robot is improved.

The present invention discloses a kernel data competition detection method based on static program analysis and fuzzy testing. The method identifies shared resources in the code through static program analysis, obtains access operations for shared resources, records the read and write properties of the access operations, and determines access pairs of potential data competition. By analyzing the semantics of kernel synchronization primitives, shared resource access pairs that are unlikely to generate thread interleaving are filtered to generate a more accurate data competition pair set. Kernel resources and system calls are identified through static analysis, and a system call dependency graph is constructed. The data competition pair set and the system call dependency graph are used to guide the generation and mutation of kernel fuzzy test cases. A thread-aware fuzzy test generator is used to randomize the priority of threads to explore thread interleaving, and runtime feedback information is used to update the edge weights of the dependency graph, improve the system call dependency graph, and guide the mutation of subsequent fuzzy test inputs.

The application relates to the field of outdoor AR positioning and anchoring technology, in particular to a kind of persistence anchoring method and system of outdoor large scene augmented reality content, comprising: collecting image sequence of target area in different time periods and different weather, and carrying out semantic filtering and feature extraction, and constructing multi-dimensional feature index library;When starting AR application, online semantic enhancement hierarchical positioning is carried out, specifically including: LBS coarse screening and context perception, key frame selection and semantic mask, global visual retrieval, local feature matching and geometric verification, gravity constraint PnP solution, and pose fusion and smoothing;AR image is input into lightweight depth estimation network to obtain depth map, and is written into depth buffer to realize the visual effect that virtual object is correctly occluded by real object;Incremental feature fusion and old feature elimination are carried out using image data.The application realizes high-precision, long-time, cross-session stable anchoring and depth virtual-real fusion AR positioning technology.

The application relates to the field of somatic household robot communication and control, in particular to a wireless data processing method and device for a somatic household robot and a robot. In view of the problems that in existing robot communication, full-amount transmission of bus data leads to heavy burden of a wireless link, key data is difficult to be delivered in time, real-time performance and resource utilization are insufficient, a processing scheme is provided, in which intelligent forwarders collect communication total data on the robot side, and switch between full-amount forwarding and semantic filtering according to the state of the collected data; when semantic filtering, preset key data is extracted and sent to a target functional module. The scheme can reduce the occupation of wireless bandwidth by invalid or low-value data, improve the timeliness and reliability of key control and state information transmission, and enhance the communication flexibility and overall control efficiency in a dynamic deployment scene.

One or more embodiments of the invention provide a vulnerability detection method and device. According to the method and device, a to-be-detected target code is subjected to static analysis, and a taint path set in the target code is obtained through recognition; performing priority ranking on each taint path in the taint path set through rules such as DFS, and determining the taint path with higher priority; and selecting taint paths with higher priorities through a preset large language model to perform vulnerability detection. According to the method and the device, semantic filtering is carried out on the taint path set through a preset large language model, and unavailable taint paths in the taint path set are filtered out. According to the method and the device, under the condition that whether the taint path has the vulnerability cannot be judged based on the current information, the context information of the taint path is called based on the MCP, and vulnerability detection is continued based on the context information.

The invention relates to the technical field of medical information retrieval, and discloses a medical information retrieval method based on an intelligent medical science popularization platform. The method comprises the following steps: acquiring a user retrieval request, and dividing the user retrieval request into a medical concept entity and an association description entity; based on the medical concept entities, extracting core concept nodes and adjacent extension nodes in a preset medical knowledge graph, and generating an initial retrieval range; based on the association description entity, performing semantic filtering on the initial retrieval range, screening out semantic matching nodes, and constructing a target retrieval set; analyzing the medical information in the target retrieval set to generate a medical information optimization evaluation value; and performing priority ranking on the target retrieval set according to the evaluation value, and outputting a ranked medical information retrieval result. According to the method, the matching degree of the retrieval result and the user demand is improved, the user can conveniently and quickly obtain high-quality medical information, the retrieval experience is improved, and the method is suitable for an information retrieval scene of an intelligent medical science popularization platform.

The invention relates to the technical field of image generation, is applied to image generation scenes of finance and medical treatment, and discloses a text-to-image generation method and device, computer equipment and a storage medium. The method comprises the following steps: receiving an input text and generating a text semantic vector; initializing the vector to obtain an initial latent variable; inputting the text semantic vector and the initial latent variable into a U-Net network and starting iterative denoising processing; in each iteration denoising processing process, when calculation is carried out to a U-Net network cross attention module, a value domain adjustment mechanism is introduced to modulate an output value domain of the cross attention module so as to improve semantic consistency in a text guide image generation process; and a semantic filtering mechanism is introduced to improve the structural stability in the text-guided image generation process, so that the image generation result of financial and medical scenes is improved.

The invention relates to the technical field of intelligent risk operation and maintenance, in particular to a lightweight station city operation and maintenance intelligent agent system based on bidirectional semantic filtering, and the system comprises the steps: collecting and obtaining multi-source operation and maintenance data when a subway enters an alternate stage of underground operation and overground operation; analyzing the multi-source operation and maintenance data, and generating equipment efficiency reduction risk data of the sensor; identifying and obtaining risk response data based on the equipment efficiency reduction risk data; performing preventive processing on the sensor based on the risk response data, and obtaining vehicle door detection data according to data acquired by the sensor; and performing identification based on the vehicle door detection data, the equipment efficiency reduction risk data and the risk response data to obtain the decision basis reliability of the data. According to the method, time alignment and weighted fusion identification are carried out on the vehicle door detection data and the decision basis reliability, and the equipment maintenance priority index is obtained according to the data change of the sensor in the alternating stage and the decision basis reliability.

This invention discloses a small-sample fine-grained image classification method based on multimodal semantic filtering and saliency fusion, comprising the following steps: Step 1, using a multimodal semantic alignment mechanism to filter the most accurate text description from the original image to be enhanced, providing high-quality generation guidance conditions without semantic drift; Step 2: based on these guidance conditions, generating realistic enhanced images through saliency fusion to expand the training set, solving the problem of scarce small-sample data; Step 3: using a bilateral feature relation network for end-to-end feature learning and classification to achieve small-sample fine-grained image classification on the expanded training set. This invention features accurate semantic alignment, reasonable background logic, and natural foreground fusion.

This application provides a method and apparatus for identifying emotional reason fragments, belonging to the field of natural language processing technology. It employs a technique that combines input text, emotional fragments, and generated common sense knowledge into prompts input into the same large language model. The large language model, based on its deep semantic understanding, judges the relevance of common sense knowledge to the current text context, retaining only the highly relevant common sense knowledge for integration into the model input. This mechanism utilizes the same large language model system to complete the dual tasks of common sense generation and semantic filtering, maintaining consistency in judgment standards and avoiding cross-model semantic understanding bias. By pre-setting a relevance threshold to control the filtering strictness, when the relevance confidence of the large language model output exceeds this threshold, it is determined to be valid common sense knowledge, thereby reducing noise interference and improving the effectiveness of common sense knowledge. Therefore, it achieves the technical effect of improving the accuracy of the model in identifying emotional reasons.

This invention proposes a radio map construction and non-cooperative radiation source localization method based on agent interaction. The method first utilizes a drone swarm to dynamically sample sparse signals, then uses Gaussian process regression to fuse the signal propagation physical priors to generate a spatial attention weight map. These physical priors are injected as biases into a multi-head attention mechanism, enabling multiple agents to form a consistent joint representation of the radiation source target in a unified semantic space. Next, key semantic features are dynamically compressed and selected through task-driven semantic filtering and interaction. Semantic recovery is performed using masked multi-head self-attention and transposed convolutional networks to suppress noise and output preliminary results for both tasks. Finally, an embodied intelligent feedback closed-loop system is constructed. This system can perceive the environment and task status in real time, dynamically adjust the weights and interaction strategies of both tasks, and store the optimal strategy in long-term memory. This invention improves the accuracy of radio map construction, radiation source localization accuracy, and multi-agent collaborative efficiency in sparse sampling, multi-source aliasing, and dynamic complex electromagnetic environments.

The present disclosure provides a method, apparatus, device and medium for recalling structured data, relating to the field of artificial intelligence, and particularly to the field of data processing. By determining a structured data table corresponding to a query text in a structured database, where the structured data table contains multiple pieces of structured data; obtaining multiple types of keywords corresponding to the query text, and performing multi-way recall on the structured data in the structured data table based on a multi-way recall strategy corresponding to each type of keyword to obtain a first structured data set; performing semantic recall on the query text and the structured data in the structured data table, and performing semantic filtering on the structured data obtained by the semantic recall to obtain a second structured data set generated after filtering; obtaining a target structured data set according to the first structured data set and the second structured data set. This application combines keyword recall and semantic recall, which can improve the efficiency and accuracy of structured data retrieval.

The present application discloses a lightweight method for a BIM model with semantic constraints, which relates to the technical field of data processing. The method includes: constructing sub-high-precision models for each component according to the geometric information of each component in the BIM model to be processed; obtaining the semantic attribute information of the component, mapping the semantic attribute information into the sub-high-precision model, and then generating a high-precision model according to the sub-high-precision model; filtering non-main structure components in the high-precision model according to the semantic attribute information to obtain a low-precision model. The present application can perform lightweight processing on the model by means of semantic filtering.

The invention discloses a signaling analysis method, system and device, a medium and a product, and the method comprises the steps: responding to the received complaint information of a target user, and generating a first signaling data set based on network signaling data corresponding to the complaint information; identifying abnormal signaling from the first signaling data set based on a preset abnormal signaling feature library, and extracting signaling data in a preset time range from the first signaling data set based on a timestamp of the abnormal signaling to form a second signaling data set; matching the second signaling data set with a pre-constructed case rule base to determine a target case rule; extracting a signaling subset associated with the target case rule from the second signaling data set through semantic filtering; and performing case analysis on the signaling subset based on a preset analysis normal form to obtain an analysis result. According to the method, a standardized intelligent analysis process is constructed, so that the complaint analysis duration is greatly shortened, and the dependence on artificial experience is reduced.

This invention proposes a method for fusing a large language model and a knowledge graph based on a semantic embedding model. First, the input text is linked to entities. Entities are mapped to an existing knowledge base through batch deduplication, vector library retrieval, and multi-level filtering; unmatched entities are marked as new entities. Then, using linked entities as anchors, a candidate set of one-hop relations is obtained, and semantic filtering is performed sequentially on tail entities, head entities, and triples. Finally, the large language model is used to refine the relation linking results and their confidence scores; potential conflicts are sent for manual review. This invention enables end-to-end knowledge graph fusion.

This invention discloses a method for identifying violations in live-streaming e-commerce advertisements based on two-stage semantic retrieval, comprising the following steps: Step 1, video index construction; Step 2, coarse-grained retrieval of prohibited words; Step 3, fine-grained matching of advertising laws; Step 4, compliance determination by a large-scale model. This invention introduces for the first time a two-stage semantic retrieval mechanism combining video knowledge graph and prohibited word retrieval, and video content and legal vector retrieval. First, coarse-grained screening is performed using a prohibited word database, followed by fine-grained matching and reasoning using legal provisions, achieving hierarchical violation identification and semantic filtering. Finally, the recalled video content, detection requests, and matched legal provisions are input together into a large-scale language model to obtain traceable violation determination results.

The invention discloses a visual SLAM method based on scene dynamic quantitative decision, and relates to the technical field of computer vision and robot navigation, and the method comprises the steps: obtaining an image frame, and carrying out the target detection; quantitatively calculating the area proportion occupied by the high-dynamic object in the current scene to obtain a scene dynamic attitude index; comparing the index with a preset threshold value, and adaptively selecting a processing mode according to a comparison result; in a low dynamic scene, dynamic features are eliminated only by adopting fast semantic filtering so as to ensure real-time performance; and in a high dynamic scene, a refined verification module based on multi-view geometry is automatically activated, and the positioning precision is ensured. Therefore, by adopting the visual SLAM method based on the scene dynamic quantitative decision, the computing resources can be intelligently allocated according to the dynamic change of the environment, the contradiction between the system processing efficiency and the positioning precision is effectively balanced, and the comprehensive performance and robustness of the SLAM system in various real scenes can be remarkably improved.

Embodiments of the present application disclose a relationship-guided full-inductive multi-modal knowledge graph reasoning method and system. The method comprises: generating multi-modal semantic labels for a target entity and extracting its local structure context in a reasoning graph; attention filtering of the semantic labels is guided by a query relationship to construct relationship-guided semantic features; the local structure context is encoded to obtain structure features; semantic and structure features are fused through an adaptive gating mechanism to generate an initial representation of the entity; global message passing is performed on a message passing network to enhance the representation; and finally, the reasoning is completed by calculating the triple score based on the entity representation and the relationship representation. Through relationship-guided semantic filtering and adaptive gating fusion, the present application effectively suppresses multi-modal semantic noise, maintains structure generalization capability, and significantly improves the accuracy and robustness of knowledge graph reasoning under full-inductive settings.

The invention discloses an unmanned aerial vehicle visual radar tight coupling positioning method, device and equipment, a medium and a product, and relates to the technical field of unmanned aerial vehicle autonomous navigation. The method comprises the following steps: processing image data by adopting a semantic segmentation network, screening feature points according to a semantic segmentation result, and carrying out feature matching constraint on static feature points obtained by screening based on a semantic consistency principle; projecting the point cloud data to an image coordinate system, endowing semantic information and performing semantic filtering processing; based on semantic-geometric consistency as a constraint condition, plane feature extraction is performed on the structural semantic point clouds, and a plurality of semantic planes are merged and optimized to obtain a structural semantic map of the environment; and determining an observation model based on the point-surface constraint residual error and the semantic weight coefficient, and carrying out tight coupling filtering and optimization processing by adopting an IESKF to obtain updated state estimation information. According to the invention, the positioning precision and robustness can be improved.

This invention discloses an object detection and coordinate output method based on a visual large language model, relating to the field of image detection. The invention includes inputting a user's instruction to detect a target object, embedding the instruction into the original image, preprocessing the image using bilinear interpolation and centering, segmenting and encoding the processed image based on a ViT visual encoder, introducing self-attention to enhance the detection focus on the target object, and jointly decoding the visual encoding and embedded instruction within the same transformer architecture to output a detection report. This invention avoids the quantization error caused by direct normalization, offers output flexibility (generated through autoregression, with the output format dynamically controlled by the instruction and not limited by predefined templates), and semantic filtering (utilizing the common sense reasoning capability of the VLM for high-level semantic judgment in the hidden state space, eliminating the need for separate classifier training).

The application discloses a kind of end side collaborative real-time video analysis methods based on space-time semantic filtering.The method is based on the intelligent vision device and edge server constitute intelligent vision end side collaborative system, then through intelligent vision device to capture video information, through the tracking detection module and region of interest extraction module of carrying configuration.The tracking detection module adaptively determines unloading decision, with tracking algorithm in local processing each video frame, or unload to edge server deduced by object detection model.Region of interest determines the resolution of each unloading frame and detection model configuration, to ensure that analysis result can be returned in time.Tracking detection module and region of interest cooperate with each other, filter repeated space-time semantic information, to maximize processing rate, while ensuring high video analysis accuracy.

The invention relates to the technical field of big data processing, in particular to an intelligent big data analysis method and system based on AI driving. The method comprises the following steps: acquiring a user behavior activity description text big data set, performing behavior text semantic feature analysis, and performing embedded and filtering feature selection processing at the same time to obtain a user behavior semantic embedded feature subset and a user behavior semantic filtering feature subset; performing repeated representative feature screening processing and behavior activity feature classification according to the user behavior semantic embedded feature subset and the user behavior semantic filtering feature subset to obtain a behavior semantic representative feature subset corresponding to each user behavior activity, and carrying out local interpretability AI intelligent analysis and logic reason visualization display processing on the user behavior activity decision to obtain an interpretability logic reason behind each user behavior activity decision. According to the invention, the efficiency and accuracy of big data analysis can be improved.

The application relates to the technical field of medical information retrieval, and discloses a medical information retrieval method based on an intelligent medical popular science platform. The method collects a user retrieval request, divides the user retrieval request into a medical concept entity and an associated description entity, extracts core concept nodes and adjacent expansion nodes in a preset medical knowledge graph based on the medical concept entity, generates an initial retrieval range, performs semantic filtering on the initial retrieval range based on the associated description entity, screens out semantic matching nodes, constructs a target retrieval set, analyzes medical information in the target retrieval set, generates a medical information optimization evaluation value, prioritizes the target retrieval set according to the evaluation value, and outputs a sorted medical information retrieval result. The method improves the matching degree of the retrieval result and the user demand, facilitates the user to quickly obtain high-quality medical information, improves the retrieval experience, and is suitable for the information retrieval scene of the intelligent medical popular science platform.

The invention discloses a community police emergency message intelligent identification system based on multistage semantic filtering. A data access module pulls a session message; the preprocessing module extracts text information in the session message, preprocesses the text information and inputs the text information to the first-level rule engine module; the first-level rule engine module performs keyword filtering and normalization, screens out high-priority candidate texts through threshold judgment and outputs the high-priority candidate texts to the second-level semantic judgment module; the second-level semantic discrimination module converts the data into a semantic vector, carries out probability distribution, and outputs the data to the emergency scoring module after calibration; the emergency score fusion module fuses multiple factors and performs feature weighting and noise suppression calculation on the multiple factors, and a final emergency score is obtained after normalization; the dynamic threshold and strategy decision module gives an alarm when the final emergency score is greater than or equal to the dynamic threshold; and the alarm distribution and work order module generates an alarm notice and pushes the alarm notice. According to the system, emergency degree identification and alarm distribution are carried out on messages, and efficient early warning and refined distribution of community police duty are realized.