30 results about "Fuzzy mapping" patented technology

The invention discloses a surface wear volume time-space distribution prediction method combining abrasive particle and vibration monitoring, which comprises the following steps: constructing an abrasive particle size distribution driven Markov process, representing a random wear process, and realizing dynamic and probability representation of real-time wear volume time evolution; a physically constrained double-branch mapping model is constructed, fuzzy mapping of the surface wear volume and the vibration time-frequency characteristics is established, and precise observation of wear volume space distribution is achieved; the two physical mechanisms are fused on the basis of a state space modeling framework, a digital twinborn model combining abrasive particle monitoring and vibration monitoring is constructed, and accurate prediction of the abrasion volume of the friction pair part in the time dimension and the space dimension is achieved. According to the method, the wear volume space-time distribution is predicted by fusing the wear particle information and the physical correlation between the vibration characteristics and the wear process, and the precise monitoring of the wear process is realized.

The invention provides an underwater alumen ustum quantitative alarm method, device and equipment based on an AI model and a medium, and relates to the technical field of water treatment and artificial intelligence. According to the method, the collected underwater alumen ustum video image is obtained in real time, the pre-trained AI target recognition model is adopted to perform semantic recognition and segmentation on the underwater alumen ustum video image, the alumen ustum area is extracted, the white noise part is reserved, and the semantic segmentation result is obtained; calculating density features and size features of alumen ustum in the image according to the semantic segmentation result; inputting the density features and the size features into a pre-trained model based on depth nonparametric fuzzy mapping, and classifying the alumen ustum distribution states to obtain alumen ustum state classification results; comparing the alumen ustum state classification result with a preset threshold value, and triggering an alarm if the alumen ustum state classification result exceeds the threshold value. The alumen ustum state can be accurately quantified, real-time alarm and process optimization are realized, and the water treatment efficiency and the automation level are improved.

The invention discloses a visual scene graph generation method based on differentiable fuzzy logic reasoning, and solves the technical problems of semantic fuzziness, logic common sense deficiency and the like during long tail relation processing in the prior art. The method comprises the following steps: inputting a training image into a target detection module, and respectively outputting corresponding high-dimensional geometric embedding vectors to a relation classifier and a fuzzy mapping layer; outputting a visual prediction branch by the relation classifier; meanwhile, the fuzzy mapping layer outputs a fuzzy membership degree vector; the logic tensor reasoning module is combined with a common sense rule in an external knowledge base, simulates a logic reasoning process by utilizing a differentiable logic operator, and outputs a logic reasoning branch of which the relation of each pair of objects meets a preset logic rule in the training image; a gating residual fusion module performs weighted fusion on the visual prediction branch and the logical reasoning branch to generate a scene graph triple of the training image; and finally, training the network model, inputting a test image into the trained model, and outputting a corresponding visual scene graph.

The invention discloses an MOM software quality optimization method and system based on dynamic fuzzy scenario, and the method comprises the steps: collecting heterogeneous data from multiple dimensions of operation monitoring, user interaction and business process tracking, extracting a core feature subset through information measurement after cleaning normalization, and defining an operation scene in combination with the characteristics of system load and the like and core feature expression. A double-layer weight model is constructed and dynamically adjusted, fuzzy mapping of features and preset evaluation levels is established for all scenes to form a judgment matrix, the quality level is determined according to the maximum membership degree after weight and matrix synthesis, software configuration, resource allocation or workflow adjustment instructions are generated based on evaluation results to be issued and executed, and closed-loop optimization is achieved. According to the method, accurate quantitative evaluation and automatic optimization are realized.

The present invention discloses a network security situation assessment method that integrates fuzzy cluster analysis and fuzzy comprehensive evaluation and decision-making, which relates to the field of network security technology. The method includes: first, obtaining an initial data matrix, then using the translation-range transformation method to standardize it, then constructing a fuzzy similarity matrix through the similarity coefficient-quantity product method, then obtaining a fuzzy equivalence matrix based on the closure transfer characteristic, and performing dynamic clustering to construct a first-level situation factor set and a second-level situation factor set; then, based on expert advice, obtaining a single-factor fuzzy mapping set of characteristic attributes of each cluster, and using the hierarchical analysis method to obtain the corresponding weight vector; using a weighted average operator to calculate the fuzzy comprehensive evaluation matrix, and then obtaining the status of the network security situation through the maximum membership principle. Therefore, the above method can comprehensively consider the fuzzy interactions and weight distribution of each clustering factor to achieve network security situation assessment, and the assessment framework is more flexible and adaptable.

This application discloses a method, system, device, and medium for predicting underwater robot operation risks, relating to the field of risk prediction. The method inputs a risk value vector to be predicted into a basic risk prediction model to obtain a first risk level prediction value; utilizes a hierarchical analysis method to obtain corresponding weights, and determines a second risk level value based on the weights; utilizes a trapezoidal fuzzy membership function to calculate the risk membership of each risk factor value, and determines the risk level corresponding to the maximum risk membership as a discrete risk level value, which is input into a risk level fuzzy mapping model to obtain a third risk level prediction value; calculates the number of different risk levels and inputs this into a risk prediction model based on risk level statistics to output a fourth risk level prediction value; and utilizes a particle swarm algorithm to determine the weights of the four risk level prediction values, thereby determining the comprehensive risk value of the underwater robot operation. This application achieves higher-accuracy underwater robot operation risk prediction using limited data.

The application discloses a kind of building inspection method and system based on existing building structure, the method includes the following steps: based on multi-period point cloud registration calculation three-dimensional displacement, fusion principal component generates structure deformation stability index, identifies potential instability area;Through multi-scale surface fitting extraction curvature and normal vector evolution gradient, combined with signal-to-noise ratio weighting, construct local distortion sensitivity index;Gradient entropy and roughness are extracted to abnormal area, and material degradation confidence factor is generated by fusing reflection intensity variation;Based on spatial gridding calculation degradation concentration, coupled maintenance frequency and obtained weak degree score by fuzzy mapping, combined with importance weight output risk priority;Deformation rate sequence is extracted to high-risk component, trend is fitted and cumulative score is constructed, and weak score is double-target normalized, and by pareto front and euclidean distance solution, determine the highest position of comprehensive risk and label failure urgency level.The application improves the identification precision of existing building structure deformation.

The application discloses a MOM software quality optimization method and system based on a dynamic fuzzy scene, which comprises the following steps: collecting heterogeneous data in multiple dimensions from operation monitoring, user interaction and business process tracking; extracting a core feature subset through information measurement after cleaning and normalization; defining an operation scene in combination with system load and other features and core feature performance; constructing a double-layer weight model and dynamically adjusting the model; establishing a fuzzy mapping of features and preset evaluation levels for each scene to form a judgment matrix; determining the quality level according to the maximum membership degree after the weight and the matrix are combined; and generating software configuration, resource allocation or workflow adjustment instructions based on the evaluation results and executing the instructions to realize closed-loop optimization. The method realizes precise quantitative evaluation and automatic optimization.

The application discloses a weather radar wind turbine clutter recognition method and system based on fuzzy reasoning, and belongs to the technical field of fuzzy reasoning. The monitoring area of a target weather radar is divided into multiple echo observation subunits according to distance gates and scanning azimuth angles, and the reflectivity factor, radial velocity, spectral width and differential reflectivity of each echo observation subunit are continuously acquired within a preset data acquisition time sequence. By analyzing the discrete characteristics of the reflectivity factor and the radial velocity in the time sequence, echo stability characteristic quantities are constructed, echo shape consistency characteristic quantities are constructed based on the consistency deviation of the spectral width and the differential reflectivity in the neighborhood, the echo stability characteristic quantities and the echo shape consistency characteristic quantities are subjected to fuzzy mapping, and the wind turbine clutter and meteorological echoes are distinguished and recognized through a fuzzy comprehensive judgment quantity, so that the accuracy and robustness of the wind turbine clutter recognition are improved, and the weather radar data quality control under complex background conditions is suitable.

The invention belongs to the technical field of crop disease prevention and control, and particularly relates to a planting method for preventing cucumber fruit blotch germs and green mottle viruses, which comprises the following steps: acquiring soil and crop historical pathogen data, environment monitoring data and cucumber seed batch information of a planting area; querying supply chain quarantine records of the seeds, and carrying out standardization treatment on the seeds; after planting, the disease probability of each growth stage is predicted through a risk assessment algorithm, a dynamic monitoring plan is generated in combination with a fuzzy mapping algorithm, plant samples are collected, PCR detection of melon bacterial fruit blotch and RT-PCR detection of cucumber green mottle mosaic virus are carried out, and a quarantine report is issued. Based on the monitoring plan, the quarantine report and the real-time environment data, judging a comprehensive risk level, triggering a corresponding field intervention scheme, and recording an execution result to form an intervention log; after planting is finished, a prevention and control efficiency report is generated. Therefore, the problems of insufficient disease detection, lack of intervention measures and the like in the prior art are solved.

The invention discloses a target domain label generation method and device, equipment and a medium. The method comprises the steps of obtaining an initial target domain feature of target domain data, an initial label matrix of the target domain data and an initial source domain feature of source domain data; predicting an initial pseudo label of the target domain data through a random walk method, and filtering the initial pseudo label to obtain a target pseudo label; determining a first fuzzy feature matrix of the target domain data and a second fuzzy feature matrix of the source domain data through a preset fuzzy mapping rule; and calculating an update parameter according to the first fuzzy feature matrix and the second fuzzy feature matrix to obtain an optimal solution target domain feature corresponding to the initial target domain feature and an optimal solution source domain feature corresponding to the initial source domain feature, and determining a target domain label of the target domain data.

The application discloses an arc length control method, system, device and medium for electric arc additive manufacturing, and relates to the field of additive manufacturing; the control method comprises the following steps: acquiring image data of a target part; determining arc length data by using an image edge data detection method according to the image data; obtaining a fuzzy error by using fuzzy mapping according to the arc length data and set reference arc length data; determining a control quantity according to the fuzzy error; the control quantity comprises a pulse load ratio; and the deviation of the arc length data of the target part in the manufacturing process is compensated according to the control quantity, and the target part is manufactured by using a non-gas shielded welding method; and the application can realize automatic control of the arc length.

The invention provides a semantic-driven cold-chain logistics path dynamic planning method and system, and belongs to the technical field of intelligent logistics. In order to solve the problem that existing cold chain distribution is difficult to process unstructured environment information and customer flexible demands, the invention constructs a collaborative architecture of environment perception and scheduling decision, and specifically comprises the following steps: S1, based on multi-source data of cold chain logistics, defining road network environment perception and customer demand analysis rules, generating a semantic feature vector containing a road network state component and a customer demand component; s2, analyzing the semantic feature vector, and constructing a distribution constraint model by using a dynamic correction and fuzzy mapping method; s3, constructing a multi-target distribution total cost function based on the rule and the model; and S4, performing global optimization on the total cost function by adopting a parameter adaptive improved genetic algorithm, and generating an optimal cold chain distribution path and scheduling instruction. According to the method, natural language environment information can be converted into constraint conditions of a mathematical model, accurate response to a time-varying road network and refined satisfaction of customer demands are realized, and the total cost and the goods damage rate of cold chain distribution are effectively reduced.

This invention relates to a remote sensing image scene classification method based on uncertain visual-language alignment, comprising the following steps: acquiring remote sensing images as training and testing sets; using a CLIP dual encoder to map the original image and text to a shared point representation space; the uncertain encoder predicting each embedding vector along two paths; internally, the uncertain encoder performing two-level interactions on the input embedding vectors; globally modeling the image and text as two Gaussian distributions based on the mean and variance vectors, maximizing the image-text similarity that should be correctly matched in the entire batch; acquiring new remote sensing images and inputting them into the trained model, which then classifies the scenes in the remote sensing images. This application captures fuzzy mappings between modalities in the probability space, providing a more generalized representation for downstream remote sensing image scene classification tasks.

The application discloses an access control authorization method based on fuzzy mapping, which formalizes a user set into a clear object for the purpose of protecting user privacy, formalizes a role and a fuzzy role into a fuzzy attribute, and provides a clearer and more universal definition. First, a common role and a role mapping are extracted, then a fuzzy concept is constructed and a fuzzy concept lattice is constructed, according to the obtained fuzzy concept lattice, a fuzzy aggregation concept lattice is constructed, hierarchical division is carried out, and the current and the obtained role are converted into a mapping between fuzzy role sets. By introducing the concept of a fuzzy role, the protection of the user identity is realized, and the protection is reflected in the transparency of the role to the administrator. The security features of the minimum permission and the user hierarchy are met, the number of roles in the system is greatly reduced, the problem of role explosion is avoided, and the flexibility of traditional access control is increased.

The invention discloses a network security situation assessment method fusing fuzzy clustering analysis and fuzzy comprehensive judgment decision, which relates to the technical field of network security, and comprises the following steps: firstly, acquiring an initial data matrix, standardizing by using a translation-range transformation method, constructing a fuzzy similar matrix by using a similar coefficient-quantitative product method, and establishing a fuzzy clustering result; then, on the basis of closure transfer characteristics, a fuzzy equivalence matrix is obtained, dynamic clustering is carried out, and a first-level situation factor set and a second-level situation factor set are constructed; according to expert suggestions, a feature attribute single-factor fuzzy mapping set of each cluster is obtained, and a corresponding weight vector is obtained by using an analytic hierarchy process; and a fuzzy comprehensive evaluation matrix is calculated by using a weighted average operator, and then the condition of the network security situation is obtained through a maximum membership principle. Therefore, by adopting the method, the fuzzy interaction and weight distribution of each clustering factor can be comprehensively considered, the network security situation assessment is realized, and the assessment framework is more flexible and higher in adaptability.

The present application provides a blockchain-based e-commerce operation data security protection system and method, which distributes and stores the e-commerce operation data processed by homomorphic encryption through various nodes of the blockchain network; determines the trust attributes of each user's access behavior based on historical behavior data, performs fuzzy mapping on all trust attributes, and obtains the fuzzy membership of each user's access behavior, and then determines the behavioral trust of the e-commerce cloud platform in the user access process through all fuzzy memberships; determines the encrypted identity credentials of the target user based on the target user's behavior score and initial access level; performs minimum permission constraints on the target user's access rights through the analysis results of the encrypted identity credentials and the behavioral trust through smart contracts, obtains encrypted access rights, and limits the target user's authorized scope of access to encrypted data based on the encrypted access rights. By adopting the solution of the present application, minimum permission constraints on the authorized scope of access to e-commerce operation data can be achieved.

The present application belongs to the technical field of industrial machine vision, and particularly relates to an industrial workpiece surface defect detection method and system based on a TSK fuzzy mapping. The present application aims at the problems of complex noise in industrial field environment, poor robustness of defect detection model and lack of explainability, and constructs a dual-view detection framework. Among them, the TSK fuzzy module maps image features into defect rules with clear physical semantics, such as strip texture, point pit, etc.; in the training stage, the cross-view gradient consistency constraint is introduced as an anti-noise self-calibration mechanism, which forces the model to keep the optimization direction consistent under noise interference. Experiments show that the present application has an accuracy of 93.84% in the task of hot-rolled steel strip surface defect detection in the environment simulating sensor noise interference, effectively improving the reliability and traceability of industrial quality inspection.

The invention discloses a weather radar wind turbine clutter identification method and system based on fuzzy reasoning, and belongs to the technical field of fuzzy reasoning. A monitoring area of a target weather radar is divided into a plurality of echo observation subunits according to a range gate and a scanning azimuth angle, and a reflectivity factor, a radial speed, a spectral width and a differential reflectivity of each echo observation subunit are continuously acquired in a preset data acquisition time sequence. By analyzing discrete characteristics of a reflectivity factor and a radial velocity on a time sequence, an echo stable characteristic quantity is constructed, an echo form consistent characteristic quantity is constructed based on consistency deviation of a spectral width and a differential reflectivity in a neighborhood, fuzzy mapping is performed on the echo stable characteristic quantity and the echo form consistent characteristic quantity, and the echo stable characteristic quantity and the echo form consistent characteristic quantity are calculated. The wind turbine clutter and the meteorological echo are distinguished and recognized through the fuzzy comprehensive judgment quantity, the accuracy and robustness of wind turbine clutter recognition are improved, and the method is suitable for weather radar data quality control under the complex background condition.

The invention relates to a load balancing and loss optimizing system and method for a dry-type transformer. The system comprises an attribute acquisition and storage module, a spatial correlation graph construction module, a mapping rule module, an evaluation module, a reverse fuzzy control module, a dynamic adjustment module and a closed loop verification module. The method comprises the steps of collecting static and dynamic attributes of a transformer and a load to establish a database; constructing a weighted undirected graph for quantizing the spatial association relationship; determining a fixed mapping and fuzzy mapping relationship between the load and the transformer based on a preset rule; continuously evaluating the overload state of the transformer and the system loss; when overload is detected, calculating a power distribution coefficient and a membership degree adjustment step length through reverse fuzzy control, and dynamically adjusting load distribution; and the adjustment effect is optimized through closed-loop verification. According to the invention, the coordination of load balancing and loss optimization is realized through differential mapping and accurate control, and the problems that the traditional regulation and control mode is large in sweep range and easy to cause power supply impact are effectively solved.

The present application relates to the technical field of flaw testing, in particular to a tension clamp forging flaw testing system and method based on optical means, which comprises a vibration detection unit, a fuzzy mapping unit and a segmentation testing unit. The conversion mapping module of the present application performs gray level histogram statistics on the de-fuzzed tension clamp image, calculates the cumulative distribution value of the gray level, linearly maps the cumulative distribution value of the gray level to the target gray level range, generates a new gray level after rounding, realizes image equalization through pixel gray level replacement, enhances the image through image equalization, makes the detailed features on the surface of the tension clamp more obvious in the equalized image, reduces the covering of the detailed features on the surface of the tension clamp by the rough background in the de-fuzzed tension clamp image, increases the difference between the gray level of the detailed features on the surface of the tension clamp and the surrounding rough background area, and uses the gray level to distinguish the detailed features and the rough background to increase the accuracy of the flaw testing.

The application provides a kind of intelligent welding current adaptive regulation method and system, belongs to industrial control system technical field.The method realizes the online accurate identification and classification of hidden welding defects such as internal incomplete penetration of weld, porosity, etc.by adopting deep convolutional neural network;At the same time, the current waveform data corresponding to the time interval of the welding defect is accurately matched by timestamp, the inverse mapping model of historical defect characteristics and current parameters is established and the current deviation prediction value is output, and the optimal current compensation and waveform correction function are iterated out by gradient descent optimization algorithm, which solves the problem of fuzzy mapping of defect space characteristics and current time parameters and lack of targetedness of parameter adjustment;Finally, the accurate current regulation instruction of the un-welded area is generated, and the adaptive adjustment of welding current according to different welding defect types is realized, so as to improve the welding quality.

This invention discloses a method, device, and storage medium for multi-parameter fusion sludge discharge control in sedimentation tanks of steel plants, within the field of wastewater treatment technology for the steel industry. The method includes: acquiring real-time multi-parameter data; calculating pre-calibrated multi-parameter data based on the real-time multi-parameter data and theoretically predicted values ​​of pre-acquired multi-parameter data; performing fuzzy mapping on the pre-calibrated multi-parameter data to obtain the membership degree of each parameter; constructing a fuzzy output set from all triggered rules in a pre-defined fuzzy rule set based on the membership degree of each parameter; and defuzzifying the fuzzy output set to obtain the sludge discharge pump frequency conversion command. This invention can alleviate the technical problems affecting sludge discharge efficiency caused by excessive or insufficient sludge discharge leading to water quality exceeding standards, interference of traditional sensors by ferromagnetic sludge, response lag due to manual reliance, and wasted water and electricity costs in existing technologies.

This invention relates to the field of page-turning clock technology, and more particularly to a page-turning clock flipping drive method and system supporting voice interaction control. The method collects historical operating data of the drive unit, extracts voice parsing response data to determine the response deviation factor of command parsing, and quantifies the uncertainty of voice interaction; acquires historical pose deviation data, uses error membership functions for fuzzy mapping, calculates the deviation membership degree, and determines the correction cost of flexible closed-loop feedback; integrates the feedback correction cost and the response deviation factor to dynamically calculate the drive compensation coefficient; and dynamically adjusts the output torque of the page-turning drive module in real time based on this coefficient, achieving end-to-end adaptive optimization from command reception to action execution. This effectively solves the problems of inaccurate command response, stiff drive process, and high noise in traditional page-turning clocks under complex sound fields, significantly improving the smoothness, accuracy, and robustness of the page-turning action.

This invention discloses an intelligent closed-loop rehabilitation method and system based on multi-source information fusion, fatigue evolution prediction, and adaptive regulation. It collects inertial motion signals and electromyographic signals through multimodal wearable sensors. After signal preprocessing and time synchronization, a multidimensional correlation-sensory attention coding mechanism and a gated recurrent network are used to perform feature encoding and future signal prediction on the multidimensional time-series signals. Based on the predicted signals, a multidimensional fatigue index system is constructed to achieve continuous, feedforward assessment of fatigue states during training. Furthermore, a fuzzy mapping method is used to map the multidimensional fatigue states into four risk levels, and training parameters are dynamically adjusted according to the risk level to form real-time, intelligent closed-loop training control. This invention can improve the scientific, precise, and individualized level of rehabilitation training while ensuring training safety, and is particularly suitable for rehabilitation training of patients with neurological diseases and those with impaired motor function.

The invention discloses a closed-loop control image deblurring method for guiding inverse process solution through a blurring mechanism, and relates to the field of deep learning and image deblurring calculation. The method aims at solving the problem that an existing algorithm depends on pairwise fuzzy-clear image pairs and is difficult to adapt to real image deblurring. The method comprises a mechanism prior guidance module, a deep coding and decoding network, a fuzzy mapping space converter, a fuzzy kernel extractor and a loss cost calculation module. A blurring kernel and a clear image initial value are estimated through a blurring generation principle, a blurred image is reconstructed through a deep coding and decoding network, double-closed-loop cross learning is realized by using a loss cost result, and regularization constraint is introduced to improve optimization stability. According to the method, a large amount of training data is not needed, high-quality deblurring can be achieved in a few-sample scene through closed-loop optimization and a frequency domain information retention mechanism, the method is particularly suitable for high-speed motion blurred scenes such as vehicle-mounted image collection, and the accuracy of subsequent image recognition and analysis is effectively improved.

This invention discloses a method and apparatus for predicting the state of underwater dynamic targets based on T-S fuzzy logic. The method includes: first, acquiring the input variables of the underwater dynamic target at time t-1 for T-S fuzzy inference; then, using the fused underwater dynamic target motion model represented by equation (6) at time t-1 to perform extended Kalman filtering to predict the motion state of the underwater dynamic target at time t; then, taking t = t+1 until the tracking ends. Where X... t X t‑1 Let Y and Y represent the state vectors of the underwater dynamic target at times t and t-1, respectively, and u represent the process noise vector of the input variable. t A represents the output vector of the motion model at time t. mix B mix These are the state transition matrix and the input matrix, respectively, and C represents the pre-set output vector Y. t With state vector X t The relationship matrix. This invention provides more accurate prediction of the state of underwater dynamic targets and can meet the requirements of underwater dynamic target tracking systems.

The invention relates to the technical field of flaw testing, in particular to a strain clamp forging flaw testing system and method based on an optical means. The system comprises a vibration detection unit, a fuzzy mapping unit and a segmentation test unit. The conversion mapping module carries out grey level histogram statistics on the deblurred strain clamp image, then calculates the cumulative distribution value of the grey level values, linearly maps the cumulative distribution value of the grey level values to a target grey level range, generates a new grey level value after rounding off, and realizes image equalization through pixel grey level replacement. Through image enhancement of image equalization, detail features of the surface of the strain clamp are more obvious in the equalized image, the situation that the detail features of the surface of the strain clamp are covered by a rough background in the deblurred strain clamp image is reduced, and the difference between the gray value of the detail features of the surface of the strain clamp and a surrounding rough background area is increased. And the accuracy of flaw testing is improved by using the gray level for distinguishing detail features and rough backgrounds.

The application discloses a sliding input method, device and equipment and a readable storage medium. First, a sliding track of a user on a virtual keyboard is collected; then, feature points and passing points are determined, the feature points are sliding points capable of dividing the sliding track into multiple track segments, and the center points of keys other than the keys containing the feature points in the keys corresponding to each sliding track segment are the passing points; based on the coordinate keys and error correction keys corresponding to the feature points and the coordinate keys corresponding to the passing points, a pinyin link corresponding to the sliding track is determined; finally, based on the error correction keys, the pinyin link corresponding to the sliding track is decoded to obtain a candidate result. In the scheme, the pinyin link space can be increased through the key fuzzy mapping of the feature points, and thus the correct pinyin link corresponding to wrong sliding, multiple sliding and less sliding can be found, so that the decoding effect of the sliding track is improved.

The application discloses a visual scene graph generation method based on differentiable fuzzy logic reasoning, and solves the technical problems of semantic ambiguity and lack of logical common sense in the prior art when processing long-tail relations. The application inputs a training image into a target detection module, and respectively outputs corresponding high-dimensional geometric embedding vectors to a relation classifier and a fuzzy mapping layer; the relation classifier outputs a visual prediction branch; at the same time, the fuzzy mapping layer outputs a fuzzy membership vector; a logical tensor reasoning module combines common sense rules in an external knowledge base, simulates a logical reasoning process by using differentiable logical operators, and outputs a logical reasoning branch of each pair of object relations in the training image satisfying a preset logical rule; a gating residual fusion module performs weighted fusion on the visual prediction branch and the logical reasoning branch to generate a scene graph triple of the training image; and finally, the network model is trained, a test image is input into the trained model, and a corresponding visual scene graph is output.