103 results about "Source model" patented technology

The invention relates to the field of tunnel engineering, in particular to a tunnel multi-physics field coupling simulation and support design method considering construction disturbance, which comprises the following steps of: modeling a disturbance source, collecting disturbance factor information introduced in a tunnel construction process or an external environment, establishing a disturbance source model to represent the spatial position, intensity and evolution characteristics of a disturbance source; constructing a three-dimensional disturbance propagation function; coupling the three-dimensional disturbance propagation function with a multi-physical field model, and introducing disturbance as a source item into a tunnel heat-water-force coupling model; according to the construction method customization mechanism, modeling parameters and solving strategies are adjusted according to the characteristics of different tunnel construction methods, and corresponding disturbance source model parameters, boundary conditions and constitutive relation adjustment coefficients are set according to the different construction methods. The problems that disturbance modeling is rough, a disturbance propagation mechanism is idealized, the disturbance propagation mechanism is disjointed with a multi-physical field model, and a construction method is not included in a disturbance response system are solved.

The invention provides a BIM (Building Information Modeling) full life cycle splitting and cross-model synchronization method and system. The method comprises the steps that BIM model components are coded based on a full life cycle coding rule, private attributes of all engineering stages and component codes of the engineering stages are bound and stored in component extension attributes, and a BIM master model containing multi-stage attribute information is generated; setting a BIM model splitting rule based on a rule engine; the code of each component in the BIM model is analyzed, the codes are compared with screening conditions in the splitting rules according to levels, and component data corresponding to the successfully matched codes are packaged and stored as sub-models; establishing association mapping between the models; and when the source model is modified, the synchronization engine executes synchronous updating of the target model according to the association mapping between the source model and the target model associated with the source model, so that the cross-stage multi-participant cooperation efficiency is improved, and a solid technical support is provided for digital transformation of the construction industry.

Disclosed are example embodiments of systems and methods for virtual try-on of articles of clothing. An example method of virtual try-on of articles of clothing includes selecting a garment from a pre-existing database. The method also includes loading a photo of a source model wearing the selected garment. Additionally, the method includes generating a semantic segmentation of the model image. The method also includes extracting the selected garment from the photo of the model. Additionally, the method includes determining a correspondence between a target model and the source model by performing a feature point detection and description of the target model and the source model, and performing feature matching and correspondence validation. The method also includes performing garment warping and alignment of the extracted garment. Additionally, the method includes overlaying and rendering the garment.

The invention discloses a physical simulation method and device for seismic response certainty of an urban underground rail transit network, and the method comprises the steps: constructing a mixed seismic source model based on the active fault parameters and kinematic parameters of a target region, the seismic source model integrates a low-wavenumber deterministic slip component and a high-wavenumber random slip component on a fault fracture surface to generate a broadband seismic wave field source reflecting the physical process of a seismic source; the constructed seismic source model is used as input, a regional crustal velocity structure model is combined, propagation of seismic waves in a layered geological medium is simulated through a frequency-wavenumber domain method, and a seismic oscillation space-time field is obtained through calculation; and establishing a finite element model of the urban underground rail transit network, applying the seismic oscillation time-space field as non-consistent seismic oscillation input to the finite element model, and solving through dynamic time-history analysis to obtain the dynamic response of the whole underground rail transit network in the earthquake process. According to the invention, a complete physical chain from seismic source fracture to structural response can be reproduced.

A computer-implemented method of performing probabilistic seismic hazard analysis (PSHA) includes receiving a seismic source model, a ground motion model, and a probability distribution, constructing a joint probability distribution, using adaptive importance sampling to obtain an optimal sampling density, computing a PSHA hazard estimate and a hazard deaggregation distribution, and providing the PSHA hazard estimate and the hazard deaggregation to a user. Another computer-implemented method includes estimating a mean and fractile PSHA hazards for each of a series of iterations by drawing a random sample set from a proposal sampling density, evaluating an indicator function for each sample and multiply by an earthquake occurrence rate and a likelihood ratio to produce a weight vector, and estimating a mean hazard from an average of the weight vector.

The invention discloses a machine tool feeding system fault diagnosis method based on multi-level current characteristic distillation. The method comprises the following steps: collecting vibration and current time domain signals and respectively converting the vibration and current time domain signals into two-dimensional time-frequency diagrams; training a source model by using the vibration time-frequency diagram; cloning and freezing a source model, constructing a target model with the same structure, inputting the current time-frequency diagrams into the source model and the target model at the same time, and optimizing the target model through combination of multi-level feature distillation and a multi-target loss function; and finally, fault diagnosis based on the current signal is realized by using the optimized target model. According to the method, source domain data does not need to participate in training, knowledge migration from a vibration mode to a current mode is realized, the problems of weak current signal fault features and scarcity of labeled samples are effectively overcome, and the fault diagnosis precision and cross-working-condition adaptive capacity under the small sample condition are improved.

The invention belongs to the technical field of unmanned driving environment perception, and provides a night semantic segmentation method and system based on passive multi-level collaborative distillation, and the technical scheme is as follows: firstly, initializing a teacher network and a student network based on pre-training model parameters under a normal illumination condition; then, teacher network parameters are fixed, and a teacher network is updated according to student network parameters in a momentum smooth propagation mode; then, based on a prediction result of a pre-training source model on the night image, selecting first K pixel points with the highest confidence coefficient to generate a pseudo tag; and finally, inputting the normal illumination image and the night image into a teacher network and a student network respectively, and realizing multi-level knowledge migration through structure perception level alignment, semantic consistency constraint optimization and frequency domain collaborative fusion. No night annotation data is needed, source domain original data does not need to be accessed, and the data cost and the privacy risk are remarkably reduced.

The application provides a high-efficiency expression migration method using base expression space transformation, comprising a preprocessing stage and a facial expression migration stage; the preprocessing stage comprises: obtaining a source model O, the source model O being a base expression model comprising expression meshes under multiple different expressions; performing facial expression reconstruction on the source model O and a target model T to obtain expression description parameters of the two; performing similarity estimation according to the expression description parameters to obtain an expression parameter conversion matrix and a similarity matrix; the facial expression migration stage comprises: inputting a character picture into the base expression model, calculating expression description parameters of a target expression model according to the expression description parameters in the base expression model and the similarity matrix, and completing expression migration.

This information processing apparatus comprises: a synthesis source generation unit that is configured to train a plurality of synthesis source models using a data set which contains certain data and train a plurality of synthesis source models using a data set which does not contain said certain data; a selection unit that is configured to select a plurality of sets of two synthesis source models from the synthesis source models; a synthesis unit that is configured to synthesize the two synthesis source models of each of the plurality of sets to generate a synthesis model; and a distribution calculation unit that is configured to calculate a first distribution regarding outputs produced upon the input of said certain data into the plurality of synthesis models which contain, as synthesis sources, the synthesis source models having learned said certain data, and a second distribution regarding outputs produced upon the input of said certain data into the plurality of synthesis models which do not contain, as the synthesis sources, the synthesis source models having learned said certain data. This makes it possible to evaluate, from synthesized models, whether certain data is used in training synthesis source models.

The specification discloses a model training method and device, equipment and a readable storage medium. According to a to-be-trained transformation network and a source model trained by a poisoned sample, a target model to be trained is determined. A clean sample obtained is input into the target model to obtain a prediction result output by the target model. According to the prediction result and a label of the clean sample, parameters of the transformation network are adjusted. When a prediction request is received, to-be-predicted data is input into the trained target model to obtain a prediction result of the to-be-predicted data. It can be seen that the reconstruction of the source model based on the transformation network avoids the input and / or output of the source model attacked by the poisoned sample being transformed by the transformation network without adjusting the model parameters of the source model, so as to avoid the source model obtaining an incorrect prediction result for the sample preset by the attacker, thereby ensuring the model performance and realizing a high defense effect and protecting the safety of private data.

The application discloses a kind of passive cross-domain adaptive image classification methods based on multi-modal pre-training model, it is related to transfer learning technical field, first by quantitative estimation target domain data on source model existing category similarity matrix;Then the category similarity matrix of source model is introduced into CLIP (multi-modal pre-training model), directional generation confusion category text feature is used for classification, and the corresponding classification result is obtained;Then the classification result of source model and the classification result of CLIP are weighted, and the output of two models is respectively aligned with the weighted result;Again, the category feature center library of each model is calculated according to the category similarity matrix;Finally, the projection of input image on two feature center libraries is aligned using attention network to realize the alignment of feature space.The application dynamically estimates the similarity degree between categories in downstream task from the angle of model, learns using the complementary information between models, and realizes the fine-grained classification between similar categories, and has certain explainability.

The invention discloses a structured light and texture feature fused three-dimensional face model registration method and system, and relates to the field of face model registration, and the method comprises the steps: firstly carrying out the point-by-point analysis of a source model, and calculating the significance degree of local geometric features and the richness degree of texture information of the source model; an exclusive confidence score is generated for each point on the model. Then, respective geometric and luminosity weights are dynamically generated for each point based on the confidence score. A higher geometric weight is automatically given to an area with a prominent geometric structure and a single texture; however, the luminosity weight can be improved in a region which is geometrically flat but rich in texture. Finally, the group of adaptive weights capable of accurately reflecting the reliability of local information are integrated into a fusion cost function, and a registration optimization process is guided to intelligently focus on a more reliable information source, so that interference caused by feature imbalance is effectively overcome, and three-dimensional face model registration with higher precision and robustness is realized.

This provides a real-time simulation and model prediction method for arc additive manufacturing based on event sequences. [Solution] In a real-time process of arc additive manufacturing of metal structures, the method includes the steps of: activating units in real time by event sequence and guiding the heat source in real time; setting parameters for the heat source model, thermal simulation, and mechanical simulation, and performing a real-time thermal-mechanical simulation of arc additive manufacturing; and constructing a theoretical model for a simplified calculation of the residual stress field and correcting the calculation model prediction.

The invention discloses a broadband seismic oscillation simulation method and system based on probabilistic seismic risk analysis, and relates to the technical field of field seismic oscillation, and the method comprises the steps: carrying out the seismic risk decoupling through employing a Chinese probabilistic seismic risk analysis method, determining a target potential source area corresponding to the maximum contribution degree with probability significance and a potential source magnitude and an epicentral distance of the target potential source area; calculating a fault size parameter of the target potential source area based on the potential source magnitude and a preset scaling rate; calculating fault distribution parameters of the target potential source area based on the fault size parameters and the mixed seismic source model; combining the three-dimensional longitudinal wave velocity model and the shallow structure model to construct a three-dimensional calculation model; and performing deterministic low-frequency simulation and deterministic high-frequency simulation on the three-dimensional calculation model based on a spectral element method and a three-way stochastic finite fault method respectively, and then performing broadband time program sequence synthesis to obtain a broadband seismic oscillation simulation result. The technical problem of low simulation result accuracy in the prior art is relieved.

Techniques are provided for adaptive usage of storage resources using data source models and data source representations generated using the data source models. One method comprises obtaining sampled data generated by sampling source data from a data source; fitting a data model to the sampled data to obtain a representation of the sampled data; obtaining a classification of the sampled data into one of multiple predefined retention models; and adapting a usage of one or more storage resources that store the retained data based on the representation and the classification. The adaptive storage resource usage may comprise, for example: (i) varying a data retention model based on an age of the sampled data; (ii) evicting cache data based on the representation; (iii) moving the retained data to a different storage tier; and (iv) determining an amount of time to store the retained data.

The invention relates to a BlendShapes transmission method and a BlendShapes transmission system based on UV mapping. The method comprises the following steps: receiving a basic model of a source model, at least one BlendShapes deformation model corresponding to the basic model of the source model and a basic model of a target model; constructing a mapping relation between the target model and the source model based on the UV grid space; calculating the vertex offset from the basic model of the source model to the BlendShapes deformation model of the source model; calculating a vertex displacement vector of the target model based on the vertex offset and the mapping relation; and superposing the vertex displacement vector of the target model to a basic model of the target model, and generating a BlendShapes deformation model corresponding to the target model. By the adoption of the method, topological limitation can be broken through, and when any topological difference exists between the source model and the target model, BlendShapes transmission can still be achieved.

A method of semiconductor manufacturing process parameter optimization based on Bayesian average Kriging and evidence theory is proposed. By introducing Bayesian model averaging into the basic Kriging model, different variance functions are adaptively weighted, which effectively improves the robust estimation of spatial correlation structure and alleviates the bias caused by model mis-specification under small sample conditions. At the same time, the introduction of evidence theory not only realizes the fusion of multi-source model prediction results, but also explicitly describes the cognitive uncertainty at the model level, enhancing the interpretability of the prediction results and the value of process diagnosis. The method shows good applicability in high-cost and data-scarce manufacturing systems, and can provide reliable proxy modeling support for complex processes such as semiconductor etching.

This invention discloses a source model method for concave-convex bodies considering random effects, belonging to the field of civil engineering technology. The method includes the following steps: S1, determining the source and concave-convex body distribution parameters; S2, calculating the random concave-convex body model; S3, classifying the random uniformity; and S4, calculating the coseismic displacement of the ground surface. This invention employs the aforementioned method for a source model of concave-convex bodies considering random effects, primarily utilizing the rearrangement of random quantities to form concave-convex body characteristics without altering the redistribution law of random quantities. This establishes a complete source model method for concave-convex bodies considering random effects, enabling the source model to more realistically reflect the actual situation. This improves the accuracy of surface and stratum deformation calculations, provides effective support for the seismic design of fault structures, and has strong practicality.

The embodiment of the invention provides a transferable black box injection disturbance method and device for a heterogeneous graph neural network, and relates to the field of graph neural network security of artificial intelligence. The method is used for solving the problems that an existing disturbance method for a heterogeneous graph neural network is insufficient in universality and mobility of a disturbance strategy, high in calculation overhead and poor in perturbation efficiency due to the fact that a common vulnerable point of'relation sub-graph semantic difference 'in a heterogeneous graph structure is not fully mined and utilized and excessively depends on gradient information of a source model or a substitution model. And different target models are difficult to adapt. According to the method, firstly, an original heterogeneous graph is split into relation sub-graphs according to relation types, key vulnerabilities of the relation sub-graphs are mined through multi-view semantic voting scores, malicious features are generated based on cross-relation semantic aggregation, disturbance sub-graphs are constructed and integrated into a disturbance heterogeneous graph, the whole process does not need to depend on model gradient information, and the perturbation efficiency is improved. The method can pertinently utilize the semantic difference of the relation subgraph, improves the universality and mobility of attacks, and reduces the calculation overhead.

The present application relates to the technical field of radiation dose calculation, and particularly relates to a method and system for automatically modeling radiation dose, wherein on one hand, a plurality of intelligent optimization algorithms are used to perform step-by-step and region-by-region optimization, to obtain a radiation source model corresponding to each intelligent optimization algorithm, and then to obtain an optimal radiation source model, so that automatic modeling can be realized, the efficiency is high, and the radiation source model can be optimized to reduce errors; on the other hand, step-by-step and region-by-region optimization is performed in a search interval of a target empirical energy spectrum, so that convergence is easy, and the problem that a solution found due to an excessively large search range is easy to fall into local optimization can be avoided.

Aligning a source and target model includes calculating a shape descriptor for a plurality of edge points, grouping the edge points by shape descriptor and selecting representative points for each group. Target and source point pair features (PPFs) are calculated between pairs of representative points on the target and source models, where PPFs defines the relative position and orientation of point pairs. Target PPFs are matched with each source PPF and the point pairs associated with the matched PPFs are transformed to align the location and edge direction of a target PPF point with the location and edge direction of a source PPF point. An angle is determined to align the second target PPF point with the second source PPF point, and a modal angle is found among the determined rotation angles. An output transformation is calculated using the transformations associated with the modal angle.

This invention discloses a method and system for intelligent prediction and dynamic scheduling of irrigation district water resources based on multi-source model fusion, belonging to the field of water conservancy informatization. The method achieves prediction of inflow floods and various types of water demand through multi-source data acquisition and preprocessing, and multi-model fusion. It generates scheduling schemes through intelligent evaluation and decision-making, completes coordinated control of gates and pumps and engineering safety early warning, and combines visualization display and intelligent safety monitoring. It also conducts full lifecycle management and dynamic optimization of the entire model. The system includes corresponding modules for data processing, multi-source prediction model layer, intelligent evaluation and decision-making model layer, intelligent control and early warning, visualization display, safety monitoring, and model management, with each module operating collaboratively. This invention achieves intelligent scheduling of irrigation district water resources throughout the entire process, improves prediction accuracy and water resource utilization efficiency, has strong drought and flood prevention capabilities, and exhibits good model versatility and scalability, making it suitable for intelligent management of various irrigation districts.

A cross-modality knowledge transfer system is provided for adapting one or more source model networks to one or more target model networks. The system is configured to perform steps of providing the TI paired datasets through the source feature encoders of the one or more source model networks, extracting TI source features and TI source moments from the TI paired data by the BN layers of the one or more source model networks, providing the TI paired datasets and the unlabeled TR datasets through the one or more target model networks to extract TI target features and TR target moments, training jointly all the feature encoders of the one or more target model networks by matching the extracted TI target features and TR target moments with the TI source features and TI source moments along with mixing weights, and forming a final target model network by combining the trained one or more target model networks.

The application provides a water environment model coupling data processing method and system, and relates to the technical field of hydrology and water quality monitoring. The method comprises the following steps: determining the downward trend of soil permeability characteristics and estimating the uncertainty range of runoff coefficient; using a regional source-sink pollution model, obtaining source model data according to the uncertainty range of runoff coefficient; using a one-dimensional river network model, obtaining error information and outputting prediction data according to the source model data; and determining the prediction deviation mode by comparing the prediction data and real-time monitoring data and correcting the parameters of all related models in the whole model chain. The method aims to solve the technical problem that the prediction of the river network model deviates from the actual measurement data in a nonlinear manner in a three-model coupling system, which ultimately leads to the inconsistency between the permitted discharge amount set based on the system and the actual water body absorption capacity, and to realize the accurate analysis of the response relationship between the urban pollution load and the water environment quality and the real-time and self-adaptive adjustment of the permitted discharge amount.

This invention discloses a transferable adversarial texture coating generation method and system. The scheme includes a multi-model collaborative training section, establishing a source model and target model collaborative training and two-layer optimization mechanism. The inner layer introduces target feature perturbations to generate adversarial camouflage samples, while the outer layer employs weighted average loss and transferability regularization constraints, enabling the adversarial texture coating to have transferability across detector models and target carriers. In the feature space alignment section, feature layers of different architecture models are matched, and dynamic feature layer matching and multi-scale gradient alignment loss functions are calculated to achieve consistent alignment calibration of shallow and deep feature spaces. A transferability evaluation index is introduced during the alignment process to evaluate the alignment effect in real time. In the three-plane mapping texture generation section, target geometric features are extracted, surface curvature distribution regions are calculated, dynamic weights are assigned to different regions, and weighted fusion texture generation is performed to achieve accurate texture mapping and coverage of the adversarial texture coating on complex three-dimensional surfaces of different target carriers.

The application discloses a cross-subject electroencephalogram signal classification method based on online test time domain adaptation, and steps of the method comprise the following steps: 1, pre-processing original EEG data, including removing noise, segmenting, extracting time-frequency features by using short-time Fourier transform, and obtaining source domain data and target domain data; 2, constructing a source model, a student model and a teacher model based on a CNN network, and training the source model by input data to obtain a pre-training source model; 3, initializing the student model and the teacher model by using the pre-training source model; 4, online optimizing the student model and the teacher model based on a mutual learning strategy on target data flow and realizing classification of electroencephalogram signals. The application can realize rapid classification of electroencephalogram signals under the condition of protecting the privacy of patients, so that the real-time demand of the classification system of electroencephalogram signals in an actual scene can be met.

This invention discloses a holiday load prediction method based on ANN and transfer learning, comprising the following steps: performing feature engineering operations on the preprocessed dataset to form a full load dataset; extracting the holiday load dataset and the festival load dataset from the full load dataset; constructing a full data source model using an ANN, and training the full data source model using the full load training dataset; using the full data source model as a pre-trained model to train the festival / holiday load dataset, using the same parameters as the full data source model, to obtain the final festival / holiday load training model; predicting the load values ​​at each time point of the prediction day, obtaining the prediction results for the load at each time point of the prediction day. By using ANN to predict the holiday load data and the festival load data separately for holiday prediction, the characteristics of each holiday can be better learned, thereby improving the prediction accuracy.

The invention provides a domain self-adaption method and device based on utility-driven sample fusion and dynamic knowledge distillation and electronic equipment, and relates to the technical field of risk control and artificial intelligence, and the method comprises the steps: constructing a utility function based on a domain similarity score and a prediction reliability score, screening out source domain samples with first N% of utility value based on the utility function, fusing with a target domain sample to form an enhanced training set; based on the domain similarity score, determining the importance weight of each source domain sample with the first N% of the utility value in the enhanced training set, and correcting the distribution offset between the source domain sample with the first N% of the utility value and the target domain sample based on the importance weight of the source domain sample with the first N% of the utility value; and performing dynamic knowledge distillation on the corrected enhanced training set by using a composite loss function to obtain a target model. According to the method, the high-quality source domain sample which is close to the target domain in distribution and has high source model prediction accuracy and high confidence can be accurately screened out.

The invention discloses a consistency calibration method and system for scores of a multi-source model, and particularly relates to the technical field of distributed calibration, and the method comprises the steps: carrying out score access packaging, carrying out the structured packaging of the scores of the multi-source model, and generating a calibration parameter package containing a calibration mapping parameter and a calibration fingerprint; performing parameter preloading activation, and realizing consistent switching of calibration parameters among distributed nodes through a two-stage release mechanism and by taking a calibration version identifier as an anchor point; link version locking: binding a calibration version identifier at a request entry and carrying out unvarnished transmission along a link to ensure that the same request completes division by using the same set of calibration mapping in a distributed multi-node environment; and consistent monitoring self-healing is carried out, the consistency risk is quantified through multi-granularity monitoring, and anomaly detection, automatic loss stopping and controlled rollback are realized based on release of coupling closed-loop control of convergence credibility and a consistency risk index.

The invention discloses a passive domain adaptive method based on prototype guided contrast learning, and belongs to the technical field of computer vision and transfer learning. The method specifically comprises a source model training stage and a target domain self-adaption stage. A target domain initial prototype is constructed through source domain pre-training model category features, and cross-domain features are fused by adopting'average mapping + maximum mapping 'so as to reduce distribution differences; meanwhile, prototype orthogonal loss and'instance-instance 'and'instance-prototype' double contrast learning mechanisms are introduced, and feature discrimination and semantic association are enhanced. Besides, by integrating multiple loss items to construct an overall optimization target and utilizing end-to-end training to optimize model parameters, the problems of insufficient data utilization rate, false label noise interference and neglect of semantic association of an existing passive domain adaptive method are effectively solved, and the target domain feature representation robustness and the model cross-domain migration performance are remarkably improved.