144results about How to "Improve recall" patented technology

The invention relates to an electric power marketing business abnormity real-time detection method and system based on stream-oriented computation, and belongs to the technical field of electric power system optimizing.The method comprises the steps that data snapshots are extracted from an electric power marketing business system, difference comparison is conducted on the data snapshots and historical snapshots of an intermediate library, and standardized increment events are generated and stored; capturing an incremental event in real time through a data change capturing tool and pushing the incremental event to a message queue; a streaming computation engine consumes the event stream, sequentially performs data cleaning, association with a static dimension table and sliding window statistical feature calculation, and constructs a feature vector; and performing parallel analysis and weighted fusion on the feature vectors based on a business rule base and an online machine learning model to generate a comprehensive risk score, and outputting an abnormal event when the score exceeds a threshold value. According to the method, the problems of exception identification lagging and complex work order process in a traditional batch processing mode are solved, the crossing of the business risk from hour-level detection to minute-level real-time perception is realized, and the timeliness and accuracy of power marketing risk management and control are improved.

The invention discloses a low-overlapping-rate point cloud registration method based on topology-metric decoupling, and belongs to the technical field of computer vision and three-dimensional reconstruction, and the method comprises the steps: scanning an object through a three-dimensional laser scanning collection terminal, collecting point clouds at two different angles as a source point cloud and a target point cloud, and achieving high-robustness registration in a difficult scene with an extremely low overlapping rate; according to the method, through the asymmetric association strategy, the zero solution problem under the low overlapping rate is solved through the one-way matching union set, and the recall rate is remarkably increased; meanwhile, through a DGTG geometric pruning and probability manifold optimization mechanism, a voxel quantization error is eliminated by using anisotropic covariance and Lie algebra iteration, a translation error and a rotation error are greatly reduced, and a precision bottleneck caused by voxelization is effectively broken through. According to the method, the existence of a solution can be ensured under the condition that the overlapping rate is extremely low, voxel quantization errors can be eliminated through manifold optimization based on covariance, and high-precision and high-efficiency registration of low-overlapping-rate point clouds is achieved.

The invention provides a multi-source security knowledge fusion driven security risk management large model construction method and application. The method comprises the following steps: constructing a campus security knowledge graph and a semantic vector database based on multi-modal campus security data; performing approximate nearest neighbor search in the semantic vector database based on user query to obtain context evidence related to the user query, splicing the context evidence with the user query to obtain a retrieval enhancement prompt, the user query being campus security risk information of any mode; and training based on the campus security knowledge graph to obtain a security risk management large model, and inputting the retrieval enhancement prompt into the security risk management large model to obtain a risk solution. According to the scheme, the attention confidence, the feed-forward network confidence and the knowledge alignment confidence of all levels are aggregated in the reasoning process to generate the comprehensive confidence score, so that the illusion problem caused by no data reference in the model reasoning process is effectively avoided.

The invention relates to the technical field of digital archive management and information retrieval, and discloses a digital archive multi-modal data semantic enhancement fusion retrieval method and system.The method comprises the steps that a policy cycle time axis and a policy term evolution graph are constructed, tense logical reasoning is conducted on archive seals, and permission effectiveness evolution is derived; the temporal permission feature vector and the content semantic vector are fused to generate a multi-modal representation vector, and cross-policy-cycle semantic enhancement retrieval is realized by combining query expansion and temporal permission filtering, so that the problems of missing detection and misjudgment of policy and regulation archives in seal permission historical evolution and term cross-cycle retrieval are solved.

The invention discloses a power production event type identification and grade evaluation method based on artificial intelligence, and belongs to the field of power safety production, and the method comprises the steps: receiving event description information inputted by a user, extracting event key elements through a natural language understanding model, converting the event key elements into query vectors, and obtaining the query vectors; searching related knowledge fragments in a pre-constructed electric power professional knowledge base, combining the event description information, the key elements and the knowledge fragments into enhanced cue words, inputting the enhanced cue words into a large language model subjected to field fine tuning, and finally, performing comprehensive analysis by the large language model to obtain the enhanced cue words. And generating a structured research and judgment result containing the event type, the level, the research and judgment basis and the submission requirement. According to the method, accurate and efficient research and judgment of the electric power security event are realized by fusing the retrieval enhancement generation technology and the field fine tuning large model, and interpretability and traceability of a conclusion are ensured by outputting research and judgment basis through the requirement model.

This invention relates to a method, apparatus, electronic device, and storage medium for intelligent agent memory management. The method includes: acquiring interaction content between the intelligent agent and a user; performing semantic segmentation and metadata extraction to generate memory fragments; storing the memory fragments in a multi-level directory structure of a local file system, without deleting memory fragments already stored in the multi-level directory structure; constructing a time index, keyword index, semantic index, and association index for the memory fragments; and, in response to a retrieval request from the intelligent agent or user, retrieving memory fragments matching the retrieval request from the time index, keyword index, semantic index, and association index respectively to obtain a candidate memory set, performing fusion and sorting to obtain sorted memory data. This invention improves the efficiency of intelligent agents in accumulating and reusing knowledge and experience, and also enhances privacy protection and long-term memory reliability.

The invention relates to the technical field of tobacco plant identification and counting, and discloses an unmanned aerial vehicle tobacco plant identification and counting method and system based on improved deep learning, which combines technologies of image classification, image segmentation, feature point extraction, machine learning, target detection and the like with refined intelligent identification of tobacco. According to the method, high-precision image recognition and analysis are realized by utilizing the efficient feature extraction and calculation capability of the method, point plant counting can be accurately performed, the efficiency and precision of agricultural production are greatly improved, the problem of calculation complexity of a traditional method in large-scale high-resolution image processing is solved, and powerful technical support is provided for precision agriculture. Therefore, the invention provides an unmanned aerial vehicle tobacco plant identification and counting method based on improved deep learning target detection and multi-dimensional post-processing, and high-precision and robust detection and counting of tobacco plants in large-scale tobacco field orthoimages are realized by constructing a three-stage assembly line of image preprocessing, improved target detection and multi-dimensional post-processing.

This invention provides a method and apparatus for 3D target detection in autonomous driving, relating to the field of autonomous driving technology. The method includes: acquiring point cloud data collected by LiDAR; for each point in the point cloud data, defining a neighborhood of a set radius centered on the current point, counting the number of points within the neighborhood, and normalizing the number of points within the neighborhood to obtain a normalized value; determining the density feature of the current point based on the negative of the normalized value; extracting semantic features of each point in the point cloud data through multiple SDSA layers, predicting the foreground confidence of each point based on the semantic features, and performing point sampling by combining the density features and the foreground confidence to generate sampling points; aggregating the features of each sampling point to generate aggregated features for each sampling point, and predicting 3D target bounding boxes based on the sampling points and their aggregated features. This improves the sampling bias problem caused by the unevenness of the point cloud data.

The embodiment of the invention provides an image retrieval method, device and equipment for target image information. The method comprises the following steps: acquiring to-be-retrieved target image information; analyzing and extracting the target image information to obtain target image feature information; matching the target image feature information in a preset image data structured index database, and retrieving to obtain a retrieval result containing matched target image data and structured description information associated with the target image data; outputting a retrieval result; wherein the image data structured index database is obtained through the following processes: obtaining multi-modal original data containing images, videos and text descriptions of the same target object; performing content analysis and feature extraction on the multi-modal original data to obtain corresponding structured description information; and storing the structured description information as an index structure supporting feature-based matching retrieval according to a preset storage format. According to the embodiment of the invention, the deep semantic content of the image can be efficiently understood and accurately matched.

The invention discloses an approximate repetition detection method and system fusing local retrieval and multi-dimensional decision. The method comprises the steps that a to-be-detected document is preprocessed, and text content and service meta-information are extracted; utilizing MinHash to generate a compact signature; quickly recalling a candidate set from a massive historical library through an LSH index; adopting a paragraph-level weighted Jaccard algorithm to accurately calculate the structural similarity between the candidate document and the document to be detected; when the similarity exceeds a threshold value, intelligent decision making is carried out according to a preset priority chain considering multi-dimensional business rules such as release mechanism levels and time, and finally reserved documents are determined; and recording the structured audit log in the whole process for tracing. The method realizes approximate repetition detection with high recall, high precision, low time consumption and interpretable and auditing decision process, and is especially suitable for massive de-duplication scenes of strong normative texts such as government affair official documents, news announcements and the like.

The application discloses an image clustering method, an image clustering device and a computer storage medium. The image clustering method comprises the following steps: obtaining a portrait similarity of a first snapshot large image and a second snapshot large image; when the portrait similarity is less than a large image similarity threshold, performing image clustering on a person face snapshot small image pair of a person face snapshot small image pair with a person face similarity greater than or equal to a first person face clustering threshold and a person body snapshot small image pair with a person body similarity greater than or equal to a first person body clustering threshold; and when the portrait similarity is greater than or equal to the large image similarity threshold, performing image clustering on a person face snapshot small image pair of a person face snapshot small image pair with a person face similarity greater than or equal to a second person face clustering threshold and a person body snapshot small image pair with a person body similarity greater than or equal to a second person body clustering threshold. The image clustering method can adjust the similarity of the person face snapshot small image and the person body snapshot small image according to the similarity of the snapshot large image, thereby improving the recall rate of image clustering.

The invention discloses a small target detection method based on multi-core enhancement and multi-branch weighted fusion. A small target image detection network model based on RT-DETR improvement is researched and designed, an RT-DETR network is used as a backbone, and multi-scale features are extracted from shallow to deep; in a feature fusion network, a sub-pixel rearrangement downsampling convolution module (SRDC), a channel segmentation global multi-core enhancement module (CSGME) and a multi-branch adaptive weighted fusion module (MBAWF) are introduced to form an improved bidirectional feature fusion structure. The SRDC module retains fine-grained texture information while performing down-sampling; the CSGME module improves the feature global context representation capability through multi-scale large kernel convolution and a frequency domain attention mechanism; and the MBAWF module realizes multi-branch feature adaptive weighted fusion. Finally, the features are input into a decoder, target classification and bounding box regression are completed in a multi-scale parallel mode, small target high-precision detection is achieved, and the detection accuracy of the RT-DETR in a small target scene is effectively improved.

The invention belongs to the technical field of deep learning image classification, and provides a cancer tissue pathology image fine-grained classification method based on SMCNet. Comprising the following steps: S1, setting an SMCNet network model; s2, collecting a cancer pathology image fine-grained classification public data set, training a classification network based on the formed SMCNet network model, and storing training weights; s3, performing network performance evaluation on the training result of the SMCNet model by utilizing the evaluation indexes of the accuracy, the classification precision, the recall rate, the F1 score and the AUC score; and S4, in combination with the training weight and a result visualization program, reporting a classification prediction result of the test set, and visualizing the classification prediction result by using a thermodynamic diagram. The SMCNet model adopted by the invention has stronger feature extraction capability and high discrimination degree for fine-grained categories, space-channel features are subjected to DRA attention of feature extraction by using a self-attention module, and the design thought of model lightweight is considered on the basis of realizing an optimized feature extraction function.

The present application relates to the technical field of computer vision and road engineering, and particularly relates to a pavement crack lightweight segmentation and quantification method based on enhanced feature fusion. The method comprises the following steps: decoding and reading pavement images and performing verification, then performing grid division and cutting, and performing pixel normalization processing on the effective sub-images after cutting; a lightweight crack segmentation network is constructed; the preprocessed standard images are input into the lightweight crack segmentation network, and feature extraction, feature fusion, target detection and pixel segmentation are sequentially completed, and a crack detection frame and a binary segmentation mask are output; the lightweight crack segmentation network is trained, the optimal model is saved according to the iteration optimization of the verification set index; the segmentation mask output by the model is denoised, the crack geometric size is calculated, and the final detection quantification result is output. The method has the advantages that the small-scale crack recognition capability is improved, the crack length and width can be automatically and accurately measured, the crack size measurement error is small, and the method is suitable for large-scale pavement inspection operation.

The invention discloses an educational academic literature tracing method and system based on retrieval enhancement generation. The method comprises the following steps: performing fine-grained analysis on the multi-source literature, constructing a document object containing fields such as a title, an abstract, a methodology, an experimental result and a conclusion, and establishing a field-level index and a citation network feature score; identifying an academic intention queried by a user by utilizing a large model, dynamically loading a weight mapping table according to the academic intention, and performing weighted rearrangement on multiple paths of retrieval results to obtain candidate core literatures; generating a traceability text with an explicit label based on the candidate literature; and executing reference consistency verification by using the natural language inference model, and executing illusion correction according to a verification result. According to the method, through structured field analysis and dynamic rearrangement of intention driving, the problems of semantic fragmentation and reference fictition when a general RAG is used for processing complex academic literatures are solved, and the academic traceability preciseness and accuracy are remarkably improved.

The application discloses a rotating machinery fault diagnosis method based on delay-weighted visual graph and Netpro2vec, and belongs to the technical field of mechanical state monitoring and fault diagnosis. The method comprises the following steps: collecting a rotating machinery vibration signal; converting the signal into a delay-weighted visual graph, wherein, when judging whether two data points are visual, adjacent nodes of the two data points are ignored, and a weight is assigned to a visual edge, and the weight is defined as the minimum vertical distance between the two points and all data points therebetween; based on the graph, a low-dimensional feature vector is extracted by using the Netpro2vec method, specifically, a word sequence is generated by calculating a node distance distribution and a transition matrix, and a document is formed, and then a vector representation is obtained by learning through a document embedding model; and finally, the feature vector is input into a classifier to obtain a diagnosis result. The application enhances the noise resistance through a delay mechanism, and finely depicts the signal structure through a weighting mechanism, so that the precision and robustness of fault diagnosis are significantly improved.

The application provides a Chinese-oriented named entity recognition method and system, and belongs to the technical field of natural language processing; comprising: an encoding stage step, context encoding of input Chinese text; a feature extraction stage, a double-branch structure of global semantic branch and local boundary branch in parallel is adopted, wherein the local boundary branch carries out boundary feature enhancement through multi-granularity convolution combined with a learnable Gaussian Laplacian operator; a feature fusion and prediction stage step, global and local features are fused through a gating mechanism, and multi-label classification prediction is carried out based on a preset six-tuple grid label system; a decoding stage step, named entities are obtained through bidirectional search strategy decoding according to predicted relationship probability; through boundary perception feature extraction and enhanced label system, the recognition accuracy and recall rate of nested entities and discontinuous entities in Chinese text are significantly improved.

The application discloses a Chinese human phenotype ontology recognition method and related equipment, which can be applied to the technical field of text information processing. The application can effectively improve the recall rate by expanding the human phenotype ontology concept synonym set through a large language model, generating an expanded dictionary with the initial dictionary, significantly enhancing the discrimination ability of the recognition model for semantically ambiguous concepts through two-stage comparative learning, balancing the precision and recall rate by constructing an initial dictionary based on the human phenotype ontology official database and the Chinese human phenotype ontology official data, merging and disambiguating the current input text after accurate matching and semantic similarity matching to obtain a fused candidate result list, and finally obtaining the structured recognition result corresponding to each current input text by post-processing each candidate fragment in the fused candidate result list, which can provide effective data support for disease diagnosis.

The application belongs to the technical field of log auditing, and discloses a numerical control system log auditing method and terminal based on LSTM, which comprises the following steps: a new log analysis tool based on the Drain log analysis tool is used to reconstruct three types of log exclusive analyzers, and log analysis is performed into template statements and variables; the log template statements are correspondingly converted into log keys; a bidirectional long short-term memory model is used to identify the log mode; and through the prediction of the log keys within a certain window size and the calculation of the Gaussian error of the generated log, the time sequence-based abnormal detection of the log is performed. The application shows high precision on the classical data set, and the accuracy, recall rate and F value are higher than those of the traditional method. The application improves the precision of log analysis, perfects the system log processing procedure, realizes full-automatic log analysis, realizes two main abnormal models in the system, tests the hyperparameters of the model, and perfects the procedure details of log analysis to input model.

The invention relates to the technical field of protocol analysis, in particular to a network protocol collaborative combination analysis method based on a tree hierarchical structure, which comprises the following steps of: acquiring different network protocol analysis requirements, training a basic network protocol analysis model and a coarse granularity model, and acquiring a hierarchical structure by analyzing the coarse granularity and the fine granularity of application requirements. The method comprises the following steps: identifying and combining network protocol analysis models of the same hierarchy, constructing a tree-shaped layered structure, extracting a model identification result and an operation relationship in the tree-shaped layered structure, designing a two-dimensional mapping table, guiding input traffic through the two-dimensional mapping table, analyzing the input traffic layer by layer according to the tree-shaped structure, and positioning the next step through an operation result of each step until requirements are met. And an efficient collaborative analysis process is obtained. According to the method, the problems of high model training difficulty, high resource consumption and accumulation of errors along with layer propagation existing in a single network protocol analysis algorithm for multi-granularity and multi-layer hybrid application requirements are solved.

The invention discloses a water conservancy knowledge base and intelligent question and answer construction method based on a large model and RAG, and the method comprises the steps: 1, combing a water conservancy knowledge system, and sorting water conservancy professional documents and data as materials of the knowledge base; step 2, constructing a water conservancy knowledge library management system through a J2EE system, a B / S architecture and a distributed object storage system; step 3, based on the RAG technology, forming different knowledge bases according to the sorted knowledge classification and confidentiality security level of the water conservancy industry, and establishing a water conservancy knowledge intelligent question-answering system on the knowledge bases; 4, performing extended application on the basis of the constructed water conservancy knowledge intelligent question-answering system; the problems that in the prior art, due to the fact that unstructured data processing is weak when a water conservancy knowledge base is constructed, a large number of image-text mixed documents such as PDF and PPT cannot be effectively utilized, and the recall rate and the accuracy rate are not high are solved.

The invention discloses a method for detecting small defects on the surface of lightweight steel, and belongs to the technical field of industrial machine vision. The method is based on an improved YOLOv8n architecture and cooperatively works through three core technical means: firstly, a Swin Transform module is introduced into a backbone network to model long-distance spatial dependence and suppress complex background interference; secondly, a high-resolution P2 detection branch is constructed, shallow details and up-sampling semantic features are fused, and microdefect characterization is enhanced through bidirectional refining circulation; finally, P2 exclusive adaptive threshold focus loss ATFL is adopted, threshold update is only limited to P2 detection branch samples, and precise optimization of difficult and tiny defects is achieved in cooperation with a gradient directional return mechanism. According to the scheme, the detection recall rate and the positioning precision of the tiny defects are effectively improved while the model parameter quantity is remarkably reduced, and high-precision and light-weight industrial deployment is realized.

The present application relates to the field of computer image processing, in particular to a feature enhancement and self-adaptive regression tea leaf picking point positioning method and system. The basic principle of the method is: firstly, the collected tea bud image is preprocessed for clarity; then, a two-stage model architecture of detection first and then positioning is adopted, an improved YOLOv5 network is used in the detection stage to robustly detect multi-scale buds and suppress background interference; then, the target suitable for picking is screened out and cut into a single bud image; in the positioning stage, an improved YOLOv11-Pose network integrated with an adaptive convolution kernel module is used to accurately regress the picking point coordinates; finally, the coordinates are mapped back to the original image and output. The core technical effect of the present application is: through the synergistic optimization of the two-stage process and the targeted improvement of the model components, the problem of inaccurate picking point positioning and poor robustness caused by image degradation, multi-scale targets, complex background and variable bud morphology in the natural environment is effectively solved, providing a high-precision solution for tea leaf automatic picking.

The present disclosure provides an audio recall method, a model training method, a device and an electronic device, relates to the technical field of cloud computing, in particular to the technical field of deep learning, intelligent search and voice technology, and the audio recall method comprises: acquiring a first audio; segmenting the first audio to obtain N audio segments, any two adjacent audio segments in the N audio segments partially coincide, and N is an integer greater than 1; recalling a second audio corresponding to each of the N audio segments from a sample pool.

The application relates to the field of information technology, in particular to an AI-based advertisement material multi-modal intelligent retrieval method and system. The method comprises the following steps: receiving multi-modal advertisement materials, and extracting content descriptions of the advertisement materials; converting the multi-modal advertisement materials and the content descriptions into unified multi-modal joint representation vectors by using a multi-modal fusion model, and constructing a material index library; receiving demand information of a new advertisement provided by a user, wherein the demand information comprises at least one of a content description, a target audience description and a delivery scene description; extracting embedding vectors of the demand information, performing retrieval in the material index library according to the similarity between the embedding vectors and the multi-modal joint representation vectors, and obtaining advertisement materials with a similarity higher than a preset threshold as candidate results; and generating a predicted effect tendency of the candidate results, and outputting the candidate results as retrieval results after screening according to the predicted effect tendency.

The invention discloses a UI front-end code intelligent generation method and system based on mixed layout analysis and prefix tree index. According to the method, traditional image processing + light CNN + character detection three-channel mixed layout analysis is sequentially executed at a browser end, and a rectangular frame set F of UI screenshots is obtained; a Z-order character string signature S of the F is used as a key, and a compressible prefix tree index is constructed in a memory; after the screenshot is uploaded, the screenshot is matched with the longest prefix of the two channels through elastic deformation compensation, if the matching length is larger than or equal to 80%, a code is directly returned, otherwise, improved IOU secondary fine arrangement is started, and a Top-1 code is selected and output; and meanwhile, user data security is guaranteed through differential privacy and AES page-level encryption, and online evolution and edge caching are supported. The first loading time of the system is less than or equal to 300ms, the end-to-end code generation time is less than or equal to 260ms, and the query time is lt; compared with the prior art, the speed is increased by three times, the accuracy rate is increased by 6.3%, and the method can be completely operated in an edge server-free environment.

The application discloses a knowledge retrieval method and system based on vector and graph fusion, a device and a medium, relates to the technical field of knowledge retrieval, and obtains natural language query data; encodes the natural language query data into query vectors; performs semantic retrieval on the query vectors to obtain candidate nodes and acquires semantic similarity scores of the candidate nodes; generates structure scores of the candidate nodes according to structural features of the candidate nodes in a graph database; applies a gating coefficient to fuse the semantic similarity scores and the structure scores of the candidate nodes to generate fusion scores; screens seed nodes from the candidate nodes according to the fusion scores; and applies the seed nodes to generate target retrieval results of the natural language query data. The application fuses the semantic similarity scores and the structure scores of the candidate nodes through the gating coefficient, thereby screening seed nodes that are both semantically relevant and important in the graph structure, accurately aligning user intentions, and further obtaining accurate target retrieval results.

The embodiment of the invention provides a student behavior data set labeling method, and the method comprises the steps: carrying out the seat region detection of a first image of a target classroom through employing a trained target detection model, and obtaining a seat detection frame set, and the first image comprises seats disposed in the target classroom; based on the shooting view angle of the first image, each seat detection frame in the seat detection frame set is allocated to a grid position formed based on a preset row number R and a preset column number C to generate a structured seat template, and R and C are integers greater than or equal to 1; a second image of the target classroom is matched with the structured seat template, student behaviors corresponding to each seat are identified, and the second image comprises students in the target classroom; and identifying the second image based on the behavior of the student corresponding to each seat, and adding the identified second image to the data set of student behavior identification.

The application discloses a video information acquisition method, device and equipment and a storage medium. The method comprises the following steps: acquiring video data; wherein the video data comprises a video and initial text information of the video; performing entity recognition on the initial text information to obtain entity information; obtaining supplementary text information of the video by using the entity information; taking the initial text information and the supplementary text information as associated information of the video; wherein the associated information is used for being saved into a movie and television library together with the video, and the initial text information and the supplementary text information are used for searching the video in the movie and television library. In the above manner, the recall rate and the relevance of the video can be improved.

This invention discloses an image segmentation method and medium for densely populated live cell regions. The method includes acquiring multiple first data pairs and preprocessing each first data pair. Each first data pair includes a single-modal master image of a live cell region and a paired real auxiliary modality image. A feature extraction network is trained using the preprocessed first data pairs, and the feature extraction network outputs a predicted auxiliary modality image corresponding to the single-modal master image. The preprocessed first data pairs and the predicted auxiliary modality image are input into a pre-trained segmentation model, and the dynamic weights of the predicted auxiliary modality image are adjusted to optimize a second loss function to obtain optimal weight parameters. The single-modal master image of the live cell image to be segmented is input into the trained feature extraction network, and combined with the optimal weight parameters, it is input into the pre-trained segmentation model to obtain the segmentation result. This method is particularly effective for accurate segmentation in densely populated regions.