Patents
Literature
Patsnap Eureka AI that helps you search prior art, draft patents, and assess FTO risks, powered by patent and scientific literature data.

604results about How to "Improve recall" patented technology

RAG-based vertical domain knowledge multi-round question and answer method

The invention belongs to the technical field of artificial intelligence, and discloses an RAG-based vertical domain knowledge multi-round question-answering method, which is based on reconstruction of documents in a knowledge base and construction of dense and sparse vector databases, assists in understanding and expanding original questions of customers, and realizes effective processing of the questions. And retrieving from the vector database by utilizing hybrid query to obtain a plurality of small document blocks and metadata information, sorting the small document blocks by utilizing a rearrangement model, and performing duplicate removal and merging processing on the small document blocks after correlation sorting. By means of the method, professional complex problems in the vertical field can obtain effective semantic splitting and hierarchical association, the retrieval success rate of problem related semantics during retrieval is increased, and the accuracy rate and recall rate of original text reference are increased; meanwhile, the problems that historical dialogues are mixed with useless information, and the real intention of a user is submerged by too long document block content are effectively avoided, and efficient vertical domain knowledge multi-round questions and answers can be achieved.
Owner:NEUTON HEALTH (WUHAN) CO LTD +1

Retrieval enhancement generation-based retrieval method, product, equipment and medium

The invention discloses a retrieval method based on retrieval enhancement generation, a product, equipment and a medium, and relates to the technical field of natural language processing. According to the scheme, an entity in initial query information is identified, so that a system can be helped to capture key information and a semantic structure in user query, and the accuracy of query understanding is improved; entities in the initial query information are expanded based on the knowledge graph, so that implicit background information can be supplemented, and a system can better process incomplete or fuzzy queries; due to the introduction of the knowledge graph, the system can better understand and process the query of the professional field by utilizing the specific structured knowledge of the field; and meanwhile, the knowledge graph can be conveniently expanded or replaced into professional knowledge bases in different fields, so that the adaptability of the system is improved. In the semantic matching process, keyword matching and semantic matching are combined, the query intention of the user and the content of the text block to be retrieved can be more comprehensively understood through multi-layer semantic matching, and the retrieval precision is remarkably improved.
Owner:LANGCHAO ELECTRONIC INFORMATION IND CO LTD

Scientific and technical literature flow chart entity and relation extraction method based on retrieval enhancement

The invention provides a scientific and technical literature flow chart entity and relation extraction method based on retrieval enhancement, and relates to the technical field of artificial intelligence. A visual encoder is supervised and trained, a vector knowledge base is constructed for the literature database, the target flow chart is received, vector conversion and similarity retrieval matching are carried out, and an image-text matching result is determined; and supervising and training the multi-modal large model, transmitting an image-text matching result to the multi-modal large model, executing identification interpretation and relation extraction, and performing structure conversion to generate a target knowledge graph. The technical problems that a traditional method depends on manual data labeling, time and labor are wasted, the processing requirements of a large number of scientific and technical literatures are difficult to meet, automatic and intelligent extraction means are lacked, and the efficiency is low when large-scale literatures are processed are solved.
Owner:DOCUMENT & INFORMATION CENT OF CHINESE ACAD OF SCI

Document knowledge question-answering method, document knowledge storage method and document knowledge question-answering device

The invention provides a document knowledge questioning and answering method, a document knowledge storage method and a document knowledge questioning and answering device, and relates to the technical field of intelligent questioning and answering. All layers of nodes of a knowledge tree are traversed according to a target retrieval mode on the basis of the query statement to obtain a query result, the knowledge tree comprises multiple layers of nodes, the multiple layers of nodes at least comprise first-layer nodes, second-layer nodes and third-layer nodes, semantics of the first-layer nodes are semantics of text contents of knowledge point levels in the document, and semantics of the third-layer nodes are semantics of text contents of knowledge point levels in the document; the semantics of the second-layer nodes are the semantics of the paragraph-level text content in the document, and the semantics of the third-layer nodes are the semantics of the full-text-level text content in the document; and outputting an answer text corresponding to the query statement based on the query result. The method provided by the invention not only improves the recall rate of knowledge and the correlation of answers, but also enhances the overall experience of the user, so that the user can obtain more comprehensive and accurate answers in complex queries.
Owner:CHENGDU CELIS TECH CO LTD

Room obstacle target detection method and system based on multi-modal information

The invention provides a room obstacle target detection method and system based on multi-modal information, and relates to the technical field of target detection, and the method comprises the steps: obtaining an RGB image and a depth image of a to-be-detected room, extracting the visual features of the RGB image, extracting the depth features of the depth image, and carrying out the feature coding of the depth image, splicing the extracted visual features and depth features in the channel dimension to obtain a fused feature map, and obtaining a depth enhanced feature map based on the fused feature map and the coded depth image; pixel-level segmentation is carried out on the depth enhancement feature map through a full convolutional network, a semantic segmentation result of the obstacle target is obtained, candidate regions are generated through a region suggestion network, target classification is carried out, and an initial obstacle target detection result is obtained. And according to the obtained obstacle target semantic segmentation result and the initial obstacle target detection result, removing overlapped detection frames through a non-maximum suppression algorithm to obtain a final obstacle target detection result.
Owner:BEIJING XINGWANG SHIP POWER TECH CO LTD

Method, system and equipment for retrieval enhancement generation based on large language model and medium

The invention discloses a retrieval enhancement generation method, system and device based on a large language model and a medium. The method comprises the steps that a text file is fragmented to obtain a plurality of parent text blocks, a unique identifier is generated for each parent text block, and all the parent text blocks and the corresponding unique identifiers are stored in a relational database; fragmenting each parent text block again to obtain a plurality of child text blocks, and performing text summarization on each parent text block to obtain a plurality of parent text block summarized texts; generating one or more hypothetical questions for each parent text block; vectorizing the sub-text blocks, the summary text of the parent text block and the hypothetical question of the parent text block, and storing in a vector database; querying in the vector database and extracting the unique identifier based on a query problem text input by a user to obtain a plurality of parent text blocks; and forming prompt words by the plurality of parent text blocks and questions proposed by the user, and generating a final answer. The accuracy and recall rate of semantic retrieval are improved.
Owner:NO 30 INST OF CHINA ELECTRONIC TECH GRP CORP

Test case generation method and system based on large model and retrieval enhancement generation

The invention provides a test case generation method and system based on a large model and retrieval enhancement generation, and the method comprises the steps: obtaining a software description document, carrying out text fragment segmentation and vectorization, and constructing a software knowledge base; obtaining and vectorizing a demand document, and recalling a content optimization demand document in the reference software knowledge base based on the retrieval enhancement generation framework; constructing a first Prompt instruction, and splitting the optimized demand document into demand fragments of different test points by using a large language model; and finally, constructing a second Prompt instruction for the demand fragment of each test point, generating a test case by using a large language model, and establishing a mapping relationship between each test point and the test case for linkage output. According to the method, the understanding ability of the large model on the demand document is effectively enhanced; meanwhile, the requirements are decomposed into test points, mapping is carried out on the test points and related test cases, and completeness and accuracy of the requirements and the test cases can be conveniently sorted.
Owner:GUANGDONG UNIV OF TECH

Multi-mode RAG knowledge question-answering method and device applied to vertical field

The invention provides a multi-modal RAG knowledge question answering method and device applied to the vertical field, relates to the technical field of artificial intelligence, supports multi-modal input of to-be-answered texts including texts, pictures, audios and videos, and makes full use of semantic information of multi-modal data. And utilizing a plurality of preset retrieval models to retrieve the knowledge fragments corresponding to the to-be-answered text to obtain a plurality of retrieval results. Various retrieval results are fused through an RRF algorithm, and scores are calculated by using a target reordering model for reordering, so that the recall rate is improved. And based on the retrieval content and the preset knowledge frequency information, fine tuning sorting is further carried out, so that the method is more accurate and meets user expectation. According to the method, the natural language understanding ability of a large language model, a target reward model and related information recalled by the RAG are combined to generate optimal solution data, and meanwhile, a target scoring model and domain expert knowledge are utilized to provide more professional high-quality solution data meeting actual requirements.
Owner:LINKAGE TECHNOLOGY (NANJING) CO LTD

Intelligent production scheduling early warning method and system for hardware machining

The invention relates to the technical field of intelligent production scheduling early warning methods, in particular to an intelligent production scheduling early warning method and system for hardware machining, and the method comprises the following steps: obtaining real-time production scheduling operation data on a hardware machining production line; preprocessing the real-time production scheduling operation data to obtain preprocessed data; performing dynamic time window division based on the preprocessed data; performing multi-scale feature extraction on the divided time window data; performing abnormal mode recognition on the extracted multi-scale features by using a quantum inspiration depth self-encoding network; based on the abnormal mode recognition result, abnormal reason diagnosis is carried out; early warning information is generated and pushed; and the model is updated online according to feedback information, a complex abnormal mode can be accurately identified through an innovative quantum inspiration deep self-encoding network, and the accuracy, the precision rate and the recall rate of anomaly detection are greatly improved. Test results show that the accuracy rate reaches 98.7% and is far higher than that of a traditional method.
Owner:QINGDAO HAINAKUN TRADING CO LTD

RAG-oriented multi-branch hybrid balance retrieval optimization method and system

The invention discloses an RAG-oriented multi-branch hybrid balance retrieval optimization method and system, and relates to the technical field of RAG retrieval, and the method comprises the steps: extracting a semantic vector of a user question, carrying out the similarity matching of the semantic vector of the question and the semantic vector of each text in a knowledge base, and generating a first candidate text set; extracting keywords in the user question, performing secondary filtering to obtain a keyword list, and performing keyword matching on the keyword list and each text in the knowledge base to generate a second candidate text set; combining the first candidate text set and the second candidate text set, performing duplicate removal, and inputting the first candidate text set and the second candidate text set into a Reranker model to generate a third candidate text set; judging whether the third candidate text set is empty or not, if not, outputting the third candidate text set as a retrieval result, otherwise, performing secondary retrieval, and re-determining and outputting the retrieval result; and sending the user question and the retrieval result to the large language model. According to the method and the device, the accuracy of RAG retrieval can be improved, and the user experience is improved.
Owner:山东省大数据中心

Lightweight track surface defect real-time detection method and device

The invention discloses a lightweight track surface defect real-time detection method and device, relates to the technical field of image data processing, and improves the detection precision and detection efficiency of track defect detection by optimizing a model structure and a training strategy. The method comprises the following steps: pre-training a T-YOLOv8 teacher model by using a data set of a track image; the T-YOLOv8 teacher model is obtained by taking an original YOLOv8 model as a basis, introducing an NEW-IAT module into a backbone network and replacing a neck network with a CBAM-C2f module; carrying out distillation training on the lightweight S-YOLOv8 student model by using the trained T-YOLOv8 teacher model, and carrying out optimization and verification on the trained lightweight S-YOLOv8 student model; the lightweight S-YOLOv8 student model is based on an original YOLOv8 model, an SPPF layer is reserved, the rest part of a backbone network is replaced by a lightweight new backbone network, and the lightweight new backbone network comprises three modules of HGStem, LightHGBlock and DWConv; and the optimized lightweight S-YOLOv8 student model is used to carry out real-time detection on track surface defects to obtain track defect information.
Owner:KUNMING UNIV OF SCI & TECH +2

Operation site safety management and control method and system based on multi-modal large model

The invention discloses an operation site safety management and control method and system based on a multi-modal large model, and the method comprises the steps: collecting site violation data, and carrying out the preprocessing, and obtaining multi-modal data; constructing a multi-modal large model; carrying out training and optimization on the multi-modal large model by utilizing the preprocessed multi-modal data; and using the optimized multi-modal large model to realize operation site safety management and control. According to the invention, a visible light image, a safety violation knowledge base and other modal data are fused, and a comprehensive safety monitoring and management platform is constructed by using a large model and a big data processing technology, so that real-time monitoring, early warning and intelligent response of potential safety hazards of a complex operation site are realized, and the safety management level of the operation site is improved.
Owner:CHINA ELECTRIC POWER RESEARCH INSTITUTE CO LTD

Multistage multi-mode fused OLED defect detection device

The invention discloses a multi-stage and multi-mode fused OLED defect detection device, which aims at the problem that the precision of a traditional image-based detection scheme is limited, innovatively introduces polarization and multi-spectral information in an OLED image acquisition stage, enhances defect characteristics, further improves the detection capability of a system for fine defects, and aims at the defects of an OLED screen body, which are rich in variety and high in accuracy. According to the invention, the problems of complex and changeable shapes and incapability of complete detection by a single device are solved, and different types of defects are detected by stages by adopting a double-point detection system from the cause of formation and material characteristics of the defects. Detecting cracks, scratches, foreign matters and fragment defects by using a detection point I based on polarization fusion according to the characteristic that the polarization information is sensitive to the roughness and morphological characteristics of the surface of an object to be detected; by utilizing the characteristic that bubbles and smudginess are sensitive to spectral characteristics, the device II based on multispectral fusion is used for detecting the defects of bubbles and smudginess on the OLED screen body.
Owner:ZHEJIANG UNIV

Vector indexing method and device, electronic equipment and readable storage medium

The invention relates to the technical field of vector indexing, and provides a vector indexing method and device, electronic equipment and a readable storage medium. The method comprises the following steps: inputting a to-be-queried vector into a target index structure, performing hierarchical neighbor graph index processing on the to-be-queried vector, and determining a clustering center corresponding to at least one to-be-queried vector; obtaining a corresponding mapping relationship from a preset mapping table based on the clustering center corresponding to each to-be-queried vector, and determining at least one target cluster according to each mapping relationship; performing reverse indexing processing on the at least one target cluster and the to-be-queried vector to obtain a target result vector sequence; and carrying out vector selection processing on the target result vector sequence to obtain at least one target return result vector, so that the vector query efficiency is improved, the query result accuracy is improved, and the expansibility and flexibility of a vector indexing process are enhanced.
Owner:BEIJING LONGZHI DIGITAL TECH CO LTD

Aluminum alloy handle surface defect detection method, device and equipment and storage medium

The invention relates to an aluminum alloy handle surface defect detection method, device and equipment and a storage medium. The method comprises the following steps: respectively acquiring standard illumination original images and low illumination original images of a plurality of angles of the surface of the aluminum alloy handle, and preprocessing to obtain a preprocessed image pair of each angle; processing is carried out through three parallel convolution branches and residual connection, and an enhanced feature map of each angle is obtained; performing illumination compensation processing to obtain an illumination equalization image of each angle; performing optimization to obtain a de-noised image of each angle; and splicing and reconstructing the de-noised image of each angle to obtain a reconstructed surface image, and carrying out multi-scale feature fusion and surface defect detection on the reconstructed surface image to obtain a surface defect detection result. According to the method, the precision and recall rate of defect positioning on the surface of the aluminum alloy handle are improved.
Owner:SHENZHEN JINYUREN TECH CO LTD

Track plate crack segmentation method and system based on multi-scale attention mechanism

The invention provides a track plate crack segmentation method and system based on a multi-scale attention mechanism, belongs to the technical field of image segmentation based on computer vision, and aims to solve the problem that the pixel proportion of a track plate crack in an unmanned aerial vehicle image is small. A neural network pre-trained on a large-scale data set is used as an encoder, extraction and representation of crack detail information by a model are improved, a feature fusion module based on a mixed attention mechanism is adopted, the model can eliminate interference of background redundant information and focus on crack features so as to improve track slab crack segmentation precision, and track slab crack segmentation accuracy is improved. And Focal loss and Dice loss functions are introduced to solve the problems that positive and negative samples are unbalanced and the positive samples are difficult to learn in a crack segmentation task.
Owner:BEIJING JIAOTONG UNIV

Multi-semantic feature collaborative Chinese address matching method

The invention discloses a Chinese address matching method based on multi-semantic feature collaboration. According to the method, an address matching problem is converted into a dichotomy problem. The method comprises the following steps: firstly, carrying out address element analysis on a Chinese address, and training a Word2vec model by utilizing a training corpus to generate a word embedding vector representation corresponding to the Word2vec model; then, extracting multi-semantic features of the address elements by using a text cyclic convolutional neural network and a graph attention network respectively; and finally, performing local reasoning and reasoning combination on the address multi-semantic features based on an enhanced sequence reasoning model to realize accurate matching of Chinese addresses. The method is remarkably superior to an existing Chinese address matching method based on deep learning, excellent results are obtained in the matching accuracy, the recall rate and the F1 value, and the method is of great significance to application such as geocoding.
Owner:LANZHOU JIAOTONG UNIV

Abnormal network traffic identification method based on small sample data enhancement

The invention provides an abnormal network traffic identification method based on small sample data enhancement, and the method comprises the steps: S1, converting one-dimensional traffic data into a two-dimensional gray-scale map, and employing a twin network combined with a dynamic attention mechanism to analyze the differences between various types of traffic data pair by pair; parameters of an attention mechanism in the twinborn attention network are analyzed, and inter-class difference features are identified; s2, selecting a training domain by using the inter-class feature difference matrix, introducing a comparative learning method to judge the difference between real data and generated data, and generating an enhanced data set for each class of traffic; s3, after the enhanced data set issued by the central server is received, each edge server uses the enhanced data set and local sample data to train the twinborn attention network in the step S1; due to different local sample data sets, each edge server is trained to obtain a personalized abnormal traffic identification and classification model suitable for local data distribution, and abnormal traffic is effectively classified.
Owner:NORTHEASTERN UNIV CHINA

Threat intelligence entity detection method based on Bert and deep learning model

The invention relates to a threat intelligence entity detection method based on Bert and a deep learning model, and the method comprises the steps: firstly, capturing the basic semantic information of a text through a pre-trained BERT model, and constructing a grammatical structure diagram; then, sending the grammar structure diagram into a diagram attention network for processing, and analyzing a complex dependency relationship between entities; meanwhile, a CLS vector output by the BERT model and a global vector obtained through Text-CNN processing are spliced, and an HCV containing global context information and local detail features is formed; in addition, obtaining word time sequence context information and importance association between entity words; and finally, fusing vectors from different modules, putting the fused vectors into a conditional random field layer for entity identification, and obtaining output of a threat entity. The method shows more excellent performance when processing terminologies and complex contexts in the field of network security.
Owner:SHANDONG COMP SCI CENTNAT SUPERCOMP CENT IN JINAN +1

Intelligent anomaly detection system and method for time series data based on large model

The invention discloses an intelligent anomaly detection system and method for time series data based on a large model, and the system comprises a feature analysis module which is used for analyzing the periodicity, drifting performance and stability of the time series data; the algorithm type selection module is used for selecting a proper algorithm according to a skewness value of data distribution and adjusting algorithm parameters according to periodicity, drifting and stability characteristics; the model training module is used for supporting dynamic loading and parallel training of different anomaly detection algorithms by adopting an algorithm factory mode; for periodic data, model training is enhanced through segmentation and aggregation processing; for data drift, enabling the model to gradually adapt to data change by adopting an incremental learning method; and the real-time detection module is used for continuously receiving real-time data, loading the trained model for anomaly detection after preprocessing, and immediately outputting a result once an anomaly is detected. According to the system and the method, the accuracy and the timeliness of anomaly detection in a dynamically changing service environment can be improved.
Owner:CHINA UNITECHS

Hospital infection risk prediction method and system based on random forest algorithm

The invention discloses a hospital infection risk prediction method and system based on a random forest algorithm, and relates to the technical field of risk assessment, and the method comprises the steps: obtaining the physiological features of a historical patient at a preset observation starting point, and obtaining historical physiological information; acquiring an infection outcome of a historical patient at a preset observation terminal point, and obtaining a historical infection outcome; wherein the infection outcome comprises an outcome label, and the outcome label is used for representing whether a hospital infection event occurs in the corresponding historical patient before the preset observation terminal point; constructing a risk prediction model based on a random forest, wherein the risk prediction model is used for predicting the corresponding outcome label according to the physiological features of the patient; training the risk prediction model according to historical physiological information and historical infection outcomes; acquiring target physiological information according to the physiological features of the target patient; and inputting the target physiological information into the trained risk prediction model to obtain a target outcome label so as to estimate the risk of hospital infection of the patient in advance.
Owner:SICHUAN ACADEMY OF MEDICAL SCI SICHUAN PROVINCIAL PEOPLES HOSPITAL +1

Lightweight wisdom classroom learning behavior detection method and system

The invention belongs to the technical field of computer vision and artificial intelligence, and discloses a lightweight wisdom classroom learning behavior detection method and system, and the method comprises the steps: extracting student behavior features in a classroom image, and obtaining preliminary multi-scale features; designing a deep convolutional neural network to perform deep spatial-temporal feature representation on the extracted features, and introducing a focusing modulation module to enhance the features; designing a dynamic detection head to fuse and process features of different scales; and calculating the difference between a prediction result and a real label by using the optimized classification loss function and bounding box loss function, optimizing model parameters, and outputting a final student behavior detection result through the model. And finally, the system performs detailed analysis on the detected student behaviors and generates a visual classroom behavior report. According to the method, more accurate student behavior detection in a classroom scene is realized, a complex classroom environment and diversified student behavior modes can be effectively processed, and the method has relatively high practicability.
Owner:HUAZHONG NORMAL UNIV

Multi-dimensional combined API gateway flow control method and system

The invention discloses a multi-dimensional combined API gateway flow control method and system, and relates to the technical field of flow control, and the method comprises the steps: collecting the monitoring data of an API gateway and the resource utilization data of a node, collecting the context information when the flow is abnormal, generating a first feature set, and calculating the feature importance parameter of each feature; the features with the feature importance parameters larger than a feature selection threshold value serve as key features, an anomaly feature matrix is generated, an anomaly classification model and an anomaly influence range prediction model are trained, the anomaly type and the influence range of current traffic anomaly are predicted, a traffic control strategy is designed, and a strategy historical performance score is calculated. The method comprises the following steps: calculating a strategy fitness score of each flow control strategy, calculating a strategy income and a strategy failure risk of each flow control strategy, calculating a strategy score value of the flow control strategy for the current flow abnormity, and taking the flow control strategy with the highest strategy score value as an optimal flow control strategy, thereby realizing control and self-healing of the flow abnormity.
Owner:JIANGSU FINSTONE TECH CO LTD

Improved YOLOv8 chip surface defect detection model and construction method and application thereof

The invention discloses an improved YOLOv8 chip surface defect detection model construction method and application, and the method comprises the steps: constructing a network structure of an improved YOLOv8 target detection model, which comprises a Backbone module, a Neck module and a Head module; the Backbone module comprises a multi-layer DCNv3 network, a C2f network and an SPPF network, and outputs a first effective feature layer, a second effective feature layer and a third effective feature layer after performing multi-layer convolution processing on the picture; the Neck module adopts a BiFPN bidirectional feature pyramid feature fusion structure, inputs three effective feature layers output by the Backbone module into a ConcatBiFPN network for bidirectional adaptive fusion, and then outputs the three effective feature layers through a CBAM attention mechanism network; the Head module comprises a Deect network structure, and the Deect network structure is connected with a CBAM attention mechanism network and outputs a target detection result; the improved YOLOv8 target detection model structure is used for accurately recognizing the surface defects of the chip, and the adaptability to different types of defects is improved.
Owner:JIANGSU UNIV

Method for improving recall rate of large model retrieval enhancement system based on knowledge density

The invention provides a method for improving the recall rate of a large model retrieval enhancement system based on knowledge density. The method aims at solving the challenges of a traditional retrieval enhancement system in the aspects of document truncation, context integration and the like. The method comprises the following steps: firstly, intelligently segmenting a document into text blocks with medium lengths through an intelligent document segmentation technology, marking whether the text blocks and the text blocks can form a context or not, and further segmenting the text blocks into shortest text blocks; secondly, performing text vectorization processing by applying an embedded model, performing dense and sparse vector calculation on a text, and storing a result and a relationship between text blocks into a database; in addition, through preliminary mixed retrieval and score calculation based on knowledge density, most relevant medium-length text blocks are recalled accurately, a long document is formed, and background knowledge is formed. According to the method, accurate related contents can be provided, context information can be integrated, the recall rate of a large model enhanced retrieval system can be remarkably improved, and good balance is achieved between complexity and performance.
Owner:SOUTHEAST UNIV

PET waste plastic intelligent sorting method and system based on multi-feature fusion

The invention discloses a PET waste plastic intelligent sorting method and system based on multi-feature fusion, and the method comprises the following steps: collecting an RGB-HSI original data set of PET waste plastic, carrying out the dimension reduction processing of the HSI data set based on a genetic algorithm to generate MSI data, and forming a feature fusion data set with RGB data; constructing a feature fusion network based on a YOLACT model, and training by adopting a feature fusion data set; the feature fusion network comprises an RGB branch, an MSI branch and a fusion prediction module, RGB is used for extracting features of RGB data, the MSI branch is used for extracting features of MSI data, and the fusion prediction module is used for combining output of the RGB branch and output of the MSI branch to make comprehensive prediction; and installing the trained feature fusion network to an industrial personal computer to realize intelligent sorting of the PET waste plastics. Compared with a traditional recognition algorithm, the method can improve the recognition precision.
Owner:ZHANGZHOU LUHAI ENVIRONMENT PROTECTION IND DEV CO +1

Engineering structure crack detection and measurement method based on machine vision

The invention belongs to the technical field of machine vision and deep learning, and discloses an engineering structure crack detection and measurement method based on machine vision, and the method comprises the steps: collecting original image data of an engineering structure, discretizing the original image data into image blocks, carrying out the preprocessing, building a crack classification model through a Resnet-50 neural network, and carrying out the training, the method comprises the following steps of: firstly, constructing an image block, identifying a crack image in the image block, then constructing a SegFormer segmentation model based on a Transformer framework to segment the crack image to obtain a crack mask graph, evaluating the performance of the two models, then analyzing the two models by adopting an orthogonal skeleton line method, obtaining a contour line and a skeleton line of a crack, and calculating the length of the crack. According to the method, the ResNet-50 image classification model and the SegFormer segmentation model are combined to realize high-precision crack identification and segmentation of the building structure crack, the contour line and the skeleton line of the crack are obtained and the length of the crack is measured by means of the orthogonal skeleton line method, and the crack of an engineering structure is well detected and subjected to risk assessment.
Owner:SANYA SCI & EDUCATION INNOVATION PARK WUHAN UNIV OF TECH +1

Chinese named entity recognition method based on stacked grid structure information enhancement

The invention discloses a Chinese named entity recognition method based on stacked grid structure information enhancement, and belongs to the technical field of natural language processing and named entity recognition. A stacked grid structure and a hierarchical compressed sensing module are used for introduction and lossless compression of word vectors, a long input sequence and information loss are avoided, and the recognition efficiency is improved. Meanwhile, the vocabulary fusion process and the context feature coding process are decoupled to reduce unnecessary attention calculation, and the calculation efficiency is improved; font structure information of Chinese characters is introduced as a supplement, the relevance of entities on multiple granularities of fonts, characters and words is mined through multiple cross attention, and the context representation ability of a text is enhanced; an entity recognition method based on duplex collaboration is used, collaborative learning between span extraction and label classification is achieved through soft label constraint and a multi-task joint scoring function, label classification training is completed through full samples, error propagation caused by span extraction errors is reduced, and the recognition accuracy is improved.
Owner:YANSHAN UNIV

Unsupervised domain adaptive target detection method and system based on double-path structure

The invention provides an unsupervised domain adaptive target detection method and system based on a double-path structure, and the method comprises the steps: carrying out the target detection of an unmarked target domain data set through an unsupervised domain adaptive target detection model of the double-path structure, performing label filtering fusion on the invariant feature detection result set and the feature detection result set obtained by detection to obtain a target false label of the target domain data set; training a model by using the source domain data set, the target domain data set and the target pseudo-tag, and updating the target pseudo-tag once when a new target pseudo-tag obtained based on target detection is trained once; and inputting the to-be-detected target domain image into the trained model for detection to obtain a target detection result, performing label filtering fusion on the result obtained by dual-path detection to obtain a pseudo label with high confidence, and performing training fine tuning on the model by using the data set with the pseudo label. The recall ratio and precision ratio of model detection can be improved, and then the detection precision is improved.
Owner:MILITARY INTELLIGENCE RES INST OF THE CHINESE PEOPLES LIBERATION ARMY ACAD OF MILITARY SCI

Decentralization financial platform loan intelligent contract vulnerability detection method and system based on graph neural network

The invention relates to the technical field of software security, in particular to a decentralized financial platform loan intelligent contract vulnerability detection method and system based on a graph neural network, and the method comprises the steps: collecting intelligent contract source code data, and carrying out the vulnerability marking of the intelligent contract source code data according to whether there is a vulnerability in the data; compiling intelligent contract source code data into an abstract syntax tree, and constructing a contract code heterogeneous graph; constructing a graph neural network model, and performing vulnerability prediction training on the graph neural network model by using the contract code heterogeneous graph to obtain an intelligent contract vulnerability detection model, the graph neural network model captures structural features of a contract code heterogeneous graph layer by layer by using a hierarchical attention enhancement mechanism and predicts and identifies vulnerabilities in contract codes based on the structural features; and converting the to-be-detected smart contract source code into a target contract code heterogeneous graph to perform vulnerability prediction on the target contract code heterogeneous graph. According to the method, the semantics and data flow characteristics of the contract can be accurately captured, and efficient and accurate detection of the intelligent contract vulnerability is realized.
Owner:Chinese People's Liberation Army Cyberspace Force Information Engineering University