29 results about "Domain identification" patented technology

A cross-domain recognition and drift compensation method for electronic noses based on contrastive variational autoencoders (CVAEs) is proposed, belonging to the interdisciplinary field of electronic nose systems and artificial intelligence. The method involves constructing a labeled dataset, building an initial C-VAE network and establishing a basic loss function, connecting the initial C-VAE network with a classifier to form a joint learning network, introducing a classification loss function to construct a combined C-VAE loss function, and training to obtain an initial model. A target domain dataset is then constructed and input into the initial model to obtain the latent space vector of the target domain. Mean-variance transformation is used to achieve style transfer and data augmentation of source domain features. The latent feature structure of the target domain is optimized through various constraints. A joint training set is constructed, and the depth alignment of features between the source and target domains is achieved by constraining the distribution differences of features between the two domains. Training is performed under the joint loss function, and progressive scheduling is applied to obtain a cross-domain gas recognition model. This method effectively solves the nonlinear drift problem of electronic nose sensors.

The application discloses a training method of a property customer service question and answer model, and comprises the following steps: inputting a multi-stage knowledge mask strategy into a BERT model for pre-training to obtain different language granularity information; and performing domain recognition training based on the language granularity information to obtain the property customer service question and answer model. According to the application, the multi-stage knowledge mask strategy and the domain recognition are trained in the BERT model, so that the user's question can be accurately understood, and the user experience is improved.

Embodiments of the present application provide a model training method and device, electronic equipment and storage medium, and relate to the technical field of computers. The method comprises: retrieving a set of documents from a preset document library based on a set of keywords; clustering the set of documents to obtain a plurality of subsets of documents; extracting at least one document from each subset of documents as a first document; determining positive samples and negative samples from the first document through a plurality of large language models and training an initial model to obtain a target model. After training the baseline model on the few-sample labeled dataset of the multi-model voting, the baseline model is enhanced using an active learning strategy. First, the baseline model is used to evaluate a randomly selected set of documents, and the sample with the lowest prediction is iteratively annotated. Using the large model debate idea, when the accuracy, recall rate and other indicators of the model no longer have significant improvement, the iteration is stopped, so that the trained model has high target field recognition ability.

The application relates to a spatial transcriptome spatial domain identification method and system based on double contrast learning. The method comprises the following steps: performing Gaussian smoothing processing on gene expression data to obtain smoothed gene expression data; constructing an initial adjacency matrix based on spatial coordinate data, performing edge pruning to obtain an optimized adjacency matrix; inputting the smoothed gene expression data and the optimized adjacency matrix into a graph convolution network encoder to obtain low-dimensional latent features and reconstruction features, and calculating a reconstruction loss; inputting the low-dimensional latent features into a multi-branch multi-layer perception fusion module to obtain enhanced features; constructing a regional contrast learning task, calculating a regional contrast loss, constructing a topological contrast learning task, and calculating a topological contrast loss; optimizing model parameters through a total loss, using an optimized model to perform spatial domain identification, and obtaining a spatial domain identification result. The spatial domain identification accuracy can be improved and the cost can be reduced.

This application relates to a zero-interaction permission recognition method and system based on facial recognition and ERP / MES integration, belonging to the field of information security recognition technology. The recognition method includes: real-time acquisition of facial images through a multi-camera network deployed in different scenarios; preprocessing the facial images to construct a facial feature relationship network, extracting facial feature vectors, and performing liveness detection; outputting identity features when the liveness detection confidence exceeds a preset confidence threshold; comparing the identity features with a pre-stored employee feature database in the ERP / MES system in real time, and binding employee identity identifiers and real-time permission data based on the comparison results; automatically triggering corresponding device automatic authorization commands based on permission binding tags matching the current scenario; and recording the execution events of the device automatic authorization commands to generate structured audit logs. This application can achieve high-precision identity verification without the need for active user cooperation, and instantly trigger corresponding levels of device control and data access permissions accordingly.

The application discloses a kind of multi-agent (Agent) dynamic cooperation and information aggregation method for complex query task, the method includes: by extracting the key semantic feature vector of query text, similarity matching is carried out with pre-set field feature library, accurate field recognition and shunt are realized;For real estate field query, further identify subtask type, dynamically load the most suitable professional agent from special sub-agent pool;Agent detects parameter gap automatically during execution, and obtains complete parameters through multiple rounds of directional interrogation;Finally, call professional tools to obtain original information, and carry out semantic deduplication, conflict detection and multi-source fusion, to generate comprehensive answer.The application realizes the specialization of complex query processing, intelligent interaction, efficient dynamic scheduling and powerful information integration, and improves the processing efficiency and answer quality of complex query, especially professional field query.

The application discloses a kind of based on data link's communication data's cleaning method, it is related to data processing technical field, the data quality characteristic of each data group is identified and obtained, data quality coefficient is generated, if it exceeds quality threshold, send cleaning instruction;From the cleaning priority value of data processing in each sub-region generated by data condition set, the cleaning priority value in each sub-region is sorted, and the sorting obtained is used as cleaning sequence;The cleaning feature of data in each sub-region is identified and obtained, and from the pre-constructed test cleaning scheme library, several corresponding candidate schemes are matched out, and the recommended scheme is selected from several candidate schemes;Several data quality coefficients are continuously obtained, the time consumption of data quality coefficient exceeding corresponding quality threshold is predicted, if it is shorter than expected, then send prompt instruction.The busy degree of each sub-region is formed evaluation, and the working state of each sub-region is adjusted according to the evaluation result, to prolong its working life.

The application discloses a coal soluble organic matter molecular category content prediction method based on a solvent domain, relates to the field of coal chemical analysis, and comprises the following steps: S1, acquiring sample data and recording solvent domain identification; S2, performing component identification on each sample to obtain relative content; S3, merging component relative content into a molecular category content candidate set; S4, constructing a feature matrix with component relative content as input and performing data preprocessing; S5, performing unsupervised evaluation on the molecular category content candidate set; and S6, training a prediction model and outputting a content prediction result. The sample is grouped through sequential solvent sub-domain processing, component features are unified, and molecular categories are merged, unsupervised target category screening is performed in combination with principal component analysis and hierarchical cluster analysis, accurate prediction of the coal soluble organic matter molecular category content is realized, and data support is provided for clean and efficient utilization of coal and optimization of a conversion process.

This invention discloses a two-dimensional visualization method and related apparatus for identifying small-signal stability regions, comprising: constructing a full-order small-signal model of the system; calculating the stable operating points of the system node voltages and the stable operating points of the system branch currents; importing the stable operating points of the system node voltages and the stable operating points of the system branch currents into the full-order small-signal model of the system; and solving for the characteristic matrix of the system. A sys The initial conditions; based on the feature matrix of the system. A sys The initial conditions determine the eigenvalues ​​with the largest real parts. λ max Based on the eigenvalue with the largest real part λ max The method and related devices for identifying the stability domain boundary can reflect the multi-parameter coupling effect and efficiently identify the stability domain boundary of the system.

This invention discloses an open-domain adaptation method based on feature decoupling and data reconstruction, belonging to the field of artificial intelligence technology. The method includes: S1, dividing the model into known and unknown categories and performing differentiated data augmentation on the source and target domains; S2, constructing an end-to-end model including a feature decoupling module, a reconstruction module, and a dynamic gradient inversion layer; S3, training the model using a phased dynamic strategy, adaptively fusing four dimensions of information—classification confidence, class center distance, reconstruction error, and feature variance—during the inference phase; and optimizing training using gradient accumulation and pruning strategies; S4, saving the best model for inference based on the validation set AUROC. This invention effectively improves the cross-domain recognition accuracy, generalization ability, and training efficiency of the model under interference from unknown categories.

The application discloses a multi-scale space transcriptome domain identification method based on a block wavelet graph, and comprises the following steps: acquiring original gene data of a space transcriptome; pre-processing the original gene data to obtain processed gene data; performing multi-scale feature extraction on the processed gene data to obtain feature data information; training a block-level model by using the feature data information to obtain a trained block-level model; outputting a block reasoning result by using the trained block-level model, and performing clustering analysis after global recombination of the block reasoning result to obtain a space domain division result. The application can accurately capture cross-scale biological signals, break through hardware limitations, and realize efficient and accurate space domain identification of super-large-scale µST data.

The application particularly relates to a spatial multi-omics data integration, alignment and omics inference method based on pair graph contrastive learning, comprising the following steps: obtaining omics expression count matrices and spot spatial coordinates of multiple tissue sections; constructing a section space graph based on the spot spatial coordinates, and pre-processing the omics expression count matrices to obtain initial features of each omics modality; for data of the same omics modality in different tissue sections, jointly learning by using a single-omics integration module to obtain single-omics embedding representation; for single-omics embedding representations of different omics modalities in the same tissue section, adaptively weighting and fusing by using a weighted nearest neighbor algorithm to obtain multi-omics integrated representation; and for multi-omics integrated representations of different tissue sections, performing cross-space alignment of spots in each section by using a labeled domain guided truncated iterative closest point method, so that cross-batch, cross-omics and cross-section collaborative integration can be realized, and the accuracy of spatial domain recognition and cross-omics inference is improved.

The present disclosure discloses a wafer surface scratch defect detection method, device, medium and system, which belongs to the technical field of semiconductor manufacturing and comprises the following steps: based on a plurality of original images of a wafer to be detected, an image of a wafer region of interest of the wafer to be detected is obtained, and a plurality of connected domains are segmented; the plurality of connected domains are divided into first connected domains or second connected domains; for the first connected domains, the types of defects in the first connected domains are identified according to the characteristic values of the connected domains; for the second connected domains, corresponding third connected domains are determined, and the second connected domains and the corresponding third connected domains are merged to generate corresponding defect regions to be detected; and the types of defects are identified according to the characteristic values of the defect regions to be detected. The first connected domains and the second connected domains with discontinuous possibilities are identified in the connected domains in the wafer surface image to be detected, and the second connected domains after merging form the final defect regions to be detected; and the types of defects are identified based on the finally determined defect regions to be detected, thereby improving the scratch defect identification accuracy.

The application discloses a non-solid medium quality detection method and system, which is applied to the technical field of non-solid medium printing and comprises the following steps: acquiring an original measurement speed sequence and a flow state image; determining the current flow speed of each non-solid medium flow column from the original measurement speed sequence; performing connected domain identification and adhesion domain segmentation on the flow state image to obtain a non-solid medium flow column region; for each non-solid medium flow column region, determining a current flow interruption detection result based on the change degree of pixel values along the flow direction of the flow column, calculating a current skew angle based on the circumscribed contour boundary of the non-solid medium flow column, and calculating a current flow column width based on the edge pixel point coordinate values along the width direction of the flow column; and integrating the current flow interruption detection result, the current flow speed, the current skew angle and the current flow column width into a jet flow quality detection result. Through the above method, full-process high-precision non-solid medium quality detection is realized, and flow column optimization and product control are realized.

This application discloses a drawing recognition method, apparatus, device, storage medium, and product. The method includes: acquiring a drawing to be recognized; inputting the drawing to be recognized into a pre-trained functional area recognition model, a symbol recognition model, and a line recognition model to recognize the drawing elements and their position information, whereby the drawing elements include functional areas, symbols, and lines; constructing association relationships between drawing elements based on preset relationships and position information between functional areas, symbols, and lines to obtain multiple structured data sets to be verified; performing mutual verification between corresponding two drawing elements based on each structured data set to be verified; and determining the drawing recognition result based on the drawing elements if the verification passes. This method can improve the efficiency and accuracy of drawing recognition.

The embodiment of the application provides a kind of regional identification method and device, apply in computer vision technical field.The method comprises: obtaining the first size information and the first coordinate information of local image area, obtaining the second size information of complete image area, according to the first size information of local image area and the second size information of complete image area, the first size information and the first coordinate information of local image area are corrected, third size information and third coordinate information are obtained, according to second size information and third size information, length proportion parameter and width proportion parameter are determined, according to length proportion parameter, width proportion parameter and third coordinate information, the second coordinate information of complete image area is determined, according to the second coordinate information and the second size information of complete image area, complete image area is divided into N sub-regions, and the text information in N sub-regions is identified.

This invention discloses a method, apparatus, device, and medium for electromyography (EMG) recognition in scenarios with missing source domain data. The method comprises: acquiring source domain EMG signal fragments, inputting them into a TCN feature extractor to extract features, and obtaining fixed-dimensional feature vectors; constructing a source domain gesture recognition model, inputting the feature vectors into a gesture classification head and a domain discriminator head, calculating cross-entropy loss and domain classification loss respectively, and freezing all parameters of the TCN feature extractor and the source domain gesture recognition model after adversarial training until convergence; guiding the user to complete a movement cycle in the target domain through a human-computer interaction interface, acquiring EMG signals and movement labels as calibration data pseudo-labels; constructing a nonlinear transformation module and inserting it into the backend of the frozen TCN feature extractor, generating calibration features aligned with the source domain by performing nonlinear mapping processing on the original feature vectors of the target domain; fine-tuning the nonlinear transformation module using calibration data, and constructing a target domain recognition model to achieve accurate recognition in scenarios with missing source domain data.

This disclosure relates to the field of spatial transcriptomics technology, and proposes a method, apparatus, and device for spatial region identification based on spatial transcriptomics data. The method includes: constructing a joint expression matrix based on candidate spatial transcriptomics data, where each element of the joint expression matrix represents the gene expression level of a single detection gene within a single spatial grid; performing structured encapsulation, data standardization, and dimensionality reduction on the candidate spatial transcriptomics data based on the joint expression matrix to generate a target dataset; calculating a dual adjacency matrix based on the grid distances and gene expression similarities between spatial grids within the target dataset; inputting the dual adjacency matrix and the joint expression matrix into a spatially aware graph convolutional network to extract neighborhood aggregation embedding features, and performing cluster analysis on these features to determine the spatial region identification result. The technical solutions provided by one or more embodiments of this disclosure can effectively improve the accuracy and stability of spatial region identification.

The application discloses a wind turbine blade multi-type damage detection method and device based on image processing, comprising two processes of blade damage area identification and damage type judgment. The application firstly carries out image grayscale, filter enhancement, segmentation and morphological processing algorithms on the blade image collected by the intelligent inspection unmanned aerial vehicle platform, realizes the identification of the blade damage area; then obtains the geometric features and gray features and other information of the blade damage area based on the connected domain analysis principle; finally, designs a blade damage type identification classifier according to the extracted feature parameter information to accurately determine the blade damage types such as skin peeling, coating damage, sand eye, oil stain and crack. The application has the advantages of simple operation, low detection cost, high detection efficiency and the like, and makes up for the deficiencies of the traditional blade operation and maintenance detection method, such as long time consumption, large labor and material cost investment and the like.

A test code generation method, comprising: receiving a test code generation request; on the basis of the test code generation request, determining a domain tag of a source domain and a domain mapping relationship, the domain tag of the source domain being used for indicating a domain to which source code of a test object belongs, and the domain mapping relationship comprising a mapping relationship between the source domain and a target domain; performing assembling on the basis of the domain tag of the source domain and the test code generation request to obtain a code generation prompt; inputting the code generation prompt into a code generation model to obtain test code of the test object in the source domain; and then, on the basis of the domain mapping relationship, mapping the test code of the test object in the source domain to a feature space of the target domain, so as to obtain test code of the test object in the target domain. The method determines the source domain and the target domain by means of adaptive domain identification, generates the test code of the source domain by understanding and analyzing test requirements of different domains, and performs code mapping on the basis of the domain mapping relationship, thereby implementing efficient generation of cross-domain test code.

This application relates to a multilingual speech content recognition method and system, belonging to the field of speech signal processing technology. The recognition method includes: acquiring raw audio stream data and performing noise reduction filtering to segment it into multiple audio segments; extracting acoustic feature vectors from the audio segments; inputting the audio segments into a speech recognition model group to obtain text segments and confidence scores; fusing all text segments to generate preliminary recognized text and segmenting it into text blocks, combining real-time received semantic feedback tags, and translating it block by block into target language text through a streaming translation model; performing bidirectional translation verification on text blocks with confidence scores below a set threshold; inputting the translated text stream and acoustic feature vectors into a semantic analysis model, fusing the output results to obtain semantic feedback tags; constructing a cross-modal association graph through a graph neural network, outputting key information summaries, and triggering alarm signals. This application can improve speech recognition accuracy while ensuring recognition efficiency and perform risk response.

The present application relates to the technical field of sentiment analysis, and particularly relates to a cross-domain aspect sentiment triple prototype constraint and relation correction method, comprising: performing context encoding, local semantic aggregation and fusion with domain identification on input text to obtain basic representation features of the text; constructing aspect phrase prototypes and opinion phrase prototypes based on target domain corpus, and performing phrase-level reorganization on source domain samples and screening cross-domain enhanced samples; generating candidate aspect spans and candidate opinion spans based on word boundary prediction results; performing relation correction and sentiment classification on candidate span pairs; and performing joint confidence constraint decoding by combining pairing scores and sentiment scores to output final aspect sentiment triples. The present application has the beneficial effect of introducing transfer priors under the condition of scarce target domain annotation data, reducing the interference of cross-domain expression differences on term matching, and improving the accuracy and stability of aspect term and opinion term pairing.

This invention proposes an image cross-domain recognition method based on relaxed-domain adaptive width learning, belonging to the technical fields of image recognition and transfer learning. The method includes: preprocessing and feature extraction of source and target domain image data; extracting width features of the source and target domains respectively through a width learning network and concatenating them to construct a joint width feature matrix; constructing an inter-domain difference metric matrix and transforming the joint width feature matrix to obtain a domain alignment term; constructing an orientation matrix and a non-negative drag matrix to establish a label enhancement matrix; constructing a relaxed-domain adaptive width learning model containing a domain alignment term, a source domain recognition reconstruction term, and a structural consistency term; and employing an alternating optimization strategy to achieve joint optimization, obtaining a domain alignment transformation matrix and a recognition projection matrix, and predicting the category of image samples in the target domain. This invention effectively alleviates the distribution offset problem in image cross-domain recognition and significantly improves recognition accuracy and the model's generalization ability.

The application discloses a kind of open source software reuse identification method, system, device and storage medium, it is related to computer field.Recognize the license information of each target root directory under the open source code file in the software project to be detected.According to the recursive order from the bottommost directory to the target root directory, each level directory is successively taken as the current directory.According to the suspicious degree score of each subdirectory under the current directory and the license information of each open source code file directly located at the next level of the current directory, the suspicious degree score of the current directory is determined.The target root directory with suspicious degree score reaching preset score value is taken as the suspicious directory with open source software reuse.Not affected by the code structure change of open source software, according to the license information, the possibility of open source software reuse in the target root directory is analyzed in the path level as a whole, which can effectively identify open source software and has good cross-platform and cross-language.

The application discloses a cross-domain pedestrian re-identification method based on feature enhancement, which comprises the following steps: firstly, a model is established, a data set is obtained, and the data set is input into the model; in the first training stage, an instance normalization Gaussian process is established, instance normalization statistics are collected, and the instance normalization statistics are added to the Gaussian process, new instance normalization statistics are sampled as instance normalization re-translation and rescaling parameters, features of pictures are extracted for classification, cross-entropy loss is calculated, back propagation is performed, and parameters of the model are updated; then, in the second training stage, a batch normalization Gaussian process is established, batch normalization parameters are collected, the batch normalization parameters are added to the Gaussian process, new parameters are sampled for batch normalization operation, features of pictures are extracted for classification again, cross-entropy loss is calculated, back propagation is performed, and parameters of the model and hyperparameters of the Gaussian process are updated; finally, a final model is obtained through iteration, and pedestrian re-identification is performed. The application improves the cross-domain recognition generalization pedestrian re-identification ability and recognition performance.

This invention provides a spatial transcriptome multi-slice alignment and domain identification system based on geometric distance matching, belonging to the fields of artificial intelligence and bioinformatics. Technical solution: Based on a geometric distance matching algorithm and constructed using an MVC architecture, it includes two core functional modules: multi-slice data alignment and spatial domain identification, as well as three supporting modules: front-end interaction, back-end service, and data processing. By identifying common cell types across multiple slices, Euclidean distance is calculated after standardizing cell spatial coordinates to achieve nearest neighbor matching. A similarity threshold is then used to filter high-matching cell pairs to complete multi-slice alignment. Beneficial effects: The system is compatible with multi-platform data and mainstream formats, provides 2D / 3D visualization interaction functions and result export services, and solves the problems of complex models, high resource consumption, and high usage thresholds in existing technologies. It achieves efficient and accurate multi-slice data alignment and spatial domain identification, and is suitable for multiple research scenarios such as developmental biology and oncology.

A test code generation method, comprising: receiving a test code generation request; on the basis of the test code generation request, determining a domain tag of a source domain and a domain mapping relationship, the domain tag of the source domain being used for indicating a domain to which source code of a test object belongs, and the domain mapping relationship comprising a mapping relationship between the source domain and a target domain; performing assembling on the basis of the domain tag of the source domain and the test code generation request to obtain a code generation prompt; inputting the code generation prompt into a code generation model to obtain test code of the test object in the source domain; and then, on the basis of the domain mapping relationship, mapping the test code of the test object in the source domain to a feature space of the target domain, so as to obtain test code of the test object in the target domain. The method determines the source domain and the target domain by means of adaptive domain identification, generates the test code of the source domain by understanding and analyzing test requirements of different domains, and performs code mapping on the basis of the domain mapping relationship, thereby implementing efficient generation of cross-domain test code.

Disclosed in the present invention is a Widget state machine for ARINC661 protocol, containing a domain recognition algorithm and a context management algorithm. The domain recognition algorithm is used for executing a traversal process of a widget tree and recognizing whether a domain change has occurred. On the basis of a recognition result of the domain recognition algorithm, the context management algorithm manages a context, comprising: (1) entering a sub-domain: storing an original current domain context in the top of a stack, calculating a new current domain context from an original sub-domain context and then, on the basis of the attribute of the sub-domain in a new current domain being accessed, calculating and updating the context of the sub-domain; (2) moving within the domain: keeping the current domain context unchanged and, on the basis of the attribute of the sub-domain in the current domain being accessed, calculating and updating the context of the sub-domain; and (3) returning to an upper-level domain: popping the context at the top of the stack to restore the stack to the context of the upper-level domain. By means of introducing the concept of "domain", the present invention achieves unified management of Widget contexts.