204results about How to "Improve understanding" patented technology

The invention relates to the technical field of control, in particular to a double-arm robot operation skill learning method and system based on big language model reasoning. The method comprises the following steps: firstly, carrying out context semantic modeling on an input natural language task instruction based on a large language model to generate a task semantic graph; in combination with a semantic entity in the task semantic graph and visual perception data collected by a robot, determining a three-dimensional space position of a target object through a multi-modal matching model, and constructing an environment semantic graph containing object nodes and spatial relation edges; generating an action sequence by using a language model according to the task semantic map and the environment semantic map, and generating a collaborative operation strategy based on the two-arm tail end state and an obstacle map; and finally, collecting feedback data of the sensor in real time when the action sequence is executed. According to the method provided by the invention, the understanding and execution capability of the two-arm robot on the unstructured natural language instruction is remarkably improved.

The invention provides a multi-parameter water quality data fusion analysis method and system. The method comprises the steps that water quality parameters are collected to form a three-dimensional data cube; constructing a space-time tensor model by using Tucker decomposition and a graph convolution network, and generating a core tensor matrix; constructing a dynamic constraint library and embedding the generative adversarial network; training a generative adversarial network by using Transform and physical constraint loss, generating synthetic data and verifying the synthetic data; performing space-time fusion by using meta learning weight distribution and Bayesian deep learning to generate a weight matrix and a confidence interval; missing data are restored through physical constraint interpolation and Gaussian process regression, and SHAP and LIME interpretation and path diagrams are generated; and performing real-time analysis by using an edge-cloud collaborative architecture to generate an intelligent report. Through physical constraint modeling, dynamic weight distribution and edge-cloud collaborative architecture, the problems that synthetic data violates physical laws, weight staticization, response lag and insufficient interpretability are solved.

The invention discloses an action quality evaluation method based on video and skeleton bimodal fusion, and the method comprises the steps: inputting a video frame sequence containing a complete action process into a fragmentation network, and generating continuous fragments; secondly, inputting each video clip into a video modal feature extraction network and a human body skeleton feature extraction network, respectively obtaining a video modal feature and a skeleton modal feature, fusing the video modal feature and the skeleton modal feature through a cross-modal attention mechanism, and generating a fused clip-level feature representation; and finally, inputting each segment-level feature representation into a multi-layer Transform encoder, obtaining a global time sequence feature of the action through a self-attention mechanism, inputting the global time sequence feature into a multi-layer perceptron (MLP) regression network, and generating an action quality score. According to the method, a prediction result is ensured to be more stable in numerical value and more accord with human perception in semantics, and the accuracy and consistency of action quality evaluation are remarkably improved.

The application discloses a kind of multi-modal medical image segmentation method, device, computer equipment and storage medium, comprising: obtaining the multi-modal medical image of the same examinee and pre-processing and spatial registration;The target segmentation concept is standardized and analyzed to generate prompt;Using basic segmentation model obtains multiple candidate segmentation results of each modality;Constitute credibility gating model, assess the credibility score of each candidate result and carry out screening, obtain the credible instance of each modality;Establish the corresponding relationship between cross-modal credible instance to form instance group;Each instance group is executed evidence weighted fusion and conflict resolution, and the final segmentation result of each target is obtained;Output multi-instance segmentation result and corresponding instance-level uncertainty index.The application can effectively reduce the false detection and result fluctuation of basic model in multi-modal image, improve the consistency and reliability of cross-modal segmentation, and provide robust segmentation result for clinical screening.

The application discloses a visual positioning method based on memory self-correction. The existing visual positioning method uses fixed image and text representation to capture cross-modal semantic consistency, which limits the flexibility of adjusting image representation according to different text information. In order to cope with this limitation, the application proposes a new memory self-correction network, which dynamically refines the image representation according to the query, thereby improving the semantic consistency between the text and the image to realize visual positioning. A semantic related filtering module (SRFM) and an adaptive memory fusion module (AMFM) are constructed to explicitly model the relationship between the image and the text. SRFM focuses on filtering image information irrelevant to the query, while AMFM adaptively fuses text-related representation with initial image features to enhance the understanding ability of the MSCN model. Comprehensive experiments on three datasets verify the superiority of the proposed method compared with existing methods.

The invention provides a page operation method and device, electronic equipment, a computer program product and a computer readable storage medium. The method comprises the steps of performing instruction prediction according to an operation request, page information corresponding to the operation request and at least one piece of target historical request information, and determining an instruction prediction result corresponding to the operation request; the target historical request information comprises a historical operation request, page information corresponding to the historical operation request and an instruction prediction result corresponding to the historical operation request; and operating a page corresponding to the operation request based on the instruction prediction result. Through the application, the success rate of page operation based on the instruction prediction result can be improved.

This invention provides a method and system for semantic novelty search of scientific and technological literature based on a large-scale pre-trained model, belonging to the field of information retrieval technology. The method includes: acquiring the target user's query input text and retrieval constraints; performing preprocessing and standardization to obtain preprocessed query input text; performing multi-granularity semantic encoding fusion to determine the semantic vector of the preprocessed query input text; performing domain-adaptive fine-tuning to obtain an optimized BERT model; constructing a feature classifier to generate feature semantic representations to obtain feature semantic representation results; performing comprehensive novelty assessment to obtain a comprehensive novelty rating; and performing automated difference analysis to obtain difference analysis results. This invention solves the technical problem in existing retrieval schemes based on pre-trained models that use single-granularity semantic representations, which cannot fully capture the complex semantic structure of professional terms, resulting in inaccurate novelty search results.

The invention discloses an automobile demand prediction method and system based on large model time sequence cross-modal alignment, and belongs to the technical field of automobile demand prediction. The automobile demand prediction method based on large model time sequence cross-modal alignment comprises the following steps: acquiring time sequence data and prompt data related to a hydrogen fuel cell automobile; inputting the time sequence data into the time sequence coding branch in a reverse embedding manner to generate a time sequence embedding representation; inputting the prompt data into an LLM enhanced coding branch for coding, and generating a prompt embedding representation; and calculating a channel-level similarity matrix between the time sequence embedding representation and the prompt embedding representation, aggregating to the time sequence embedding representation based on the channel-level similarity matrix, generating an aligned cross-modal feature, and generating a demand prediction result of the hydrogen fuel cell vehicle based on a decoder. According to the method, multi-source heterogeneous information is effectively fused, the precision and robustness of demand prediction of the hydrogen fuel cell vehicle are improved, and the capability of responding to emergencies caused by external factors is enhanced.

This invention relates to the field of decorative engineering technology, and discloses a method and system for generating virtual renderings of decorative engineering projects. The method includes: S1, parsing CAD drawings; S2, identifying professional intent; S3, defining and configuring key aesthetic elements; S4, attributing and correcting model parameters; S5, conflict detection; S6, adaptive fusion and correction; S7, material mapping and lighting calculation; and S8, outputting the virtual rendering. This invention solves the problem that the complexity of CAD drawings leads to incomplete or inaccurate geometric structures in 3D models, which in turn affects material mapping and lighting calculation, thus impacting the visual realism of virtual renderings.

The invention provides a task execution method and device and related equipment, and relates to the technical field of artificial intelligence and wireless communication, the task execution method comprises the steps that a target request is acquired, and the target request is used for requesting to execute a target task; splitting the target task to obtain a plurality of sub-tasks; obtaining target information matched with the subtask in a preset database, and packaging the subtask and the target information into a target information block; generating execution path information according to the target information block, wherein the execution path information is used for representing an operation step sequence for executing the subtask; and executing the operation step sequence based on the execution path information. Therefore, the execution accuracy of the target task can be improved, namely, the execution effect of the target task is enhanced.

The embodiment of the application provides a method for converting a knowledge graph into a natural language large model training sample and related equipment, and belongs to the technical field of artificial intelligence. The method comprises the following steps: selecting a key anchor point from the knowledge graph based on a graph centrality measurement index; taking a key anchor point pair as a starting point and an ending point, extracting a symmetric element path connecting the two, and constructing a subgraph; generating multi-perspective paths for the same core entity through an anchor point switching mechanism; converting subgraph information into a (instruction, output) sample pair conforming to the instruction fine-tuning paradigm by using a structured information template, a path information template and a metadata enrichment template; and fine-tuning a large language model by using the generated corpus. By introducing the symmetric element path and the multi-dimensional instruction template, the embodiment of the application realizes semantic preservation and structured mapping of complex relationships in the knowledge graph, and effectively improves the knowledge understanding, reasoning and generalization ability of the large model in a specific field such as the power secondary equipment field.

The invention relates to the technical field of index prediction in the social economic field, and discloses a social economic index set prediction method, which comprises the following steps of collecting multi-source data related to social economy, the multi-source data comprises medical business data of a medical institution, medical insurance data of a medical insurance department and social economic environment data of an external data source; and cleaning and standardizing the collected multi-source data to remove noise, missing values and abnormal values in the data, and unifying the data format and dimension. Data are obtained through multiple channels, the data basis of social and economic index analysis is enriched, multi-aspect conditions are presented, macroscopic information is reflected, understanding of phenomena and trends is enhanced, the accuracy and scientificity of index prediction are improved, and more valuable data support is provided for decision making. The problems of limited data sources and insufficient data mastering in the previous social economic index research are solved.

The application discloses a traditional Chinese medicine property analysis system and method, and a computer readable storage medium. The system comprises an information acquisition unit, a first analysis unit, a quantification unit and a second analysis unit. The method comprises grouping patients with consistent conditions and grouping traditional Chinese medicines according to different influencing factors; analyzing the influence of different influencing factors of traditional Chinese medicines on treatment effects; analyzing the quantitative relationship between the dosages of traditional Chinese medicines with different influencing factors and the dosages of traditional Chinese medicines required to achieve the same treatment effects; and analyzing the influence of influencing factors of traditional Chinese medicines on the properties of traditional Chinese medicines. The application analyzes the influence of different influencing factors of traditional Chinese medicines on treatment effects, further obtains the influence of influencing factors of traditional Chinese medicines on the properties of traditional Chinese medicines, and thus obtains the dosages of traditional Chinese medicines suitable for the current situation according to the quantitative relationship between the dosages of traditional Chinese medicines with different influencing factors and the dosages of traditional Chinese medicines required to achieve the same treatment effects, which is favorable to increasing the accuracy of the dosages of traditional Chinese medicines, strengthening the comprehensive understanding of the properties of traditional Chinese medicines, enhancing the modernization degree of traditional Chinese medicine, improving the persuasiveness and reliability of traditional Chinese medicine, and being favorable to the international development of traditional Chinese medicine.

The invention relates to the technical field of pile body detection, in particular to a pile body integrity intelligent detection method based on multi-source data fusion, which comprises the following steps: acquiring a spectrogram corresponding to a stress wave through a time sequence, carrying out time sequence analysis on stress wave data in a current scene by taking standard reference data as a benchmark, and combining a constraint condition of the current scene to obtain a pile body integrity intelligent detection result; collecting effective observation pairs in each detection point; taking stress wave data of the effective observation pair as input, and outputting defect types and defect areas of the pile body at different depths through inversion calculation; for the obtained defect areas, determining correlation coefficients among the defect areas at different time points, and obtaining depth distribution characteristics among the defect areas; and on the basis of the depth distribution characteristics, determining an anomaly judgment standard at each detection point, and according to the anomaly judgment standard, integrating the defect areas into an output pile body detection result. And the accuracy and the efficiency during pile body detection are realized.

The application discloses an underwater target detection encoder, a detection system and a method, the encoder comprising an input projection layer, a global semantic modeling unit, a first-stage feature fusion unit and a second-stage feature fusion unit; wherein the input projection layer is used for channel mapping of multi-scale features output by a backbone network, the global semantic modeling unit establishes a dependency relationship between spatial positions in a feature map through a self-attention mechanism; the first-stage feature fusion unit introduces a frequency domain transformation path in a cross-scale feature fusion process, performs frequency screening on features and generates high-frequency perception features; the second-stage feature fusion unit introduces intermediate features in the first stage in a propagation process and performs splicing and reconstruction processing, forming multi-scale feature representation. Based on the encoder, the detection system and the method are constructed, realizing classification and positioning processing of targets in underwater images.

The invention belongs to a power market simulation method, and provides a power market transaction simulation drilling method and a related device for solving the problems that a practical operation training tool is lacked in a current power market construction process, and the adaptability with actual operation is insufficient in a market system design and rule optimization process. After a power market subject model, market load demand data and an adaptive environment are determined, power medium-and-long-term centralized bidding transaction simulation is performed according to the power market subject model and the adaptive environment by adopting a mode of cooperative interaction of a master station end and a substation end, and power medium-and-long-term rolling matchmaking transaction simulation is performed based on a marketization transaction rule. The method comprises the following steps: carrying out the whole-process closed-loop simulation of an electric power spot day-ahead market, calculating the comprehensive income score of a competitor according to a simulation result, and obtaining an electric power market transaction simulation drill result in combination with a model result and the comprehensive income score. The method is superior to the prior art in the aspects of simulation efficiency, operation stability and practical training.

The application discloses a kind of marine unmanned equipment target detection methods based on multi-modal representation learning, belong to artificial intelligence, marine high-end equipment and so on field, comprising: S1: image data, acoustic data and numerical data are respectively preprocessed, and through space-time alignment, generate after synchronization multi-modal data;S2: based on the multi-modal data after synchronization, extract multi-modal feature vector, through projection head mapping to low-dimensional embedding space, based on contrast learning loss and sequence mask reconstruction loss, large-scale pre-training is carried out, and pre-training multi-modal model is generated;S3: based on pre-training multi-modal model, freeze the parameters of encoder, design prompt vector and multi-modal data are spliced, and through self-regularization constraint, parameter efficient fine-tuning is carried out, and classification model is generated;S4: based on pruning and quantization operation, the classification model is compressed, and deployed to the edge computing unit of marine unmanned equipment.The normal target detection capability of marine unmanned equipment is improved.

The application provides an abnormal value screening method based on particle component reconstruction, and relates to the technical field of environmental monitoring. The method comprises the following steps: obtaining standard ion component data, standard carbon component data and standard inorganic element component data; obtaining multiple reconstruction component data; determining target PM2.5 data; determining suspicious reconstruction component data judgment results of the multiple reconstruction component data at each moment; determining single-variable time sequence abnormal point judgment results of each kind of reconstruction component data at each moment; determining multivariate time sequence abnormal point judgment results of the multiple reconstruction component data at each moment; and determining whether each kind of reconstruction component data at each moment is abnormal data. According to the application, the component reconstruction characteristics under different pollution levels can be evaluated, the understanding of the relationship between particle components by the model is enhanced, the stability and accuracy of identifying whether the reconstruction component data is abnormal data are improved, the calculation efficiency is optimized, and the environmental monitoring auditors are assisted to make decisions.

The application provides a multimodal large model modal missing data completion method, relates to the technical field of data processing, and comprises the following steps: step 1, missing value detection and encoding processing are performed on multimodal input data, the correlation between internal features of the data is analyzed, structured feature representation is generated through feature reorganization, feature calibration parameters are generated in combination with the correlation between local features, and numerical feature vectors and mask matrices of each mode are obtained; step 2, the numerical feature vectors and the mask matrices are input into a feature fusion model, and the context dependence relationship between modes is analyzed, the major axis parameters of the feature distribution ellipse are calculated, the ellipse features are corrected by rotation, and preliminary corrected fusion feature representation is generated. Through feature distribution ellipse correction and global context retrieval, the accuracy and intelligence of multimodal missing data completion are improved.

The application provides a kind of multi-source heterogeneous database natural language query method based on knowledge graph enhancement, it is related to data query technical field.Method includes according to user question and target table set, in vector database is retrieved, obtains relevant list information;Table name is extracted from DDL list, and the corresponding knowledge graph subgraph in graph database is inquired;Based on list information and knowledge graph subgraph, structured Prompt is constructed;The structured Prompt is input into LLM model, and the original response text is obtained;SQL list is extracted from the original response text, and the SQL list is tested to obtain the final SQL;The final SQL is executed to the target database, and the query result is obtained;User question and query result are spliced, and input into LLM model, to obtain natural language abstract and original structured data.Deeply analyze the implied association between database tables, so that the LLM model generates more accurate SQL statements, effectively solves the problem that the accuracy of SQL generation in the prior art is low in the multi-table association scene.

This invention discloses an AI data processing system for speech recognition devices. The system includes a data acquisition unit, a preprocessing unit, a noise detection unit, a joint adjustment unit, a feature optimization unit, and a recognition output unit. Each unit establishes a closed-loop collaborative processing link through data interaction. The data acquisition unit is used to simultaneously acquire the original speech data of the object to be recognized and the environmental noise data of the corresponding scene. The preprocessing unit is used to perform unified and standardized preprocessing operations on the two types of data. This invention relates to the fields of speech recognition and artificial intelligence technology. This AI data processing system for speech recognition devices, through a dynamic noise model and intelligent adjustment algorithm, can adapt to different noise environments in real time, generating accurate dynamic adaptation parameters, thereby effectively suppressing noise and enhancing the clarity of the speech signal. This significantly improves the accuracy of speech recognition in complex noise environments.

The embodiment of the invention provides a training method and device for a trajectory prediction model of an automatic driving vehicle. The method comprises the following steps: acquiring a historical driving data set of a target vehicle; based on the historical driving data set, constructing a target unlabeled data set, a target labeled data set and a preference optimization data set; based on the target unlabeled data, pre-training the initial visual motion model to obtain a pre-trained model; training the pre-training model based on the target labeled data set to obtain a training model; and based on the preference optimization data set, performing direct preference optimization training on the training model to obtain a trajectory prediction model. The technical problem that the accuracy of automatic driving track prediction is low is solved.

The invention belongs to the technical field of automatic control, and particularly relates to a multi-modal motion scene compression bar cooperative control system which comprises a multi-modal sensing fusion module, a dynamic coupling relation online identification module, a self-adaptive cooperative control decision module and a distributed high-precision execution module. The multi-modal sensing fusion module is used for collecting and processing original heterogeneous data from the visual sensor, the force sensor and the position sensor; the module carries out timestamp alignment and space coordinate system unification on various sensor data. By designing a multi-modal perception fusion module and a cross-modal attention fusion unit, depth feature extraction and adaptive weighted fusion of visual sense, force sense and position information are realized, the problem that heterogeneous data is difficult to process uniformly in a traditional method is effectively solved, the comprehensive perception and understanding ability of a system to a complex dynamic environment is remarkably improved, and the system has a wide application prospect. And high-quality state information input is provided for subsequent accurate control.

The invention relates to the technical field of data processing, and particularly provides an intelligent prediction method, system and device for industrial power consumption and a storage medium, and the method comprises the steps: collecting power consumption load, production operation, environment and time calendar multi-source data, carrying out the fusion, cleaning and abnormal value processing, constructing a preprocessing time series data set, and generating state marking features in the abnormal processing; feature engineering is carried out on the data set, and lagging, sliding statistics, trend, periodicity and external causal features are extracted; standardizing the numeric features, combining the standardized numeric features with the category features, and constructing a model input feature set; and inputting the set into a pre-trained hybrid prediction model, fusing output results of the time sequence deep learning model and the integrated learning model therein, and finally outputting point prediction and interval prediction of the industrial electrical load. According to the method, through multi-source fusion and refined feature engineering, the cognition and prediction precision of an industrial power consumption complex mode is remarkably improved.

The application provides an interpretable ground subsidence sensitivity control factor identification method, and belongs to the technical field of geological disaster prevention. The technical scheme comprises the following steps: S1, obtaining regional surface deformation information based on multi-temporal synthetic aperture radar interferometry (MT-InSAR); S2, resampling the surface deformation monitoring to construct a regional ground subsidence sensitivity evaluation unit; S3, selecting height, slope direction, slope, distance from fracture, distance from river, fracture density, river network density, deep groundwater level, shallow groundwater level as the regional ground subsidence influencing factors to obtain the regional ground subsidence sensitivity partition result and identify the regional subsidence main control influencing factors. The application discloses and quantifies the influence weight of environmental factors on ground subsidence and the spatial heterogeneity of the action.

This invention discloses a data product identity ID generation method based on a data grid, belonging to the field of data processing technology. It includes constructing a system communication architecture based on a data grid, employing a distributed, multi-layered data product identity ID design within the communication architecture. This solves the technical problem of achieving efficient ID generation algorithms and query mechanisms using data grid-based data product identity ID generation. This invention ensures that the generated identity ID is globally unique within the data grid, improving user understanding and interpretability of its meaning, facilitating the expression of relationships and dependencies between data products, possessing strong versatility, improving interoperability between the system and applications, allowing for flexible expansion and customization, ensuring efficient data transmission and low-latency communication, while also possessing reliability. The design employs the BGP routing protocol, enabling the selection of the optimal path, avoiding loops, and achieving data communication across autonomous systems.

This application provides a method for inverting the dynamic changes of the seabed sedimentary environment based on three-phase potential, including in-situ data processing, electrical signal calibration, mathematical model establishment, and dynamic change analysis. By deploying a seabed electrode array, signals such as spontaneous potential, resistivity, and redox potential are acquired in real time. An inversion algorithm is used to deduce seabed sedimentary environment characteristics, such as seabed interface location, suspended particle concentration, sediment density, porosity, and water content. This method, combined with multi-frequency signal analysis, can monitor changes in the seabed sedimentary environment in real time and identify sediment movement. Compared with existing technologies such as CN118153411B and CN110411923B, this invention covers richer inversion content, and the complementarity of potential signals improves monitoring accuracy and efficiency, exhibiting stronger adaptability and reliability. This method has lower equipment and maintenance costs, is suitable for large-scale, long-term monitoring, and can provide innovative solutions in fields such as deep-sea monitoring, environmental assessment, and resource exploration.

This invention relates to the fields of artificial intelligence and medical imaging technology, specifically to a method and system for generating radiology reports based on temporal reasoning and evolutionary analysis. The method includes: acquiring the patient's current medical images, current examination reports, historical medical images, and historical examination reports; encoding the features of the current and historical medical images using a visual encoding network, then inputting these features into a temporal feature extraction network to obtain temporal visual features; comparing and aligning the temporal visual features with the corresponding text in the examination reports to obtain visual text features; and inputting the visual text features into a fine-tuned large language model to generate the report text. This invention, by performing temporal modeling on images from multiple patient examinations, can effectively capture the evolutionary patterns of diseases in images at multiple time points, thereby generating a diagnostic description with temporal consistency.

The application provides a kind of intelligent construction site safety monitoring and abnormal behavior detection method, it is related to behavior identification technical field, including: based on video frame data, through target detection algorithm, personnel and cigarette in the construction site multi-modal data are positioned, to output first detection result;Based on the first detection result, video frame data and construction site multi-modal data in the preset time period, utilize 3D CNN to combine attention mechanism, to output second detection result;Based on the second detection result and construction area security level information, by isolated forest and self-encoder algorithm, get comprehensive abnormal score.

This application provides a conversation processing method and apparatus. The method includes: acquiring current input information from a user during the current dialogue turn between a user and a large language model, and determining the current semantic vector corresponding to the current input information; acquiring memory units corresponding to at least one historical dialogue turn in the user's memory bank; determining the memory relevance between the current dialogue turn and each historical dialogue turn based on the current semantic vector and the memory units corresponding to each historical dialogue turn; sorting each historical dialogue turn by the memory relevance, and determining a preset number of active memory units related to the current dialogue turn in the sorting result; and injecting the preset number of active memory units into the large language model, so that the large language model combines the active memory units to perform the current dialogue turn.