2025 results about "Structure based" patented technology

The invention provides a standard document automatic generation and multi-dimensional auditing method and system based on a large model, and relates to the technical field of artificial intelligence, and the method comprises the steps: 1, building a distributed database of a multi-source document, and analyzing a heterogeneous text through natural language processing to obtain a standardized knowledge network; step 2, extracting index elements based on the standardized knowledge network, and forming a structured parameter library through verification and verification; and step 3, based on the structured parameter library, constructing a template library, analyzing user demands in combination with semantic matching, and automatically generating a standard document outline. The document generation efficiency and quality are improved, the manual auditing cost is reduced, and the auditing comprehensiveness and accuracy are enhanced.

The invention discloses a bridge detection method and system based on a digital twin technology, and relates to the field of bridge structure health monitoring. The method comprises the following steps: acquiring a strain distribution value, a vibration spectrum value and an environmental load spectrum value in real time through a sensor network deployed in a physical bridge, generating a structural response data set, and synchronizing the structural response data set to a digital twinborn body; calculating a damage index value and an accumulated damage quantity value based on the structural response data set; inputting the damage index value and the accumulated damage quantity value into a preset safety criterion, and calculating a safety margin coefficient value and a failure risk grade value; calculating a residual life prediction value based on the safety margin coefficient value and the environmental load spectrum value, and synchronously correcting a degradation rate value of the digital twin; and generating a priority maintenance instruction according to the failure risk grade value, the residual life prediction value and the safety margin coefficient value, and feeding back maintenance effect data to the digital twinborn body to complete updating after execution. The bridge operation and maintenance efficiency and safety are remarkably improved.

The invention discloses a three-dimensional Gaussian sputtering scene reconstruction method based on structure perception refined Gaussian, and aims to solve the problems of Gaussian drift, edge blur, structure artifacts and the like of a reconstruction model due to the fact that sparse point cloud contains outliers, Gaussian morphology and normal are mismatched and a multi-dimensional optimization target is lacked in an existing three-dimensional Gaussian sputtering reconstruction method. A key frame is extracted by collecting target scene video data, sparse three-dimensional point clouds are reconstructed by using an SfM algorithm, a depth map and a normal map are generated through a Lotus model, three-dimensional Gaussian distribution is initialized after the sparse point clouds are filtered, a Gaussian covariance matrix is adjusted by using a normal consistency regular term, and the sparse point clouds are extracted. And after structure attribute analysis is carried out, a comprehensive scoring function is constructed to screen Gaussian points, and finally, a combined training framework including luminosity, normal consistency and structure continuity loss is adopted to optimize and generate a three-dimensional Gaussian scene model. The method is mainly applied to the field of three-dimensional reconstruction and multi-view rendering, and scene reconstruction precision and geometric consistency can be improved.

The invention relates to the technical field of intelligent operation and maintenance monitoring and self-adaptive rule engines, in particular to a monitoring strategy system and method based on a rule base. The method comprises the following steps: constructing a unified state vector model through multi-source heterogeneous data collection, enhancing state representation in combination with context semantics, and verifying the validity of the state representation; a dynamic rule cutting mechanism is adopted, a graph structure is constructed based on semantic redundancy and conflict relations between rules, and an optimal non-redundant rule subset is screened in combination with a greedy cutting algorithm; generating a strategy graph through rule semantic fusion, aggregating semantics by using a graph neural network, constructing a directed acyclic graph to solve action conflicts, and generating a safe and efficient response sequence; a cross-scene migration mechanism is introduced, and source scene strategy semantics are projected to a target scene through a mapping matrix, so that lightweight migration and self-evolution of a knowledge base are realized. According to the method, the dynamic adaptability of the rule base is improved, the redundancy execution risk is reduced, and multi-scene seamless migration is supported.

The invention relates to a supporting structure stress state monitoring method based on artificial intelligence, and belongs to the technical field of artificial intelligence and data processing. The method comprises the following steps: acquiring and marking strain data of a supporting structure; after abnormal values are removed, normalizing the multi-sensor data to generate a normalized strain sequence; a state monitoring model is constructed, a deep time sequence neural network architecture is adopted, and the state monitoring model comprises an input layer, a self-adaptive wavelet attention feature mapping layer, a time domain gating convolution module, a global maximum pooling layer, a dynamic feature importance reweighting layer and a full-connection classification layer; inputting a normalized data training model; optimizing a loss function through a quantile interval adaptive learning rate and a momentum updating strategy; after real-time monitoring data is processed, inputting the data into the training model according to time window slices, outputting four types of probabilities, and taking the maximum value as a prediction state; and if a plurality of continuous windows are early-warning and dangerous, triggering the terminal to give an alarm. The accuracy of monitoring the stress state of the supporting structure can be improved.

The invention relates to the technical field of automatic generation, in particular to a knowledge graph driven teaching material automatic generation method and system. The method comprises the following steps: receiving teaching target input, analyzing the teaching target input into a multi-layer structure based on a preset teaching intention meta-model, and mapping the multi-layer structure to a knowledge graph node; generating an optimal teaching path through a rule engine in combination with the node topology position and the teaching intention level; after the topological fingerprint of the target textbook system is recognized, compatibility analysis is executed, and if a structural gap exists, a transition module is automatically inserted to reconstruct a path; calling the bound multi-modal content unit according to the reconstruction path node, and completing modal coordination according to a cross-modal dependency graph; and finally, performing structured rendering on the integrated content, and outputting a textbook body adaptive to the terminal. The method has the advantages of teaching intention deep understanding, multi-modal content consistency control and cross-textbook system adaptation, and the intelligence level and generalization ability of textbook generation are improved.

The invention discloses a pilot optimization method for large-scale MIMO channel estimation based on structural compressed sensing. The method comprises the steps of establishing a channel estimation model for a large-scale MIMO-OFDM (Multiple-Input-Multiple-Output-Orthogonal Frequency Division Multiplexing) system when pilots are placed in an overlapping mode; simplifying the channel estimation model for the large-scale MIMO-OFDM system, thereby enabling the channel estimation model to correspond to a structural compressed sensing model; and obtaining an optimum pilot matrix through utilization of a pilot optimization algorithm. Through adoption of the optimum pilot matrix, according to the channel estimation of the large-scale MIMO system based on structural compressed sensing, the mean square errors MSEs of the channel estimation are clearly reduced, and the channel estimation performance is improved.

The invention relates to the technical field of building design and digitization, and discloses an automatic design method for a special-shaped facing structure based on parametric modeling. The method comprises the following steps: firstly, acquiring a geometrical characteristic parameter set of the special-shaped facing structure, wherein the geometrical characteristic parameter set comprises curved surface curvature distribution, boundary constraint conditions and a material attribute threshold value; thirdly, according to the curved surface curvature distribution, generating an initial geometric topological grid by using a dynamic subdivision algorithm, adjusting node positions according to boundary constraint conditions to obtain an optimized structural framework, and performing layered mapping on material attribute thresholds to generate a material distribution map; then, the input parametric modeling engine generates a three-dimensional parametric model, finally, a self-adaptive adjustment strategy graph is constructed through a multi-objective optimization algorithm, and a final design scheme is output. By means of the method, automation of design of the special-shaped facing structure is achieved, design efficiency and accuracy are improved, material utilization is optimized, scheme adaptability is enhanced, and design process collaboration is improved.

A system and method for latent hyperspace navigation in spatiotemporal media using hierarchical and Lorentzian autoencoders. The system compresses spatiotemporal media into navigable latent representations while preserving geometric and semantic relationships through tensor structure maintenance. A latent hyperspace manager organizes compressed representations as geodesic trajectories within a geometric manifold structure based on differential geometry principles. A geodesic trajectory mapper computes optimal navigation paths through the high-dimensional space, while symbolic anchors positioned at semantically significant locations serve as persistent reference points. Spatiotemporal routing protocols manage navigation decisions across multiple temporal scales. A strategy caching system preserves successful navigation patterns for reuse, enabling continuous learning. The system generates synthetic content during navigation to support infinite zoom capability, allowing exploration beyond original media boundaries. Cross-modal fusion combines diverse input modalities into unified representations, applicable to immersive media exploration, scientific visualization, and surveillance analysis.

The invention relates to the technical field of structure optimization design, and particularly discloses a box-type beam structure lightweight design method based on topological optimization, which comprises the following steps: constructing a multi-physics field coupled three-dimensional parameterized model, and performing cross-scale analysis on integrated structure rigidity, thermal deformation, vibration mode and fatigue life to obtain a three-dimensional parameterized model; a variable density method and level set method mixed topological optimization strategy is adopted, and an initial material distribution scheme meeting the structural function integrity is generated by modeling the relationship between density and stress strain; starting a quantum genetic algorithm and deep reinforcement learning fused intelligent optimization engine; according to the method, through the multi-physics field coupled three-dimensional parameterized model, a variable density method and level set method mixed topological optimization strategy is combined, the relation between the structural lightweight target and the additive manufacturing process is effectively balanced, multi-physics field analysis of thermal response, vibration, fatigue and the like is introduced, and the structural lightweight target is obtained. And it is guaranteed that the box-type beam meets the mechanical property and other functional requirements in actual application.

The invention provides a test risk digital twinning early warning method based on multi-domain cooperative monitoring, and belongs to the technical field of virtual-real fusion test and digital twinning, and the method comprises the steps: firstly, building a fine finite element simulation model which comprises a digital tool system, a digital sensor and a test piece and considers nonlinearity; secondly, performing nonlinear finite element simulation analysis, and constructing a multi-level mechanical response field inversion reduced-order model; thirdly, completing the construction of a complete sensor data set through a data filling algorithm, and carrying out the failure judgment of the first hierarchical structure based on the complete sensor data set; and finally, carrying out future loading level sensor data prediction and completing failure judgment of a second hierarchical structure. Carrying out full-field mechanical response inversion and online real-time correction; and performing response inversion of the region of interest to realize failure judgment of the third hierarchical structure. According to the invention, real-time dynamic monitoring and early warning of the structure test risk can be realized, the real-time performance, the robustness and the accuracy are high, and a powerful guarantee is provided for the safety and the reliability of the structure test.

The invention relates to the technical field of image processing, in particular to a dual-branch coding desert segmentation model network structure based on structure state space duality, which adopts multi-dimensional dynamic convolution to replace traditional convolution in the initial stage of an encoder, introduces a mamba2 module based on the structure state space duality into the backbone design of the encoder, and improves the robustness of the encoder. The efficiency and adaptability of the model are remarkably improved, a double-branch parallel design is adopted, one branch uses cavity convolution to extract multi-scale context information, the other branch reinforces feature expression through a mamba2 module, the model is connected in series with a space attention module and a channel attention module between an encoder and a decoder, and the algorithm is more accurate. The method has the advantages that the method is simple and easy to implement, interference of irrelevant information on segmentation results is suppressed, a deformable large kernel attention module is introduced to the tail end of a decoder, and global and local modeling capability of the model in processing desert complex boundary regions is effectively improved by combining flexibility of deformable convolution and global receptive field characteristics of large kernel convolution.

The invention relates to the technical field of knowledge graph construction, in particular to a node attenuation and edge similarity weight fused knowledge graph construction method and system, and the method comprises the steps: collecting multi-source heterogeneous data, and carrying out the preprocessing of the multi-source heterogeneous data; identifying entities in the preprocessed multi-source heterogeneous data, and extracting a relationship between the entities; establishing an initial graph structure based on the entities and the relationship between the entities, and determining the representation mode of nodes and edges in the graph; fusing time decay and space correlation factors to correct representation of nodes and edges in the constructed graph structure, and fusing semantic similarity on the basis of node and edge weight correction to further adjust an edge connection relation; after the graph structure is corrected and optimized, final knowledge graph data representation is organized and generated, and unified storage and graph calculation structured packaging are completed. According to the method, the defect that an existing map construction scheme only depends on time attenuation and neglects space factors can be effectively overcome.

The invention provides a knowledge base intelligent deep auditing method and system based on an AI large model, and the method comprises the steps: obtaining an unstructured file uploaded by a user, analyzing the unstructured file, extracting a semantic auditing unit, and constructing a semantic auditing unit set; constructing a graph structure based on context dependence according to the semantic auditing unit set, marking continuous, neighbor, reference and reverse logic edge types, and fusing a compliance attention mechanism to update node representation to obtain a node semantic representation set; identifying candidate problem items through a double-layer risk discrimination function by combining the graph structure and the node semantic representation set design according to the introduced law and regulation knowledge embedding library; and executing causal chain backtracking on the candidate problem items to generate a structured auditing conclusion. The auditing accuracy and efficiency are remarkably improved, and the manual rechecking burden is reduced.

The invention relates to a BIM-based building structure intelligent management method. The method comprises the following steps: step 1, dynamic monitoring factor mining and multi-modal sensing: collecting meteorological hydrology, geological conditions, microenvironment and natural disaster historical data; 2, neural operator-driven virtual-real fusion dynamic analysis: converting the format of the collected data, mapping the data into a BIM model, integrating a virtual-real synchronization engine, and pre-training a Fourier neural operator through real-time displacement and strain data reverse calibration model boundary conditions in a construction stage, and establishing a direct mapping relationship between an external load and a structural response; 3, multi-modal data collaboration and hierarchical permission interaction: forming a multi-dimensional feature matrix through association of timestamps and space coordinates; 4, multi-modal graph convolutional network damage identification: learning dynamic relevance between nodes through an adaptive graph convolutional layer, and extracting damage sensitive features; and step 5, dynamic life early warning of digital twinning drive: predicting the remaining service period of the structure.

The invention discloses a building structure anti-seismic toughness optimization method and system based on deep learning, and relates to the technical field of anti-seismic toughness optimizing.In operation of the system, a sensing network is deployed, multi-dimensional dynamic response parameters in the vibration process of a building structure are collected, preprocessing and sliding window characterization are carried out on a collected parameter set, and the anti-seismic toughness of the building structure is optimized; encoding the input tensor TXF, extracting a deep space-time coupling feature generation coefficient, performing multi-dimensional composite calculation on an anti-seismic stress energy coefficient SEC, a response co-motion coefficient RSC and a damage evolution index DEI to generate a system-level evaluation core index toughness spectrum evolution index REX, and comparing the toughness spectrum evolution index REX with a preset threshold value to obtain the evaluation core index. Grading and risk positioning are carried out on the current state of the building structure, a generated control strategy is sent to an execution layer, a damping adjusting device, a shock insulation layer controller or component repairing equipment is linked to complete structure regulation and control, and response is carried out after real-time acquisition and optimization.

The invention provides a digital modeling method for enhancing connectivity of a low-porosity rock core, a medium and equipment, and relates to the field of digital modeling of rock cores, and the method comprises the following steps: reconstructing a Gaussian random field based on SAXS data to obtain an initial model of a three-dimensional digital rock core pore structure; extracting pore center points of the initial model, and constructing an initial network connecting all the pore center points; based on a minimum spanning tree algorithm, identifying mutually isolated pore clusters in the initial network, and selecting a most efficient seepage path skeleton connected with the isolated clusters; based on the seepage path skeleton, constructing a throat with fractal characteristics; and according to the total volume of the throat, performing morphological corrosion operation on an original pore area in the initial model, embedding the constructed throat into the corroded model, and then performing controllable morphological expansion operation until the volume variation of the final model is smaller than a preset error threshold. According to the method, the reconstruction precision of the digital rock core in microstructure and macroscopic connectivity is effectively improved.

The invention belongs to the technical field of knowledge graph and big language model combination, discloses a big language model inference enhancement method based on structure perception knowledge graph representation, and provides a structure perception knowledge injection framework which utilizes a graph neural network to combine with multi-source structure information coding to enhance entity representation. And through a contrast learning mechanism, realizing alignment of semantics and a structure space, and meanwhile, designing an adaptive robustness fusion pruning training strategy, dynamically screening high-quality evidence sub-graphs, and guiding a language model to carry out structured reasoning. According to the method, deep integration and optimal utilization of the structure information and the semantic information are realized, the accuracy, robustness and resource efficiency of the model under a complex query task are remarkably improved, and a stronger embedding basis is provided for generation quality, so that the generation accuracy, stability and efficiency under a complex problem are improved.

The invention discloses a cartoon video generation method and system based on a video script, and relates to the technical field of video synthesis, and the method comprises the steps: based on structured script data, combining a predefined lens rule base and a reinforcement learning model, determining the type, duration and lens operation effect of a lens, and generating a lens splitting sequence through a dynamic lens splitting automatic generation mechanism, based on the split mirror sequence, generating an animation style key frame image through a diffusion model, selecting action data matched with the emotion label through a predefined action library, generating a voice waveform matched with the emotion label through a text-to-voice model, selecting a background music audio matched with the emotion label through a music library, and generating a multi-modal content stream; according to the method, the lens type, the duration and the lens operation effect are dynamically optimized through the reinforcement learning model, the defects of a traditional lens scheduling method based on static rule mapping in time continuity and narrative continuity are overcome, and the narrative fluency and the dynamic adaptability of the lens division sequence are obviously improved.

The invention discloses a laser radar multi-gas identification method and system based on deep learning, and relates to the technical field of laser radars, and the method comprises the steps: employing a differential absorption laser radar to emit lasers with different wavelengths, receiving echo signals after gas absorption, extracting multi-domain features, and employing multi-target optimization in combination with feature importance evaluation to screen key features; generating a gas portrait of the potential gas according to the key features, and obtaining the type of the interested gas; constructing a graph structure according to the gas portrait and the gas feature similarity, performing representation learning on the graph structure by using a graph convolutional network, enhancing the graph convolutional network by using an interested gas category, constructing a gas category branch, and outputting a gas category; gas concentration branches are built by adopting a multi-layer sensor, final node representation output by the graph convolutional network is used as input, and each gas concentration is output by using linear regression. According to the invention, the distinguishing capability of small difference gases is improved, and high-precision identification and concentration inversion of multi-component gases are realized.

The invention relates to the technical field of detection, in particular to a complex shell detection method and system based on structured light scanning. The method comprises the following steps: calculating a wrapped phase value and a phase modulation amplitude of each pixel point under each frequency, calculating discrete probability distribution containing candidate stripe series and a corresponding probability for each pixel point based on a multi-frequency wrapped phase difference, generating an initial pixel confidence map, and constructing an energy function; the penalty weight of the smooth item is determined according to the phase modulation amplitude and the wrapped phase gradient local variance, obtaining an initial fringe order distribution diagram, updating discrete probability distribution by using an initial pixel confidence map, re-minimizing an energy function to obtain a corrected fringe order distribution diagram, and reconstructing a final three-dimensional shape. And iteratively optimizing a projector principal point and a radial distortion coefficient in system calibration parameters. According to the scheme, the geometric details of the surface can be reserved, and the processing effect and the measurement precision of the complex curved surface and the high-curvature area are improved.

The invention discloses a robot operation track generation method based on structure perception and knowledge enhancement reasoning, and belongs to the field of artificial intelligence and machine learning, and the method comprises the steps: building a large-scale multi-modal database through obtaining sampling data of a real robot operation scene and synthetic data generated by a simulation environment; multi-source three-dimensional perception data is obtained, a large language model with a structure perception capability is input, and a machine execution instruction system with relatively strong three-dimensional space interaction understanding and dynamic modeling capability is obtained; multi-task progressive training is carried out based on spatial information understanding and interactive operation, capacity progressive unlocking is achieved, and robot tasks are strongly oriented; a large model knowledge graph enhancement method based on a three-dimensional scene object is introduced to assist in large model reasoning, and a large model which has reasoning ability and structure sensing ability and is higher in robustness to complex scenes is constructed and used for generating robot operation tracks.

The invention relates to the field of concrete detection, and discloses a concrete structure health real-time monitoring system based on multi-source sensing fusion, and the system comprises a data collection unit which carries out the collection of the physical field data of a concrete structure through a sensor array, and obtains the multi-source sensing data; performing structure response difference analysis on the multi-source sensing data to obtain concrete structure response difference data; according to the response difference data of the concrete structure, performing spatial and temporal distribution density analysis on a microcrack propagation path to obtain microcrack propagation spatial and temporal distribution density data; a data evaluation unit; multi-scale damage evolution path simulation is carried out through the structural damage time-space correlation characteristic data to obtain multi-scale damage evolution path data, the sensor array is used for collecting multi-source sensing data, structural response difference analysis is carried out, tiny changes of a concrete structure can be accurately captured, potential structural problems can be found in time, and the construction efficiency is improved. And a high-quality data basis is provided for subsequent analysis.

The invention discloses an image enhancement method and system based on semantic constraint degradation modeling. The method comprises the steps that semantic masks and multi-scale degradation features are extracted based on a low-resolution image used for training; performing deep fusion on the extracted semantic masks and the multi-scale degradation features based on a double-flow parallel architecture to generate semantic-structure fusion features; forming a multi-modal guide condition, taking the multi-modal guide condition and the semantic-structure fusion feature as input together, and reconstructing a high-resolution prediction image through a diffusion generation model; constructing a structure consistency optimization total loss based on the high-resolution prediction image and the corresponding target image, and optimizing a diffusion generation model based on the structure consistency optimization total loss; and inputting a low-resolution image to be predicted into the optimized diffusion generation model to obtain a high-resolution image corresponding to the low-resolution image. According to the scheme of the invention, comprehensive and refined understanding of low-resolution images is realized through multi-module cooperation and deep fusion.

The invention provides an exhibition hall dynamic layout planning method and system based on three-dimensional modeling, and relates to the technical field of simulation modeling, and the method comprises the steps: building a layout simulation model of exhibition hall layout planning; performing association behavior response constraint on a layout evolution topological structure in the layout simulation model according to a cross correlation coefficient of display response characteristics among different display units to obtain a layout constraint condition among the display units; determining the spatial adaptability loss of each display unit under the condition of displaying different contents according to the change characteristics of the space utilization rate of each display unit under the condition of displaying different contents and the behavior feedback characteristics of the audience when each display unit displays different contents; and target optimization is carried out on the spatial streamline of each display unit in the target exhibition hall based on the layout constraint condition and all the spatial adaptability losses, and an optimal structure path for dynamic layout planning of the exhibition hall is obtained. According to the method, cost optimization of the exhibition hall layout structure can be realized based on the simulation model, so that the robustness of dynamic layout planning of the exhibition hall is improved.

The invention discloses a nonlinear aerodynamic damping estimation method based on LSTM, and the method comprises the following steps: 1, building a nonlinear state space model of a structure based on structural response, including a state equation, an observation equation and a relation between nonlinear aerodynamic damping and structural vibration amplitude; 2, performing updating and covariance prediction on response data by using unscented Kalman filtering; 3, correcting the Kalman gain in real time by using a long short-term memory network; 4, performing state updating and covariance updating based on the corrected Kalman gain; 5, training the long-short-term memory network through an unsupervised learning mode, optimizing the filtering performance, and defining a mean square error of a posterior observation predicted value and a real observation value as a loss function; and 6, calculating the nonlinear aerodynamic damping according to the estimated nonlinear aerodynamic damping parameters. The invention further discloses a nonlinear aerodynamic damping estimation system based on the LSTM and a storage medium.

The invention provides a key component fault prediction method and system based on multi-source data fusion, and the method comprises the steps: S1, collecting key component data of a screen scarifier, constructing a multi-source heterogeneous sensor network, and completing the time-space alignment and quality optimization of multi-dimensional monitoring data; s2, a dynamic topological graph structure is constructed based on the physical connection relation of the components, the fault propagation path and strength are quantified, and multi-level feature representation covering the local state and the overall health degree is formed; s3, designing a hybrid prediction model fusing space-time analysis and memory modeling, and completing accurate description of an equipment degradation trend and early warning of potential faults through a self-adaptive feature integration mechanism; and S4, combining real-time prediction errors and historical operation and maintenance knowledge to dynamically optimize the hybrid prediction model, establishing a data-driven and knowledge-guided dual learning framework, and completing self-adaptive continuous learning along with equipment aging. The method breaks through the limitation of static state of a traditional prediction model, and the accuracy and reliability of a prediction result are remarkably improved.

The invention discloses a multi-source monitoring data fusion method based on deep learning, and the method comprises the steps: carrying out the standardization and space-time dimension alignment processing of multi-source monitoring data, constructing a multi-scale data sub-sequence, inputting a multi-scale space-time feature interaction network based on a Mamba structure, extracting and interacting multi-scale space-time features, constructing a cross-modal data topological graph structure, and carrying out the fusion of the multi-source monitoring data. And dynamically calculating and adjusting attention weights of nodes and edges of the topological graph by using an adaptive cross-modal graph attention mechanism, generating dynamically optimized cross-modal fusion features, performing collaborative feature decoding, and outputting a fusion result. According to the method, the multi-source data fusion precision and generalization ability in a complex monitoring scene are improved, and the fusion feature expression ability and decision reliability are improved.

The invention provides a model fine tuning method and system based on feedback and reinforcement, and relates to the technical field of natural language processing, and the method comprises the steps: obtaining an output text generated by a language model, and carrying out the embedded coding; identifying a structural semantic unit in the text, and generating structural mark information; constructing a low-layer capsule set based on the structure marking information, and executing dynamic routing to generate a high-layer semantic capsule set; constructing a structure expression matrix according to the mapping relation between the high-level semantic capsule set and the structure mark; and inputting the matrix into a reward scoring model to generate a reinforcement learning return value, and updating language model parameters according to the reinforcement learning return value. According to the method, closed-loop linkage of the language model structure perception capability and the strategy optimization path is realized, and the text generation structure and semantic consistency can be improved under the condition that manual annotation is not needed.

The embodiment of the invention provides a cross-modal knowledge distillation method and system based on dynamic structure perception, and the method comprises the steps: receiving a target task data set, dividing the target task data set to obtain a training set, and inputting the training set to a preset teacher model and a preset student model to obtain text features and visual features; determining a structural difference index in real time based on the parameter scale difference between the models, and obtaining a projection matrix based on the structural difference index; and obtaining a corresponding transmission matrix based on the real-time projection matrix, extracting attention distribution information of a preset teacher model, completing knowledge distillation from the preset teacher model to a preset student model based on attention distribution, and performing semantic segmentation on the target task data set to obtain a semantic segmentation result. According to the method, the defect that network models with different depths are difficult to align is effectively overcome, and the volume of the model for processing the target task is remarkably reduced.