217 results about "Model complexity" patented technology

The invention discloses a virtual space multi-person interaction synchronous control method oriented to an end-cloud collaborative architecture. The invention relates to a computer graphics and three-dimensional modeling technology, and discloses a cross-format lightweight and geometric consistency maintenance method based on a three-dimensional model. Through format-independent geometric representation and a self-adaptive lightweight strategy, efficient compression and precision maintenance of three-dimensional model cross-format conversion are realized. The method specifically comprises the following steps: performing format analysis and geometric feature extraction on an input model, and establishing a unified internal representation; adaptively selecting a multi-level LOD lightweight strategy based on the complexity of the model; the accuracy of key information is ensured through geometric feature keeping and topology consistency detection; the geometric consistency is dynamically maintained by combining error monitoring and an iterative correction mechanism; and generating a target format lightweight model and carrying out quality verification. According to the method, adaptive precision control, multi-level consistency maintenance and format irrelevant processing are combined, the model size and conversion errors are effectively reduced, and cross-platform compatibility and geometric fidelity are improved. The method can be widely applied to the fields of industrial design, game development, virtual reality and the like.

The invention discloses an apple leaf disease segmentation method based on a lightweight dual-path network, an apple leaf disease segmentation device based on the lightweight dual-path network, an apple leaf disease segmentation device and a computer readable storage medium. The problems that an existing disease segmentation method is high in model complexity and insufficient in multi-scale recognition capability are effectively solved. The lightweight encoder adopts a depth separable convolution and channel recombination technology, so that the parameter quantity and the calculation complexity are greatly reduced while the feature extraction capability is maintained, and the model can be deployed on edge equipment such as an unmanned aerial vehicle and a field robot. The enhanced cavity space pyramid pooling module constructs abundant multi-scale receptive fields through multi-branch parallel cavity convolution with different expansion rates, and can capture feature information of initial tiny disease spots and later fused disease spots at the same time.

The invention discloses a light-weight small target detection method and system for images shot by an unmanned aerial vehicle, and the method specifically comprises the steps: constructing a neural network architecture which comprises a backbone network, a feature aggregation network and a detection head; improvement of light weight and attention enhancement is implemented in the backbone network, and a multi-scale initial feature map is extracted; constructing a feature aggregation network Neck, performing cross-level fusion and refining processing on the multi-scale initial feature map, and outputting a refined feature map; a lightweight target detection head Head is constructed in combination with a large-kernel depth separable convolution module and a special decoupling head structure of a YOLOv11 network, and decoupling prediction is performed on the refined feature map; training is carried out by adopting a mixed loss function based on a normalized Wasserstein distance and modulated IoU, a trained lightweight network is obtained, and detection of a lightweight small target is realized. According to the invention, the complexity of the model is reduced, the detection speed is improved, the high detection precision is maintained, and the method is suitable for real-time detection tasks on an unmanned aerial vehicle resource limited platform.

The invention provides an inland ship target detection method and system, and relates to the technical field of ship image recognition. The method comprises the following steps: establishing an inland ship target data set; performing data enhancement processing on the inland ship target data set; a BiFPN module is adopted to replace a PANet module in the YOLOv8 model; a SimAm attention module is introduced into the neck network; the detection head is improved by adopting RepConv convolution; the Shape-IoU is introduced to replace the CIoU to serve as a bounding box loss calculation function of the YOLOv8 model; and after the improved YOLOv8 model is obtained, an inland ship target detection model is obtained after inland ship target data set training, and target detection of the to-be-detected image is realized. The method can effectively balance the detection precision and the model complexity under the condition of ensuring the real-time performance, and has obvious advantages in small target detection and complex environments.

The invention discloses an embedded AI intelligent computing power architecture method, and particularly relates to the technical field of artificial intelligence processing architecture. Collecting a resource state parameter set R of the embedded device; obtaining a to-be-executed AI task set T, wherein each task comprises model complexity, real-time requirements, expected response duration and priority; constructing a computing power resource allocation evaluation function F based on R and T, and outputting a task scheduling priority score; allocating tasks to the embedded AI module according to the allocation scheme and executing reasoning; the resource state is dynamically monitored in the task running process, and if it is predicted that resources are about to be overloaded, a scheduling function F is triggered to reconstruct a resource allocation scheme; performing iterative optimization on weight parameters in the function F based on task history feedback; by means of the method and device, optimal adaptation of multi-task concurrent scheduling can be achieved under the condition that resources are limited, the computing power resource utilization rate, the response efficiency and the system stability are improved, and the method and device are suitable for various edge side AI scenes.

The invention belongs to the technical field of uncertainty probability characterization analysis, and discloses a thermoplastic composite material leading edge structure design allowable value probability characterization method and system under a small sample condition, and the method comprises the steps: defining a plurality of candidate probability distribution models; fitting each model based on the original sample data and calculating an AIC value and a BIC value; a dynamic weight factor alpha is calculated according to the sample size n, and then a hybrid information criterion HIC value is calculated; generating a plurality of sample sets through Bootstrap self-service sampling, recalculating the HIC value on each sample set, and counting the selected optimal frequency of each model; and determining an optimal probability distribution model according to the frequency, wherein the optimal probability distribution model is used for representing a design allowable value. According to the method, the dynamic weight factor alpha is introduced, AIC and BIC criteria are effectively unified, optimal balance between prediction precision and model complexity is achieved under the condition of small samples, and engineering practicability and robustness are remarkably improved.

The invention discloses a ship water gauge scale line fitting algorithm, and relates to the technical field of ship water gauge scale calculation, and the method comprises the steps: obtaining a ship water gauge region image through an image collection device to extract a measurement point coordinate, and building a mapping relation between a pixel position and a physical draft; a weighted least square method is adopted to perform nonlinear curve fitting, a weight coefficient is calculated based on a local neighborhood standard deviation, a polynomial order p is dynamically adjusted according to an adjusted decision coefficient, and the decision coefficient is iteratively optimized to improve the balance between amplitude and model complexity. And the system repeatedly corrects model parameters through a dynamic verification mechanism of the residual sum of squares and a preset threshold value, and finally an accurate water gauge scale fitting curve is generated. The method effectively solves the problems that traditional manual observation is prone to environmental interference and large in measurement error, high robustness is kept in a complex water area environment, and reliable data support can be provided for the fields of ship load evaluation, channel safety management and the like.

The invention discloses a multi-variable measurement sensor state lightweight evaluation method based on knowledge distillation, and belongs to the technical field of electric digital data processing and multi-sensor data fusion. The method comprises the following steps: firstly, carrying out time synchronization and physical consistency constraint modeling on original data of multiple sensors, and extracting feature representation; a high-precision teacher model is trained at the cloud end, and the high-dimensional mapping relation of the sensor state is learned; intermediate features, soft output and uncertainty information of a teacher model are extracted to serve as distillation knowledge, lightweight student model training is guided, and effective migration of discrimination knowledge is achieved in combination with soft label constraint, feature alignment and an uncertainty guiding mechanism; and finally, compressing, quantifying and optimizing the student model, and deploying the student model to a vehicle-mounted end to realize real-time evaluation and dynamic updating of the states of multiple sensors. According to the method, the model complexity is greatly reduced while the evaluation precision is ensured through a knowledge distillation framework, and efficient and reliable state perception and fault-tolerant control support is provided for an intelligent driving system.

The invention belongs to the technical field of wind power prediction, and provides a double-layer cascade wind power prediction method and system based on TransformerEncoder, and the method comprises the steps: a fusion variable selection module comprises a Pearson correlation coefficient and an XGBoost regression model; the data processing module preprocesses the input original data to obtain normalized data; a fusion variable selection module screens normalized data, a Pearson correlation coefficient quantifies a linear relation of the normalized data, an XGBoost regression model captures a nonlinear relation, and a correlation feature sequence is obtained through fusion; and the model construction and prediction module processes the related feature sequence, predicts and optimizes a power value, and outputs an optimized power prediction value. According to the method, efficient screening of key variables is realized, change rules of meteorological data, fan speed data and power data are effectively identified, redundant variables are effectively reduced, the complexity of the model is reduced, dynamic characteristics of a wind power system can be better understood and learned, and higher precision and stability are shown in an actual prediction task.

The invention relates to a multi-modal data fusion 3C assembly process quality accurate prediction method, and belongs to the field of 3C intelligent manufacturing, and the method comprises the steps: carrying out the multi-modal data fusion, and forming a unified feature space containing process parameters and quality characterization; performing multi-dimensional analysis on process quality, performing hierarchical analysis on structured process parameters and unstructured images, and establishing cross-modal feature mapping; predicting the process quality, evaluating the contribution degree of process parameters to the quality by utilizing an integrated learning model based on a fused feature set, screening key parameters to reduce the complexity of the model, dynamically adjusting the prediction model, designing a'sliding window + multi-dimensional index 'monitoring system, and evaluating the model performance and the data distribution drift in real time. Establishing a time-performance-distribution three-dimensional trigger mechanism, and starting an adjustment strategy according to different priorities; a'cleaning-complementation-mechanism-mapping-self-adaption 'closed loop is realized, and the timeliness and accuracy of 3C assembly quality prediction are remarkably improved.

The invention belongs to the technical field of remote sensing image processing, computer vision and deep learning, and particularly relates to a remote sensing image erosion gully semantic segmentation method based on improved OfficientNet-UNet. Comprising the following steps of 1, data preparation and data preprocessing; 2, constructing and enhancing a data set; step 3, model construction and strategy training; and 4, performing contrast experiment and result evaluation. According to the method, detail features of ground features can be more accurately captured, an overfitting phenomenon caused by too high model complexity is reduced, so that the classification precision and boundary recognition accuracy of land coverage data are effectively improved, weights of different types of samples can be automatically adjusted in the training process, and the training efficiency is improved. Particularly, the contribution of background pixels to a loss function is reduced, so that the problem of dominant training of the background pixels is effectively relieved; the method has good expansibility.

The invention provides a power grid icing image detection method, and aims to improve the identification efficiency and precision of the icing condition of a power transmission line in the intelligent inspection of a power system. The method comprises the following steps: step 1, preprocessing and standardizing an input power grid transmission line image; 2, performing multi-scale feature extraction by using a backbone network improved based on YOLOv8 to obtain a first scale feature map, a second scale feature map and a third scale feature map; 3, performing feature enhancement on the third-scale feature map in combination with a partial self-attention mechanism; step 4, realizing fusion of different scale features through an inverted residual moving module and up-sampling operation; and step 5, inputting the fused feature map into a detection head to realize accurate detection of position coordinates and category information of the power grid transmission line image. According to the method, the identification precision of the ice-coated line is effectively improved, meanwhile, the model complexity is remarkably reduced, and the method is suitable for deployment of an embedded platform or edge equipment and has a good engineering application prospect.

The invention discloses a machine abnormal sound detection method based on enhanced self-coding reconstruction and probability modeling, and aims to solve the problem that in an existing unsupervised machine abnormal sound detection method based on a generative adversarial network, the time-frequency feature reconstruction result of machine operation sound is too smooth, so that the machine abnormal sound detection capability is insufficient. According to the method, time-frequency feature extraction is carried out on collected machine operation sound signals, modeling is carried out on machine operation sound time-frequency feature distribution under a normal working condition in combination with a probability statistical model, and the time-frequency features of the machine operation sound under the normal working condition are reconstructed and learned by using an auto-encoder of a structure enhancement mechanism. The method realizes unsupervised machine abnormal working condition determination by combining the reconstruction error and the probability determination result, can improve the determination stability and robustness of abnormal sound detection, reduces the model complexity, and is suitable for application scenarios such as industrial equipment operation state monitoring and fault early warning.

The invention discloses a substation panoramic intelligent monitoring system and a substation panoramic intelligent monitoring method, and relates to the technical field of power system monitoring, a sensing layer collects an equipment state, a topology state and digital twin simulation data, and an edge calculation layer performs secondary weight reduction on a lightweight anomaly detection model based on the data, namely, adapts to substation topology / equipment difference, and also performs secondary weight reduction on the lightweight anomaly detection model based on the data. The model complexity is reduced, real-time anomaly monitoring of an edge end is realized, and a short-term result is output; the cloud computing layer depends on AI and a digital twinborn model and fuses multi-source data to perform state prediction to generate a long-term result, and the execution control layer formulates a decision scheme according to the result. According to the scheme, the limitation of an existing fixed framework is broken through, the problems of adaptability and edge real-time response are solved, and meanwhile operation and maintenance mode transformation of the transformer substation from passive fault response to active risk prevention is supported.

The invention discloses an overhead line system dropper defect identification method based on lightweight YOLOv12s. The method comprises the following steps: acquiring an overhead line system dropper image to be identified; and inputting a to-be-identified catenary dropper image into the pre-trained YOLOv12s-Lite, and outputting a catenary dropper defect detection result. The YOLOv12s-Lite is a lightweight network obtained by carrying out lightweight improvement on the original YOLOv12s; the lightweight improvement comprises the following steps of: replacing a backbone network in the original YOLOv12s with the OfficientNetV2s; an efficient multi-scale attention module EMAttention is introduced into a neck network in the original YOLOv12s. According to the method, the model complexity is remarkably reduced while the model defect identification precision based on the YOLOv12s algorithm is maintained, lightweight detection of catenary dropper defect identification is realized, and technical reference is provided for edge end model deployment.

The invention provides an APT attack detection method, device and system based on a lightweight deep learning framework. The method comprises the following steps that: each node constructs a local sample data set based on local network flow data to train a local classification model, and uploads trained model parameters to a federated server; the federal server aggregates the model parameters of all the nodes and optimizes global model parameters, so as to issue the optimized global model parameters to each node; performing parameter updating on the local classification model by each node based on the received global model parameter, and repeatedly executing the model training step on the updated local classification model until the model converges; and each node classifies the to-be-detected traffic data based on the converged global model parameters to obtain an APT attack detection result. According to the technical scheme provided by the invention, the complexity and communication load of the model are effectively reduced while the detection precision is ensured, and the practicability and stability of the model in a distributed and resource-constrained environment are improved.

The present disclosure provides techniques and solutions for benchmarking process models by evaluating characteristics of the model, such as those reflecting model complexity. Metrics can include the number of elements in a model, the number of roles, and the number of handoffs between roles, as a few examples. Metrics for a model can be compared with reference metrics, such as those calculated from a set of other models, which can be for the same modeled process or different processes. Collections of process models can be evaluated in a similar manner, including for a set of related models that may be expressed at different levels of specificity. Metrics for individual models in the collection can be evaluated and aggregated, and then compared with aggregated metric values of other model collections, for the same or different modeled processes.

The invention discloses a lightweight model deployment method of an AI edge computing terminal and a low-power-consumption processing device, and relates to the technical field of artificial intelligence and edge computing, and the method comprises the steps: collecting the total energy consumption of the terminal and the power consumption data of a processor, a memory and a communication module in real time; by acquiring the energy consumption data of the terminal and the hardware in real time and establishing dynamic association of energy consumption and model adjustment, accurate matching of lightweight model deployment and an energy consumption state is realized, the problem of disjunction of model deployment and power consumption control is solved, sub-models are dynamically switched or calculation precision is reduced when the energy consumption exceeds a threshold value, and the energy consumption is reduced. When the energy consumption is too high, the complexity of the model is reduced in time, the balance between the energy consumption and the reasoning performance is realized, the problem that the energy consumption continuously rises or the performance is suddenly reduced is avoided, the cooperation of task execution and energy consumption management is realized by adjusting the task execution frequency and priority and combining closed-loop feedback continuous optimization, the reasoning performance is ensured, meanwhile, the consumption is accurately controlled, and the efficiency is improved. And the stability and the practicability in a complex environment are improved.

The invention discloses an intention recognition method and device based on a large model, a medium, equipment and a product. The method comprises the following steps: receiving input information of a user; acquiring background information related to the input information; according to the input information, a target model is determined from preset intention recognition models, the preset intention recognition models comprise a first model and a second model, the first model is a large model, and the model complexity of the second model is lower than that of the first model; performing intention recognition on the input information and the background information by using the target model to obtain a recognition result; determining whether the information of the identification result is complete or not; if it is determined that the recognition result information is complete, target intention information is determined according to the recognition result, and the target intention information is used for triggering generation of reply information corresponding to the input information.

The present disclosure relates to a computer-implemented method and system for self-adaptive recurrent machine-learning processing of a tabular dataset under a multi-task objective. In one or more implementations, the multi-task objective includes a decision task and a concept-based explanation task with dynamic allocation by an orchestrator of one or more CPUs and reallocation, by the orchestrator, of one or more GPUs as a function of real-time resource availability, and model complexity and estimated dataset size of recurrent method steps.

The application belongs to the technical field of computer network and artificial intelligence, and specifically provides a dynamic network adaptive routing method based on a graph neural network and a group relative strategy, which realizes adaptive perception of network topologies of different scales and types through a graph neural network encoder, eliminates the dependence on a value network through a group relative strategy optimization algorithm, thereby reducing the model complexity, and realizes cross-topology generalization, energy perception and load balancing on the basis of adaptive network dynamic change and effective avoidance of harmful nodes; compared with traditional shortest path routing, Q-Routing and PPO-MLP baseline methods, the application has significant advantages in key indicators such as average end-to-end delay, average queuing delay and data packet delivery rate, and has faster training convergence speed and higher cumulative reward promotion; in conclusion, the application has excellent routing performance and anti-destroying ability, and is especially suitable for routing scenes in a high dynamic and multi-fault environment.

The application belongs to the technical field of power electronic equipment and converter control, and specifically discloses a network-constructed converter control method and system for improving small disturbance stability. The application constructs an adaptive threshold judgment mechanism based on a deviation index: the maximum deviation index of the system in the stable parameter domain is determined according to various typical active control strategies, and the adaptive threshold is set by taking the intermediate value; in actual control, the maximum deviation index corresponding to the current control strategy is compared with the threshold, and the converter model considering or ignoring the reactive dynamic coupling is adaptively selected for controller parameter setting. Since the mechanism can dynamically evaluate the influence degree of the reactive dynamic coupling on the stability, the small disturbance stability analysis accuracy is ensured, the model complexity and the calculation efficiency are effectively balanced, the stability misjudgment caused by excessive simplification of the model is avoided, and the system stability margin misjudgment problem caused by ignoring the reactive dynamic coupling in the prior art is solved.

The invention provides a tunnel blasting equivalent load prediction method and system based on an artificial neural network for tunnel blasting dynamic response rapid evaluation. The method comprises the following steps: firstly, establishing engineering parameters for describing a single-hole blasting working condition of a tunnel, and converting the engineering parameters into dimensionless input parameter vectors; aiming at continuity parameters and grading parameters, generating sample working conditions by adopting an orthogonal test and Latin hypercube combined sampling mode, carrying out numerical simulation on each working condition, extracting a blasting triangular wave speed-time history curve of a representative measuring point in the model, and establishing a blasting triangular wave speed-time history curve with engineering parameters of a single-hole blasting working condition as input; the database takes blasting triangular wave parameters as output; an artificial neural network is trained under the loss function, rapid prediction of triangular wave parameters is achieved, and triangular waves can serve as blasting equivalent load time history to be applied to the normal direction of the plane where representative measuring points are located and used for tunnel blasting dynamic response calculation. According to the method, the calculation cost and the modeling complexity can be remarkably reduced, and the numerical calculation stability and the engineering applicability are improved.

The invention discloses an edge electric vehicle fire recognition system based on a constrained evolution algorithm, and belongs to the technical field of fire image recognition and public safety. The system comprises an image sample acquisition module, a fire image mask generation module, an image data enhancement module, a model optimization module, a fire image recognition module and an autonomous alarm module. The core innovation lies in that a model optimization module constructs model identification precision, inference delay of heterogeneous edge hardware and model complexity into a constrained multi-objective optimization problem, and a Pareto optimal solution set is solved through an NSGA-II algorithm to realize collaborative optimization of identification precision and inference efficiency; meanwhile, training data containing'hard negative samples' are generated through an innovative image data enhancement strategy, and the problem of feature confusion is solved. Different edge hardware computing power can be adaptively matched, the false alarm and missing alarm rate is remarkably reduced, the early fire recognition capability is improved, and the method is suitable for edge scenes such as public charging facilities.

The invention relates to the technical field of power line point cloud segmentation, and discloses a power line three-dimensional point cloud segmentation method, system and device and a storage medium, and the method comprises the steps: obtaining original point cloud data of a power line, inputting the original point cloud data into a multi-scale spatial feature extraction model, and obtaining spatial features; according to the spatial features, performing foreground probability prediction on the original point cloud data to obtain a foreground point set, performing multi-radius sphere neighborhood aggregation on the foreground point set to obtain local features, and performing super-point aggregation on the foreground point set to obtain global features; fusing the local features and the global features according to a feature fusion model to obtain a fused query vector; and performing classification probability prediction on the fusion query vector to obtain a point cloud segmentation result of the power line. According to the method, on the basis of keeping low calculation cost and low model complexity, the semantic recognition capability of the sparse structure target of the power line can be effectively improved, and a high-precision and high-efficiency semantic segmentation effect is achieved.

The invention discloses an engine elastic support residual life prediction method based on PCA and a BP neural network, and belongs to the technical field of vehicle part life prediction. The method comprises the following steps: firstly, acquiring multi-dimensional signals such as vibration acceleration and displacement of the elastic support of the engine through a full-life-cycle test, extracting characteristic parameters such as an effective value, inherent frequency and vibration isolation rate, and constructing an original characteristic matrix; secondly, standardizing the characteristic parameters by adopting a Z-score method, performing dimensionality reduction on standardized data by utilizing PCA, extracting key principal components, and reducing data redundancy and noise; then, taking the dimensionality-reduced features as the input of a BP neural network, constructing a multilayer neural network structure, optimizing network parameters through training, and adopting an early stop method and L2 regularization to prevent overfitting; and finally, realizing high-precision prediction of the residual life by utilizing the trained model. According to the method, PCA and BP neural networks are fused, the problems of data redundancy, model complexity and low prediction precision in a traditional method are effectively solved, and the method has the advantages of being high in generalization ability, high in automation degree and wide in applicability.

The application provides a bearing multi-scale lightweight RUL prediction method and device based on a dynamic sparse space-time graph, and belongs to the field of residual life prediction.The method comprises the following steps: adopting an adaptive multi-scale identifier to extract the potential periodicity of a time sequence, and creating a multi-scale representation robust to noise; using a ProbSparse attention mechanism to construct a dynamic and sparse connection ST graph for each scale, and updating the weight of an edge by using a decay matrix, and realizing multi-hop feature propagation by combining a Mixhop GCN, which overcomes the limitation of isolated modeling of traditional time networks and graph networks, and realizes integration of high-precision ST features; a lightweight method is proposed, which adopts two pruning types, one for full connection pruning and the other for hierarchical propagation pruning; the method solves the problems of ignoring key cross-time sensor correlation, the influence of noise on prediction and model complexity, and improves the accuracy and efficiency of RUL prediction.

This invention provides a radar motion parameter estimation method and apparatus based on information criterion values. It acquires raw radar echo data, performs range processing and target extraction to obtain a target slow-time complex sequence. Based on the target slow-time complex sequence, it constructs multi-order motion candidate models, calculates the residual sum of squares, and calculates the information criterion value based on the residual sum of squares. The optimal model order is determined based on the information criterion value, and the target radial motion parameters are inverted and output based on the phase coefficients corresponding to the optimal model order. Instead of pre-fixing the use of a first-order, second-order, or third-order model, it determines the optimal model order after constructing multi-order candidate models and evaluating them using the calculated information criterion value. This approach adaptively determines the appropriate motion model order for the target based on the phase evolution characteristics of the signal itself, and completes motion parameter estimation accordingly. This ensures estimation accuracy while suppressing unnecessary model complexity, thus improving its engineering application value.

The invention discloses a power distribution network tower and equipment parameterized three-dimensional model construction method and system, and the method comprises the steps: S1, analyzing a tower and equipment, and recognizing and extracting basic geometric elements and key features of the tower and equipment; s2, quantifying shape parameters and position parameters of a tower and equipment based on the geometric elements and the key features extracted in S1, and constructing a standardized parameterized model; s3, adjusting and optimizing the standardized parameterized model constructed in S2, and reducing the complexity of the model and ensuring the accuracy of the structure through three-dimensional parameterized design; s4, verifying the parameterized model optimized in the step S3 by using actual inspection data, and correcting the deviation of geometric and physical characteristics; and S5, extracting texture and color features in combination with the visible light image, and optimizing the parameterized model verified in the step S4. The method has the advantages of high model construction precision and the like.

To address the problems of semantic structure destruction, geometric feature distortion, and difficulty in controlling multi-level errors in the LOD simplification process of existing 3D models, this application discloses a static LOD processing method, device, and storage medium for urban digital twin scenes, belonging to the field of computer graphics and 3D modeling technology. The method includes: semantic recognition and region division of the urban building 3D mesh model to construct a semantically labeled mesh; establishing a simplification control model based on the semantically labeled mesh and determining simplification constraints; performing mesh simplification processing under constraints; preserving geometric features of the simplification results; evaluating errors in each LOD level model and performing adaptive correction; and replacing semantic components that cannot be preserved to generate multi-level LOD model data. This application reduces model complexity while achieving the coordinated preservation of semantic consistency and geometric accuracy, making it suitable for efficient rendering and visualization applications of urban-level 3D scenes.