52 results about "Probability inference" patented technology

The invention relates to the technical field of hydrological prediction, and discloses a flood prediction method and system based on machine learning, and the method comprises the steps: constructing a multi-source data probability representation layer, and converting multi-source hydrological data into probability distribution representation through a variational auto-encoder; a Bayesian depth feature extraction layer is constructed, and a random weight neural network is adopted to extract hydrological features from data represented by probability distribution; constructing a causal structure learning layer, and learning a causal relationship between variables from hydrological features by using a causal discovery algorithm; a probability inference prediction layer is constructed, and prediction distribution of hydrological variables is generated through a Markov chain Monte Carlo method; an interpretable anti-fact analysis layer is constructed, and hypothesis analysis and decision evaluation of a flood scene are supported; according to the method, the transformation from determinacy prediction to probability prediction is realized, the interpretability is enhanced through causal learning and symbol probability hybrid reasoning, and reliable technical support is provided for flood control decision making.

The invention relates to the technical field of concrete detection, in particular to a concrete structure crack damage evaluation system based on acoustic emission sensing, which comprises a signal acquisition module, a characteristic discrete evolution module, a probability inference module, an energy gradient analysis module and a boundary defining module. According to the method, a multi-dimensional evolution set is constructed by extracting the amplitude and energy standard deviation of an acoustic emission signal in a continuous time window, the synchronous change trend of the multi-parameter standard deviation is analyzed by using a Bayesian probability model, and the non-uniform expansion posterior probability is calculated to lock a key signal set of dominant damage expansion. An energy fluctuation coefficient is calculated and a space sequence is generated by combining sensor space coordinates, and a fluctuation coefficient stable interval is identified according to a gradient attenuation rule of energy along with a distance, so that a physical boundary of a crack damage dynamic active region is quantitatively defined, and accurate evaluation of a non-uniform expansion state and an active range of a concrete crack is realized.

The invention relates to the technical field of stereoscopic vision, and discloses a synchronous speed visual matching method and system, electronic equipment and a storage medium, and the method comprises the steps: synchronously collecting a stereoscopic video sequence with a predefined frame rate; executing multi-target hybrid tracking and motion induction detection, and outputting target motion information including position and velocity vectors; extracting hierarchical motion features of the target from continuous multiple frames of the stereoscopic video sequence, and performing unified space-time coding; under geometric constraints of stereoscopic vision, scale cosine similarity, direction similarity and trajectory consistency measurement are calculated and serve as observation evidences to be input into the probabilistic reasoning model for fusion, and a posterior probability representing matching reliability is output; the weight distribution of the speed similarity and the direction similarity is adjusted according to the motion characteristics of the targets in the scene, and the stable corresponding matching relation between the left view target and the right view target is established. According to the method, high-time-resolution information can be utilized, motion features and geometric constraints can be effectively fused, and the method has self-adaptive capacity.

The invention relates to a digital-intelligent automatic point alignment and full information verification method based on a power distribution terminal tester, which comprises the following steps of: acquiring electrical quantity signals and operating parameters of a terminal through the tester, and establishing communication connection through a digital-intelligent point alignment module; starting a three-remote check matrix model to perform matrix check on telemetering, remote signaling and remote control functions, calling an FFT harmonic synthesis telemetering calibration algorithm to dynamically calibrate a deviation value, introducing an abnormal attribution model based on a Bayesian network to perform probability reasoning on abnormal data, generating a report containing abnormal information and calibration suggestions, and uploading the report. The method improves the verification efficiency and overcomes the problem that the existing mode is tedious; checking accuracy is improved, and the problem that measurement deviation is difficult to correct is solved; abnormality processing capability is enhanced, defects of insufficient attribution capability are made up, stable operation of the power distribution terminal is guaranteed, and support is provided for power grid safety.

The invention relates to the technical field of dust pollution treatment, in particular to a port dry bulk cargo storage yard electronic greenhouse closing system based on multi-dimensional full-time management and control. According to the system, a time reference is established by identifying an operation event, multi-source sensor data is mapped to a unified time axis, and space-time registration is realized by combining Bayesian probability inference; constructing a physical information neural network based on the registration data to predict a continuous dust concentration field; a virtual closed boundary is generated according to the concentration threshold value, and the closing effectiveness is evaluated; calibrating physical parameters by using a CFD model in combination with a parameter inversion method; and in the rolling prediction time domain, constructing an optimization model weighted by the spraying cost, the environmental protection penalty and the control cost, solving an optimal spraying control strategy, and feeding back an execution effect to the model to form closed-loop optimization. By means of the system, the raised dust of the storage yard can be precisely treated, and it can be guaranteed that the raised dust is sealed in the electronic greenhouse in an open-air scene.

A single-cell level spatial omics multi-modal fusion method, comprising: extracting differential expression genes and nuclear spatial morphological features from spatial transcriptome data, single-cell sequencing data and histological images, and realizing field adaptation between different platforms by using conditional variational autoencoder. Based on a probability inference model, spatial transcriptome expression, single-cell omics and morphological features are fused to jointly infer the type and gene expression level of each cell. A spatial cell network is constructed by a graph attention mechanism to realize the spatial diffusion and recognition of cell types in the whole slice range. Combined with a multi-omics enhancement module, the unmeasured gene and protein expression are completed based on expression similarity, and the prediction consistency is improved through spatial correction. The method realizes high-resolution reconstruction of single-cell multi-omics information in three-dimensional space, improves the information coverage and spatial resolution of spatial omics data, and provides an efficient and low-cost solution for spatial biology and precision medicine research.

The application relates to the technical field of unmanned aerial vehicle swarm detection, and discloses an unmanned aerial vehicle swarm detection method and system based on multilayer early warning, which comprises the following steps: collecting radar signal and photoelectric signal data and unmanned aerial vehicle speed, generating initial data by unified mapping according to preset state transition and observation matrix, making prediction-residual correction by combining preset noise parameters to obtain a dynamic trajectory, smoothing the trajectory, statistically analyzing the trajectory according to height granularity, adaptively clustering to form layered height intervals, identifying dense clusters according to distance radius and core points when the number of layers exceeds a threshold, extracting cluster centers and density gradients, constructing a probability inference structure according to the above and the speed, fusing prior formation, wind disturbance and delay, outputting a dynamic threat level, calling a pre-generated heat map to position a high-threat area when the threat exceeds a threshold, combining airspace safety constraints and speed to predict a future trajectory, updating a situation, adaptively configuring radar beams and coverage, and outputting real-time detection data. The method can realize reliable airspace management.

This invention provides a collision detection method and system for robotic arms based on Bayesian inference, relating to the technical fields of robot safety perception and human-computer interaction. The method includes: constructing a generalized momentum model containing unknown dynamic terms and external joint torques based on a joint spatial dynamics model of the robotic arm; performing probabilistic inference on the unknown dynamic terms using an integrated Bayesian neural network, outputting the predicted mean and variance of the unknown dynamic terms; constructing a data-driven adaptive momentum observer based on the predicted mean and variance to generate an estimated signal of the external joint torques; and performing collision detection based on sparse Bayesian inference on the estimated signal based on the sparsity characteristics of external collision perturbations to generate a collision determination signal. This method solves the technical problems of insufficient sensitivity and high false alarm rate in existing collision detection technologies, achieving improved accuracy in external torque estimation and collision detection sensitivity.

The invention provides an enterprise cash flow intelligent prediction and risk early warning method and system, and relates to the technical field of data processing, and the method comprises the steps: obtaining enterprise cash flow data; constructing a cash flow stability tensor according to the enterprise cash flow data; dividing the cash flow stability tensor into a committed cash flow tensor and a realization cash flow tensor; determining a state description vector according to the committed cash flow tensor and the implementation cash flow tensor; according to the state description vector, constructing a hidden semi-Markov model; performing state prediction on the enterprise cash flow through the hidden semi-Markov model to obtain enterprise cash flow state distribution; constructing a risk infection graph according to the enterprise cash flow state distribution; probability inference is carried out on the risk infection graph to obtain the risk probability of the enterprise cash flow; and carrying out risk early warning according to the risk probability.

The application provides a kind of based on artificial intelligence chemical substance fast identification method and system, belong to intelligent detection analysis technical field;The method includes the preprocessing and structural conversion processing of physical signal and social context information, extracts the physical evidence characteristics and social evidence characteristics of the sample to be measured;According to physical evidence characteristics and social evidence characteristics, through double drive hypothesis generation, the bayesian inference network containing candidate substance hypothesis node, physical evidence node and social evidence node is constructed;Solving the posterior probability value corresponding to each candidate substance hypothesis, according to the posterior probability value and evidence chain analysis result executes decision, generates and outputs chemical substance identification report.The application generates double drive hypothesis by fusing physical signal and social context information, and uses bayesian network to carry out probability inference, which significantly improves the accuracy and reliability of chemical substance identification in complex scene.

The invention relates to a seal use safety control method, system and device based on multi-source feature fusion and a medium. The method comprises the steps that a to-be-stamped paper surface image and a stamping candidate area are obtained, illumination normalization and geometric correction preprocessing are conducted on the to-be-stamped paper surface image, and a standardized stamping paper surface image is obtained; performing multi-source feature extraction on the standardized stamping paper surface image to obtain a feature set; performing space and channel alignment and fusion processing on the feature set to obtain fusion features; pixel-level decoding is carried out according to the fusion features to obtain a blank probability graph, and global blank probability inference is carried out according to the blank probability graph to obtain a blank determination score and a confidence coefficient thereof; based on a preset double-threshold mechanism, generating a control decision and audit record according to the blank probability graph, the blank judgment score and the confidence coefficient; the control decision is one of blocking, warning and permission. By adopting the method, high-precision white paper identification can be realized so as to realize printing compliance control.

The invention relates to the technical field of stroke monitoring, and discloses a method and a system for detecting the stroke risk after waking of a home elder based on water and electricity information, and the method comprises the steps: synchronously obtaining household water, electricity and body data in a morning waking monitoring time period, and generating a morning waking water event sequence and an electricity load sequence through flow sequence form segmentation and electricity event steady-state analysis; morning awakening physiological indexes are formed according to the skin electricity, the heart rate and other data; determining a unified time reference according to a morning beginning behavior completion moment or a physiological composite sign, constructing an alignment time window, and mapping the three types of data to a discrete time grid; establishing double-hypothesis probability inference, respectively constructing water utilization, power utilization and body three-channel observation likelihoods, and performing joint fusion with a dependency relationship to obtain a current occurrence probability; and establishing a survival analysis model by using the time variation covariable in the aligned time window, and giving the occurrence probability of the future time period. The method does not depend on videos, and the judgment capability can be kept when the body data are missing.

The invention relates to the technical field of artificial intelligence, in particular to a student high-order thinking ability assessment method, device and equipment based on project-based learning behavior data, a storage medium and a computer program product. The method comprises the following steps: obtaining multi-source heterogeneous behavior data of a to-be-evaluated student in a project-based learning process to obtain a to-be-processed behavior data set; performing cleaning, standardization and cross-source alignment processing on the to-be-processed behavior data set, and performing fusion based on the aligned data to obtain unified behavior data; performing feature engineering processing based on the unified behavior data to extract a behavior feature set for representing high-order thinking related cognitive activities; based on a preset learning model, extracting behavior evidences from the behavior feature set, fusing the behavior evidences and performing probability inference; according to the incidence relation between the behavior evidence and the preset thinking index and the ability dimension and the probability inference result, the target assessment result is determined, and the accuracy of student high-order thinking ability assessment is improved.

The application discloses a kind of water transport ship traffic collaborative control methods based on deep learning, specifically includes: obtaining ship operation data set, time index alignment is carried out, and state vector sequence is generated;State vector sequence is encoded to feature, and deep boltzmann machine model is constructed;First conditional energy function, second conditional energy function and energy coupling matrix are established between model layer, and form hierarchical conditional energy function set;Hierarchical conditional energy function set is carried out to hidden variable probability inference, and hidden layer probability state vector sequence is generated;Parameter flow shape is divided to energy function parameter, and policy optimization is carried out under the constraint of confidence domain, and updated energy function parameter set is generated;Updated energy function parameter is carried out to energy calculation, and state probability distribution sequence is generated;State probability distribution sequence is carried out to strategy mapping, and ship traffic collaborative control strategy is generated.The application constructs hierarchical energy coupling model, realizes the collaborative control of multiple ships state.

An intelligent anti-theft electricity distribution system, method, electronic device and storage medium, the system comprises: a data acquisition module configured to acquire current data, power data and temperature data of a power distribution line; a data processing module configured to perform spatio-temporal alignment processing on the current data and the temperature data to obtain a current temperature data matrix, and construct a node current balance equation based on the current temperature data matrix and the power data; a confidence generation module configured to construct a Bayesian network based on a preset historical electricity theft case library, input a calculation result of the node current balance equation into the Bayesian network for time series probability inference calculation to obtain a electricity theft behavior confidence; an instruction generation module configured to match a preset risk level according to the electricity theft behavior confidence, and generate a risk disposal instruction according to the risk level; and an execution module configured to determine a target line to perform a target operation according to the risk disposal instruction. The technical solution provided by the present application can solve the technical problem of single detection dimension.

The invention relates to the technical field of account supervision and management, and discloses a chess and card game account transaction monitoring system which comprises the following steps: classifying asset transfer according to a business trigger source, and binding game information for game result type transaction as a traceability anchor point; a conservation verification gateway is arranged to carry out instant arbitration on game result type transactions, a traceable value balance which is adjusted only according to verified game net income and expenditure is established for an account, when point-to-point type transactions are processed, the transaction amount is compared with the balance to add a deterministic risk qualitative label, and the point-to-point type transactions are processed. According to the method and the system, credential management is realized at a data source, so that a monitoring system is converted from probabilistic inference on hybrid data into deterministic auditing based on a clear fund source, and therefore, monitoring resources are accurately focused on high-risk transactions, and supervision efficiency and auditing performance are improved.

The invention relates to the technical field of computer vision, and particularly discloses an abnormity monitoring safety early warning system based on computer vision. The system comprises a multi-source heterogeneous data acquisition module, a spatio-temporal feature self-adaptive extraction module, a multi-modal abnormal probability inference module, a dynamic risk quantitative evaluation module and a hierarchical early warning decision and execution module. High-precision detection and graded early warning of abnormal behaviors are realized, and the robustness, generalization ability and response efficiency of the system in a variable environment are improved.

The present application relates to the technical field of heat-conducting material detection, and discloses a heat-conducting gel water resistance detection method and system based on a sensor. The method comprises receiving real-time monitoring data of a heat-conducting gel in a water resistance test environment collected by a spectrum sensor and a thermal imaging sensor; dividing samples into multiple independent detection sub-regions, comparing the real-time monitoring data with historical database reference data to separate stable components and variable components; using pattern classification technology to identify unexpected change patterns, using relationship mining technology to find causal correlations and co-occurrence characteristics, and obtaining key influence factors; taking the key influence factors as input, combining with similar material standard water resistance performance data to construct a reference performance curve, determining a potential water resistance risk level according to the curve deviation degree and an intelligent evaluation framework; and using a probability inference model to predict water resistance changes within a preset time window and generating a detection result report.

This invention relates to the field of multimedia content retrieval, specifically a topology-aware hierarchical index construction and load balancing method for multimedia content retrieval. It comprises two core processes: index construction and querying. Index construction includes three stages: topology graph construction, METIS-based multi-level graph partitioning, and parallel HNSW sub-index construction. These steps are executed sequentially and are closely interconnected. Querying includes three stages: feature modulation based on a gating mechanism, cluster association probability inference, and probability-driven parallel search and fusion. Furthermore, a nearest-neighbor-aware soft-label supervised training strategy is proposed to train the gating neural network model in the query stage. This invention abandons the traditional partitioning paradigm that relies solely on geometric metrics. Through an innovative index construction process and query mechanism, with topology awareness and load balancing at its core, it achieves high recall, low latency, and high scalability in multimedia similarity search.

The application discloses a crack-resistant reinforcement configuration optimization method for high concrete face slab dam panels, S1: acquiring first data; S2: generating second data based on a geometric model of dam panels; S3: generating an indication signal based on the first data and the second data through a Pareto frontier solver; S4: generating a reinforcement network optimization instruction based on a selected working point on a Pareto frontier curve; and S5: adjusting an initial reinforcement topology network in response to the reinforcement network optimization instruction, and outputting a final reinforcement configuration scheme; the application can realize significant synergistic gain of design safety and economic benefits by deep quantization of structural risk through probability inference, construction of a double-target optimization framework of efficiency and robustness, generation of a Pareto frontier performance boundary, and output of a reinforcement scheme with higher safety under the same cost or lower cost under the same safety standard through an interactive decision tool.

The application discloses an automatic driving software dynamic test method based on a dynamic scene graph, relates to the field of software testing, and comprises the following steps: S1, dynamic fusion and structured modeling of multi-source heterogeneous traffic scene data: fusing vehicle-mounted laser radar point cloud data and V2X communication data of a roadside unit, and constructing a space-time synchronization matrix comprising vehicle pose, road topology, traffic semantics and environmental parameters; performing probability inference on driver behavior based on a hidden Markov model, and generating initial scene graph nodes comprising explicit states and implicit game logic; S2, real-time updating of the scene graph based on a double-layer evolution mechanism: dividing the updating process of the scene graph into a short-term evolution layer and a long-term evolution layer; and S3, closed-loop dynamic testing and semantic completeness verification: inputting the evolved scene graph into an automatic driving software simulation environment, and performing coverage testing, consistency testing and counterfactual testing.

The application discloses a fault probability monitoring and regulation system and method, relates to the technical field of fault monitoring, and collects to-be-determined related parameters in each period through a collection module, determines the optimal order of an autoregressive model of each related parameter by using the Akaike information criterion, and verifies the to-be-determined related parameters in each period, decides whether to perform secondary collection and replacement based on a verification result, generates a related parameter vector of each period, and starts a window translation along a period in a reverse order from the related parameter vector of each period through a monitoring module, calculates and arranges a comprehensive feature vector in each window, extracts and generates a core feature vector set through a time series self-encoder, trains a hidden Markov model by using a forward-backward algorithm, and obtains the fault probability of each period based on Bayesian inference, compares the fault probability with a probability threshold value to determine whether a fault exists, and executes regulation when the fault exists, so that high-precision fault monitoring and regulation based on collection self-checking, multi-dimensional data fusion and probability inference are realized.

The invention discloses an engineering geological survey report intelligent generation method and system, and the method comprises the steps: S1, obtaining multi-modal engineering geological survey data of a target site, the survey data comprising lithologic description, physical and mechanical parameters and position information of a plurality of drilling holes in spatial distribution; s2, on the basis of the multi-modal data, stratum labels corresponding to the multiple drill holes are inferred through a first model, and the first model is a machine learning model fusing a graph structure and probability inference; s3, on the basis of the stratum label and the engineering specification knowledge base, standardized cue words corresponding to the stratum label are generated through a second model, and the second model is a sequence generation model based on reinforcement learning training; compared with the prior art, through a data-knowledge-model closed-loop architecture and a continuous optimization mechanism based on feedback, end-to-end efficient automatic generation is realized, and the system has strong scene self-adaption and evolution capability and can flexibly serve various engineering investigation scenes.

The invention relates to the technical field of agricultural environment monitoring, and discloses a farmland greenhouse gas emission estimation system based on a remote sensing technology, and the system comprises a multi-source data sensing and crop and soil state parameter conversion module; a crop fertilizer demand characteristic dynamic diagnosis module based on an agronomy mechanism; the fertilization and farmland carbon management scene initial probability inference module is used for reversely inferring a discrete management scene containing carbon and nitrogen input; the emission and growth coupling iterative correction module is used for simulating a theoretical growth curve by using a biogeochemical process model, comparing the theoretical growth curve with remote sensing observation and iteratively correcting a scene probability; and a greenhouse gas emission weighted simulation and probability expression module. According to the method, the agricultural mechanism model and remote sensing observation are coupled, physical process self-consistent constraint is carried out on a farmland management scene containing carbon and nitrogen input, the accuracy and reliability of emission estimation are improved, and probabilistic expression of a result is realized.

The invention relates to the field of geological disaster recognition, in particular to an intelligent recognition method and system for geological disasters along a natural gas pipeline. The invention discloses an intelligent recognition system for geological disasters along a natural gas pipeline. The intelligent recognition system comprises a pipe section region division module, an environment feature acquisition module, a disaster probability inference module, a data source suitability inference module and a monitoring strategy generation module. According to the method, terrain consistency section division is carried out along the natural gas pipeline, a static label library is established for each pipe section, and the types of disasters prone to happening of each pipe section are dynamically judged by combining dynamic environment characteristics collected in real time and adopting a statistical inference method based on large-scale historical disaster samples; according to the mechanism, various difference factors such as terrains, climates and structures are fully considered, disaster identification according to local conditions and time conditions is realized, and the accuracy of risk identification and the regional adaptation capability are remarkably improved.

The present application relates to the technical field of concrete detection, in particular to a concrete structure crack damage evaluation system based on acoustic emission sensing, which comprises a signal acquisition module, a feature discrete evolution module, a probability inference module, an energy gradient analysis module and a boundary definition module.In the present application, a multi-dimensional evolution set is constructed by extracting the amplitude and energy standard deviation of acoustic emission signals within a continuous time window, the synchronous change trend of the multi-parameter standard deviation is analyzed using a Bayesian probability model, the non-uniformly expanded posterior probability is calculated to lock the key signal set of the dominant damage expansion, the energy fluctuation coefficient is calculated in combination with the spatial coordinates of the sensor and a spatial sequence is generated, the stable interval of the fluctuation coefficient is identified according to the gradient attenuation law of energy with distance, thereby quantitatively defining the physical boundary of the dynamically active area of crack damage, and the non-uniform expansion state and active range of the concrete crack are accurately evaluated.

The invention discloses a wind power generation operation optimization method and system, and relates to the technical field of wind generating set operation optimization, and the method comprises the steps: obtaining operation monitoring data and operation and maintenance event data of a wind generating set; establishing and updating a digital twin model, and outputting operation response data; constructing a Bayesian inference model and inferring to obtain the health state parameters of the key parts and the probability distribution of the remaining service life of the key parts; generating an operation control instruction; executing the operation control instruction, and obtaining control execution feedback data; and updating the digital twin model and the Bayesian inference model. According to the method, monitoring and operation and maintenance data are organized in an evidence mode, virtual observation with a consistent mechanism is formed, probability inference is carried out, life uncertainty is brought into target balance, a risk perception control instruction is output, model drifting is updated and restrained through a feedback closed loop, the reliability and auditing performance of life prediction are improved, and the reliability and auditing performance of life prediction are improved. Under the condition that safety and grid connection constraints are met, benefit improvement and load risk cooperative control are achieved.

A routing leakage detection method based on probability reasoning and path score, the method comprising: obtaining an autonomous system path in a border gateway protocol route, the autonomous system path comprising a plurality of autonomous system links, obtaining autonomous system business relationship probability data about the plurality of autonomous system links through an autonomous system business relationship probability inference algorithm, and obtaining autonomous system business relationship determination data about the plurality of autonomous system links through an autonomous system provider authorization object; the autonomous system business relationship probability inference algorithm is set according to the no valley rule, and the autonomous system business relationship comprises provider to customer, customer to provider, and peer to peer; the score of the autonomous system path is calculated according to the path score calculation formula based on the autonomous system business relationship probability data and the autonomous system business relationship determination data; the score of the autonomous system path is compared with a preset threshold value, and it is judged whether the border gateway protocol route of the autonomous system path is a leaked route.

The application discloses a kind of based on bayesian network's surrounding rock quality evaluation method and system, it is related to surrounding rock quality evaluation technical field, the method includes: constructing three-layer directed acyclic graph as bayesian network topology structure;Continuous evaluation index node in evaluation index parent node layer is discretized and handled, corresponding discrete state is generated, the discrete state of each evaluation index node in evaluation index parent node layer is mapped to uniform numerical score interval, and the score value of comprehensive score intermediate node is generated;The field measured data of the surrounding rock section to be evaluated is input as evidence into bayesian network, the probability inference of bayesian network is carried out by joint tree algorithm, according to the message passing mechanism between adjacent maximum groups in joint tree, the potential function of each maximum group in the joint tree is updated, the posterior probability distribution of target node in surrounding rock quality grade target node layer is calculated;The posterior probability value corresponding to each grade of surrounding rock quality is output in the form of posterior probability distribution.

The invention provides a three-dimensional model reconstruction method based on multi-modal perception, and relates to the technical field of model reconstruction, and the method comprises the steps: obtaining the multi-modal perception data of a building through a sensor; utilizing a deep learning model and an analysis method to perform feature extraction and probability inference on the multi-modal perception data to obtain an evidence source; analyzing the constructed macroscopic feature vector by using a decision tree classifier to obtain a classification result; on the basis of the classification result, the trust weight of each type of evidence source is adjusted by using a gating fusion network guided by space geometry, and a posterior probability graph is obtained; wherein the gating fusion network is trained to obtain a posterior probability graph, three-dimensional model reconstruction is carried out, and a three-dimensional model reconstruction result is obtained. According to the method, the problems of insufficient two-dimensional semantic segmentation precision, poor multi-modal information fusion generalization ability and poor adaptability to complex building scenes in the prior art are solved.