Surveying and mapping area situation deduction method and system under multi-objective optimization

Through deep learning and multi-objective optimization technology, a high-precision three-dimensional geographic scenario model is generated, which solves the shortcomings of traditional surveying and mapping regional situation analysis methods in data fusion and situation modeling, and achieves efficient and refined decision-making support for situation deduction.

CN119006755AActive Publication Date: 2024-11-22BEIJING GENGTU TECH CO LTD

Patent Information

Application Number
CN202411456417.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-10-18
Publication Date
2024-11-22
Estimated Expiration
2044-10-18

Smart Images

  • Figure CN119006755A_ABST
    Figure CN119006755A_ABST
Patent Text Reader

Abstract

The invention provides a surveying and mapping area situation deduction method and system under multi-objective optimization, and relates to the technical field of surveying and mapping, and the method comprises the steps: obtaining the multi-source heterogeneous geographic information data of a surveying and mapping area, carrying out the feature extraction and semantic segmentation of the multi-source heterogeneous geographic information data through a deep learning algorithm, and obtaining the posture of the surveying and mapping area; generating a three-dimensional geographic scene model of the surveying and mapping area; constructing a multi-objective optimization model of surveying and mapping area situation deduction, and determining an optimal surveying and mapping equipment layout scheme set of the surveying and mapping area based on a simulation optimization algorithm; obtaining an optimal surveying and mapping equipment layout scheme based on each scheme in the optimal surveying and mapping equipment layout scheme set; in a three-dimensional geographic scene model, simulating a surveying and mapping operation process under the optimal surveying and mapping equipment layout scheme to obtain current three-dimensional situation information of a surveying and mapping area; and fusing the multi-temporal remote sensing image data and the geographic national condition census data, and predicting three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future based on the current three-dimensional situation information.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to surveying and mapping technology, and in particular to a surveying and mapping area situation deduction method and system under multi-objective optimization. Background Art

[0002] Traditional surveying and mapping regional situation analysis and deduction methods rely on manual experience and subjective judgment, and it is difficult to fully consider a variety of influencing factors, resulting in low accuracy and efficiency of situation deduction, which cannot meet actual application needs.

[0003] At present, the situation simulation of surveying and mapping areas mainly faces the following technical challenges: Fusion and processing of massive heterogeneous data: The surveying and mapping area involves multi-source heterogeneous data such as geographic information, remote sensing images, and social economy. How to effectively integrate and process these data is a major problem.

[0004] Situation modeling in complex dynamic environments: The surveying area is affected by many factors, including natural and human factors. How to accurately establish a situation model in a complex dynamic environment is the key.

[0005] Situation optimization under multi-objective conditions: Situation optimization in surveying and mapping areas needs to consider multiple objectives such as economic benefits, ecological protection, and disaster prevention and control at the same time. How to balance and optimize these objectives is a difficult point.

[0006] Visualization and interpretation of situation simulation results: How to visualize the results of situation simulation in an intuitive and easy-to-understand way and provide clear simulation basis and explanation is an urgent problem to be solved.

[0007] Although existing technologies have made certain progress in the deduction of regional situation in surveying and mapping, they still have the following shortcomings: focusing on the analysis of a single data source and lacking in the deep integration and mining of multi-source data.

[0008] It is difficult to describe the evolution of situation in complex dynamic environment by using simplified situation modeling method.

[0009] It mainly focuses on a single optimization objective and cannot achieve trade-offs and optimization under multi-objective conditions.

[0010] The deduction results are presented in a single form, with poor visualization effects and a lack of explanation of the deduction process. Summary of the invention

[0011] The embodiments of the present invention provide a method and system for deducing the situation of a surveying and mapping area under multi-objective optimization, which can at least solve some problems in the prior art.

[0012] A first aspect of an embodiment of the present invention provides a method for deducing a situation in a surveying and mapping area under multi-objective optimization, comprising: Acquire multi-source heterogeneous geographic information data of the surveying and mapping area, wherein the multi-source heterogeneous geographic information data includes remote sensing image data collected by different types of sensors, lidar point cloud data and geographic national conditions census data; use a deep learning algorithm to extract features and semantic segment the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying and mapping area, wherein the three-dimensional geographic scene model includes a high-precision terrain model, a semantic building model and a multi-level road network model; construct a multi-objective optimization model for situation deduction of the surveying and mapping area, wherein the multi-objective optimization model takes surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost and surveying and mapping accuracy as optimization objectives, and includes multiple sub-objective functions, wherein the weight of each sub-objective function is determined by using a hierarchical analysis method; in the three-dimensional geographic scene model, determine the optimal surveying and mapping equipment layout plan set for the surveying and mapping area based on a simulation optimization algorithm; Based on each plan in the optimal surveying and mapping equipment layout plan set, an optimal surveying and mapping equipment layout plan is obtained; in the three-dimensional geographic scene model, the surveying and mapping operation process under the optimal surveying and mapping equipment layout plan is simulated, and three-dimensional point cloud data consistent with the surveying and mapping operation process is generated to obtain the current three-dimensional situation information of the surveying and mapping area; multi-temporal remote sensing image data and geographic national conditions census data are integrated, and a long short-term memory neural network algorithm is used to predict the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future based on the current three-dimensional situation information; the three-dimensional situation evolution information is input into the multi-objective optimization model, the fitness value of each surveying and mapping equipment layout plan is recalculated, and the optimal surveying and mapping equipment layout plan is updated; based on the updated optimal surveying and mapping equipment layout plan, virtual reality technology is used to generate three-dimensional visualization scenes of the surveying and mapping area in different time periods in the future.

[0013] According to a second aspect of an embodiment of the present invention, a system for deducing a situation in a surveying and mapping area under multi-objective optimization is provided, including: a first unit for acquiring multi-source heterogeneous geographic information data of a surveying and mapping area, wherein the multi-source heterogeneous geographic information data includes remote sensing image data collected by different types of sensors, lidar point cloud data, and geographic national conditions census data; a deep learning algorithm is used to perform feature extraction and semantic segmentation on the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying and mapping area, wherein the three-dimensional geographic scene model includes a high-precision terrain model, a semantic building model, and a multi-level road network model; a second unit is used to construct a multi-objective optimization model for deducing a situation in a surveying and mapping area, wherein the multi-objective optimization model takes surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost, and surveying and mapping accuracy as optimization objectives, and includes multiple sub-objective functions, and the weight of each sub-objective function is determined by a hierarchical analysis method; in the three-dimensional geographic scene model, based on simulation optimization, The algorithm determines the optimal surveying and mapping equipment layout plan set for the surveying and mapping area; based on each plan in the optimal surveying and mapping equipment layout plan set, the optimal surveying and mapping equipment layout plan is obtained; in the three-dimensional geographic scene model, the surveying and mapping operation process under the optimal surveying and mapping equipment layout plan is simulated, and three-dimensional point cloud data consistent with the surveying and mapping operation process is generated to obtain the current three-dimensional situation information of the surveying and mapping area; the third unit is used to fuse multi-temporal remote sensing image data and geographic national conditions census data, and use a long short-term memory neural network algorithm to predict the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future based on the current three-dimensional situation information; the three-dimensional situation evolution information is input into the multi-objective optimization model, the fitness value of each surveying and mapping equipment layout plan is recalculated, and the optimal surveying and mapping equipment layout plan is updated; based on the updated optimal surveying and mapping equipment layout plan, virtual reality technology is used to generate three-dimensional visualization scenes of the surveying and mapping area in different time periods in the future.

[0014] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, including: A processor; a memory for storing processor-executable instructions; wherein the processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0015] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the method described above is implemented.

[0016] The method of the invention generates a three-dimensional geographic scene model of a surveying and mapping area with high precision, rich semantics and consistent logic.

[0017] This model not only outperforms traditional methods in geometric accuracy and visual effects, but also embeds semantic ontology-driven spatial reasoning and knowledge discovery capabilities to support intelligent spatial analysis and decision-making.

[0018] The present invention provides a new three-dimensional geographic scene reconstruction and application technology approach for surveying, mapping, planning, emergency response and other fields.

[0019] By comprehensively utilizing multi-source heterogeneous geographic information data and relying on deep learning algorithms and knowledge reasoning technology, a three-dimensional geographic scene model integrating multi-level information such as terrain, landforms, and roads is constructed to achieve a comprehensive and three-dimensional expression of the geographical environment of the surveying and mapping area.

[0020] This solution aggregates the advantages of multi-source geographic data and combines a variety of artificial intelligence technologies. It can significantly improve the level of automation and refinement of geographic scene reconstruction, and provide a solid foundation for the intelligent application of geographic information in various industries. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 A schematic diagram of a flow chart of a method for deducing a situation in a surveying and mapping area under multi-objective optimization according to an embodiment of the present invention; Figure 2 It is a structural schematic diagram of a surveying and mapping area situation deduction system under multi-objective optimization according to an embodiment of the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0023] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0024] The technical solution of the present invention is described in detail below with specific embodiments.

[0025] The following specific embodiments may be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0026] Figure 1 FIG. 1 is a flow chart of a method for deducing a situation in a surveying and mapping area under multi-objective optimization according to an embodiment of the present invention. Figure 1 As shown, the method includes: S101. Acquire multi-source heterogeneous geographic information data of the surveying and mapping area, wherein the multi-source heterogeneous geographic information data includes remote sensing image data collected by different types of sensors, laser radar point cloud data, and geographic national conditions census data; use a deep learning algorithm to perform feature extraction and semantic segmentation on the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying and mapping area, wherein the three-dimensional geographic scene model includes a high-precision terrain model, a semantic building model, and a multi-level road network model; S102. Construct a multi-objective optimization model for mapping area situation deduction, wherein the multi-objective optimization model takes mapping coverage, mapping efficiency, mapping cost and mapping accuracy as optimization objectives, and includes multiple sub-objective functions, and the weight of each sub-objective function is determined by the hierarchical analysis method; in the three-dimensional geographic scene model, determine the optimal mapping equipment layout plan set for the mapping area based on the simulation optimization algorithm; obtain the optimal mapping equipment layout plan based on each plan in the optimal mapping equipment layout plan set; in the three-dimensional geographic scene model, simulate the mapping operation process under the optimal mapping equipment layout plan, generate three-dimensional point cloud data consistent with the mapping operation process, and obtain the current three-dimensional situation information of the mapping area; S103. Integrate multi-temporal remote sensing image data and geographic national conditions census data, adopt long short-term memory neural network algorithm, and predict the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future based on the current three-dimensional situation information; input the three-dimensional situation evolution information into the multi-objective optimization model, recalculate the fitness value of each surveying and mapping equipment layout plan, and update the optimal surveying and mapping equipment layout plan; based on the updated optimal surveying and mapping equipment layout plan, use virtual reality technology to generate three-dimensional visualization scenes of the surveying and mapping area in different time periods in the future.

[0027] In an optional implementation, a deep learning algorithm is used to perform feature extraction and semantic segmentation on the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying area, including: The remote sensing image data is processed by joint radiometric calibration, multi-view geometric correction and multi-scale image fusion to generate a multi-source remote sensing image dataset with consistent spectral response and spatial geometric matching; a convolutional neural network algorithm based on an attention mechanism is used to extract the texture, spectral and spatial features of the multi-source remote sensing image dataset to generate a multi-channel remote sensing image feature map; the lidar point cloud data is processed by adaptive noise filtering, multi-scale ground point extraction and point cloud segmentation based on graph cuts to obtain lidar point cloud subsets of different types of land objects; a deep learning algorithm based on a graph attention network is used to extract the geometric features, topological relationship features and semantic attribute features of the lidar point cloud subset to generate a multi-dimensional point cloud feature vector; the geographic national conditions census data is processed by spatial coordinate transformation, attribute data standardization and multi-level topological reconstruction to generate a multi-source remote sensing image dataset and lidar point cloud feature map. A set of geographical national conditions elements with consistent spatial reference of cloud subsets; a deep learning algorithm based on a gated graph convolutional network is used to extract the semantic features, spatial relationship features and hierarchical structure features of the geographical national conditions elements set to generate a multi-resolution semantic feature map; a feature adaptive fusion strategy based on a self-attention mechanism is used to nonlinearly fuse the multi-channel dynamic remote sensing image feature map, the multi-dimensional point cloud feature vector and the multi-resolution semantic feature map to generate a fused multi-source heterogeneous geographic information feature set; the fused multi-source heterogeneous geographic information feature set is input into a multi-task deep learning model to generate a high-precision terrain model, a semantic building model and a multi-level road network model of the surveying and mapping area; a geographic information semantic ontology library is constructed, and the high-precision terrain model, the semantic building model and the multi-level road network model are semantically mapped and knowledge-associated with the geographic information semantic ontology library to obtain a three-dimensional geographic scene model of the surveying and mapping area.

[0028] Exemplarily, multi-source heterogeneous remote sensing image data, LiDAR point cloud data and geographic national survey data are preprocessed. Among them, remote sensing image data are processed by joint radiometric calibration, multi-view geometric correction and multi-scale image fusion to generate a multi-source remote sensing image data set with consistent spectral response and spatial geometric matching; LiDAR point cloud data are processed by adaptive noise filtering, multi-scale ground point extraction and graph-cut-based point cloud segmentation to obtain point cloud subsets of different types of land objects; geographic national survey data are processed by spatial coordinate conversion, attribute data standardization and multi-level topological reconstruction to generate a geographic national feature set consistent with the spatial reference of remote sensing image data and point cloud data.

[0029] Different deep learning algorithms are used to extract multi-level and multi-semantic feature information from remote sensing images, lidar point clouds and geographic conditions data.

[0030] For remote sensing image datasets, a convolutional neural network algorithm based on the attention mechanism is used to extract texture, spectral and spatial features, and generate multi-channel remote sensing image feature maps; for lidar point cloud subsets, a deep learning algorithm based on graph attention network is used to extract geometric features, topological relationship features and semantic attribute features, and generate multi-dimensional point cloud feature vectors; for geographic national conditions feature sets, a deep learning algorithm based on gated graph convolutional network is used to extract semantic features, spatial relationship features and hierarchical structure features, and generate multi-resolution semantic feature maps.

[0031] For the generated multi-channel remote sensing image feature maps, multi-dimensional point cloud feature vectors and multi-resolution semantic feature maps, a feature adaptive fusion strategy based on self-attention mechanism is adopted for deep nonlinear fusion.

[0032] This fusion strategy can adaptively assign weights and optimize the combination of geographic information features from different sources and with different semantics, thereby obtaining a multi-source heterogeneous geographic information fusion feature set that is rich in information and has clear semantics.

[0033] The fused feature set covers various information such as terrain, objects, and transportation in the surveying and mapping area, laying a solid data foundation for the subsequent reconstruction of three-dimensional geographic scenes.

[0034] The fused multi-source heterogeneous geographic information feature set is input into a multi-task deep learning model.

[0035] The model adopts an end-to-end joint optimization strategy and realizes information interaction and complementarity between different reconstruction tasks by sharing hidden layer parameters.

[0036] The output of the model includes three branches: a high-precision terrain model, a semantic building model and a multi-level road network model.

[0037] Among them, the terrain model reconstruction branch adopts a regression loss function to generate a digital surface model (DSM) that is close to the actual terrain elevation value; the building model reconstruction branch adopts a multi-label classification loss function, which not only reconstructs the three-dimensional geometric shape of the building, but also semantically annotates the building's category, function and other attribute information; the road network model reconstruction branch adopts a graph structure embedding loss function to generate a multi-level three-dimensional road network with complete topological relationships and clear functional levels.

[0038] Ultimately, the multi-task deep learning model outputs high-precision, multi-semantic, and multi-level three-dimensional geographic elements of the surveying and mapping area.

[0039] Construct a geographic information semantic ontology library for three-dimensional GIS, formally define basic geographic feature classes, spatial relationship classes and thematic semantic classes, and form a conceptual hierarchy system and logical reasoning rules for geographic information semantics in the field of surveying and mapping.

[0040] Then, the generated three-dimensional geographic elements are semantically mapped with the classes and relationships in the ontology library, and rich semantic attribute information is added to the three-dimensional model, so that the three-dimensional geographic scene model has a semantic connotation that is machine-understandable and reasonable.

[0041] Finally, the ontology reasoning engine is used to perform spatial semantic consistency check and conflict resolution on the three-dimensional geographic scene model to ensure the completeness and consistency of the model in terms of logical semantics.

[0042] After the above steps, the method of the present invention finally generates a high-precision, semantically rich, and logically consistent three-dimensional geographic scene model of the surveying and mapping area.

[0043] This model not only outperforms traditional methods in geometric accuracy and visual effects, but also embeds semantic ontology-driven spatial reasoning and knowledge discovery capabilities to support intelligent spatial analysis and decision-making.

[0044] The present invention provides a new three-dimensional geographic scene reconstruction and application technology approach for surveying, mapping, planning, emergency response and other fields.

[0045] In an optional implementation, a deep learning algorithm is used to perform feature extraction and semantic segmentation on the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying area, including: The remote sensing image data is processed by joint radiometric calibration, multi-view geometric correction and multi-scale image fusion to generate a multi-source remote sensing image dataset with consistent spectral response and spatial geometric matching; a convolutional neural network algorithm based on an attention mechanism is used to extract the texture, spectral and spatial features of the multi-source remote sensing image dataset to generate a multi-channel remote sensing image feature map; the lidar point cloud data is processed by adaptive noise filtering, multi-scale ground point extraction and point cloud segmentation based on graph cuts to obtain lidar point cloud subsets of different types of land objects; a deep learning algorithm based on a graph attention network is used to extract the geometric features, topological relationship features and semantic attribute features of the lidar point cloud subset to generate a multi-dimensional point cloud feature vector; the geographic national conditions census data is processed by spatial coordinate transformation, attribute data standardization and multi-level topological reconstruction to generate a multi-source remote sensing image dataset and lidar point cloud feature map. A set of geographical national conditions elements with consistent spatial reference of cloud subsets; a deep learning algorithm based on a gated graph convolutional network is used to extract the semantic features, spatial relationship features and hierarchical structure features of the geographical national conditions elements set to generate a multi-resolution semantic feature map; a feature adaptive fusion strategy based on a self-attention mechanism is used to nonlinearly fuse the multi-channel dynamic remote sensing image feature map, the multi-dimensional point cloud feature vector and the multi-resolution semantic feature map to generate a fused multi-source heterogeneous geographic information feature set; the fused multi-source heterogeneous geographic information feature set is input into a multi-task deep learning model to generate a high-precision terrain model, a semantic building model and a multi-level road network model of the surveying and mapping area; a geographic information semantic ontology library is constructed, and the high-precision terrain model, the semantic building model and the multi-level road network model are semantically mapped and knowledge-associated with the geographic information semantic ontology library to obtain a three-dimensional geographic scene model of the surveying and mapping area.

[0046] For example, multi-source remote sensing image data from different sensors and at different phases are calibrated radiometrically to eliminate sensor response differences and atmospheric influences, so that the radiometric brightness values ​​of the image data are consistent.

[0047] Specifically, it includes: selecting the overlapping area of ​​multi-source images, extracting the radiation brightness value of stable and unchanging objects, and establishing a radiation brightness conversion function; using the radiation brightness conversion function, performing radiation correction on each source image to unify its radiation brightness value to the same scale.

[0048] Due to differences in imaging angles and sensor postures, remote sensing images from different perspectives have geometric distortion and position deviation in space.

[0049] In order to achieve accurate image registration, multi-view geometric correction is required.

[0050] Specifically, it includes: selecting high-precision control points to establish a geometric relationship model between the image and the ground; using the geometric relationship model, orthorectifying, resampling and registering images from each perspective to eliminate the geometric distortion caused by perspective differences and achieve accurate spatial matching of images.

[0051] In order to fully utilize the temporal and spatial resolution advantages of images from different sensors, multi-scale image fusion is required.

[0052] Specifically, it includes: using multi-scale decomposition methods such as wavelet transform or Laplace pyramid to decompose high-resolution panchromatic images and low-resolution multispectral images into multiple scale levels; at each scale level, using local mean matching, edge enhancement and other strategies to inject the spatial detail information of the panchromatic image into the multispectral image; using inverse transform to reconstruct the fused high-resolution multispectral image to achieve the simultaneous improvement of spatial information and spectral information.

[0053] Traditional convolutional neural networks use fixed convolution kernels for feature extraction, which makes it difficult to adapt to the multi-scale and multi-directional characteristics of ground objects.

[0054] The introduction of the attention mechanism can adaptively adjust the weights of the convolution kernel and selectively focus on significant features in different regions and scales.

[0055] Specifically, in the encoder part of the convolutional neural network, the spatial distribution and channel weights of the convolutional feature map are adaptively adjusted through the two sub-modules of spatial attention and channel attention; the spatial attention highlights the salient features of the target area by generating a spatial weight map; the channel attention emphasizes the feature channels related to the target by generating a channel weight vector; in the decoder part, the multi-scale attention feature map generated by the encoder is upsampled and cross-level feature fused to obtain a multi-channel remote sensing image feature map.

[0056] Using a convolutional neural network based on the attention mechanism, multi-level and multi-scale features of multi-source remote sensing image data sets are extracted to generate multi-channel remote sensing image feature maps.

[0057] Specifically, it includes: texture feature map, which extracts texture statistics such as gray-level co-occurrence matrix and Gabor filter from the image to characterize the surface texture pattern of the object; spectral feature map, which extracts spectral indexes such as normalized vegetation index and normalized building index from the image to reflect the differences in spectral properties of the object; spatial feature map, which extracts geometric features such as shape, size, and direction of the image to describe the spatial structural characteristics of the object.

[0058] The multi-channel remote sensing image feature map comprehensively describes the texture, spectral and spatial characteristics of the objects, providing rich discriminant information for subsequent semantic segmentation.

[0059] Ground points are an important type of land objects in point clouds. They need to be accurately extracted from point clouds to provide a reference for subsequent land object classification.

[0060] A multi-scale ground point extraction method is adopted to make full use of the smoothness and continuity characteristics of the ground.

[0061] Specifically, it includes: using a gridding method to divide the point cloud into multiple grids of different scales; within each grid, using the RANSAC (random sampling consensus) algorithm to fit the local plane equation and extract points that meet the plane characteristics as candidate ground points; between adjacent grids, using the regional growing algorithm to connect the candidate ground points with consistent normal vectors and gradual elevation changes into pieces to obtain a complete ground point cloud.

[0062] In order to achieve point cloud segmentation at the object level, it is necessary to make full use of the geometric and topological features of the point cloud.

[0063] A point cloud segmentation method based on graph cut is adopted. The point cloud is regarded as an undirected graph, and the global optimal segmentation of the point cloud is achieved through the minimum cut algorithm.

[0064] Specifically, it includes: constructing an undirected graph of point clouds based on the spatial adjacency relationship of point clouds, with each point as a node of the graph and edges connecting adjacent points; calculating the weight of the edge based on the difference in geometric features between the two endpoints of the edge connection, the greater the difference in geometric features, the smaller the weight; using the maximum flow minimum cut algorithm to find the minimum cut on the weight graph, dividing the point cloud into different connected areas to achieve point cloud segmentation; for each point cloud area after segmentation, extracting its geometric and topological features and assigning preliminary feature category labels.

[0065] Traditional point cloud feature extraction methods, such as point feature histogram and fast point feature histogram, only consider the local geometric features of points and ignore the topological relationship and semantic dependency between points.

[0066] The introduction of graph attention network can simultaneously model the local and global features of point clouds and learn the interactions between points. Specifically, it includes: converting point clouds into graph structure data, with each point as a node of the graph, and node features including the coordinates, normal vector and other geometric information of the point; applying multi-layer graph convolution on the graph, and each layer of convolution updates the node features by aggregating the neighborhood information of the node; while performing graph convolution, introducing the attention mechanism, according to the similarity of node features, adaptively adjusting the contribution of different neighborhoods to the update of the central node features; after iterative optimization of multi-layer graph convolution and attention, a high-dimensional feature descriptor of each point is obtained, which contains both the local geometric features of the point and the global semantic features.

[0067] Using a deep learning algorithm based on graph attention network, multi-dimensional point cloud features are extracted from the lidar point cloud subset to generate a point cloud feature vector.

[0068] Specifically, it includes: geometric features, which extract the three-dimensional coordinates, normal vector, curvature and other geometric properties of each point to characterize the spatial shape characteristics of the object where the point is located; topological relationship features, which extract the topological structure features such as the connectivity and adjacency relationship between each point and its neighboring points to reflect the spatial distribution pattern of the object; semantic attribute features, which extract the probability distribution, mutual exclusion relationship and other semantic features of the object category to which each point belongs to reveal the object composition of the point cloud scene.

[0069] The multi-dimensional point cloud feature vector comprehensively characterizes the geometric, topological and semantic features of the point cloud, providing an important basis for subsequent point cloud segmentation and three-dimensional reconstruction.

[0070] The spatial coordinate system used in the national geographic census data is often different from the coordinate system of remote sensing images and lidar point clouds, and spatial coordinate conversion is required to achieve the unification of spatial positions.

[0071] Specifically, it includes: selecting control points and establishing conversion relationships between different coordinate systems; using coordinate conversion relationships to uniformly convert the coordinates of point, line, and surface elements in geographic national conditions data to the target coordinate system; checking and correcting the spatial topological relationships of the converted elements to ensure the correctness and consistency of the elements in space.

[0072] The names, types and formats of attribute items in the national geographic census data are different, and attribute data standardization is needed to ensure the standardization and interoperability of attribute data.

[0073] Specifically, it includes: referring to relevant national standards, standardizing the names and types of each attribute item, unifying the naming and type standards; converting the format and units of attribute values ​​to make them conform to standard specifications; checking and correcting missing attributes and erroneous attributes, eliminating abnormal values, supplementing default values, and improving the quality of attribute data.

[0074] The features in geographic national conditions data have complex topological relationships in space, and multi-level topological reconstruction is needed to form a topologically connected and structurally reasonable feature network.

[0075] Specifically, it includes: within the feature class, based on the spatial adjacency relationship, constructing the adjacency topology between surface features and the connectivity topology between line features; between feature classes, according to the semantic association rules, constructing the subordinate topology between upper and lower level features and the composition topology between different feature classes at the same level; using the topological consistency check method to verify the correctness of the reconstructed topological relationship, and making necessary corrections, ultimately forming a feature network with complete topology and correct relationships.

[0076] Traditional feature extraction methods mainly focus on the geometric and attribute characteristics of features, ignoring the semantic connections and spatial dependencies between features.

[0077] The introduction of gated graph convolutional networks can simultaneously model the local features and global relationships of elements and learn the interactions between elements.

[0078] Specifically, it includes: converting the geographical national conditions factor network into graph structure data, with each factor as a node of the graph, and the node features include the geometry, attributes, topology and other information of the factor; applying multi-layer gated graph convolution on the graph, with each layer of convolution updating the node representation by aggregating the neighborhood information of the node, and using the gating mechanism to control the transmission intensity of the neighborhood information; while performing graph convolution, introducing the attention mechanism, dynamically adjusting the contribution of different neighborhood nodes to the update of the central node according to the similarity of the node representation; after iterative optimization of multi-layer graph convolution and attention, a high-dimensional feature representation of each factor is obtained, which integrates the local features of the factor and the global context information.

[0079] Using a deep learning algorithm based on a gated graph convolutional network, multi-resolution semantic features are extracted from a set of geographical national conditions elements to generate a semantic feature map.

[0080] Specifically, it includes: spatial relationship feature map, which extracts spatial features such as orientation relationship and distance relationship between adjacent elements, and reveals the spatial distribution pattern of ground features; semantic attribute feature map, which extracts semantic features such as attribute statistics and association rules of each feature category, and describes the semantic characteristics of feature types; hierarchical structure feature map, which extracts structural features such as inclusion and composition between elements at different levels, and reflects the organizational form of feature network. Multi-resolution semantic feature map comprehensively represents the semantic connotation of geographical national conditions elements, and provides a bridge for subsequent knowledge association.

[0081] Multi-source heterogeneous geographic information data have different feature representations and semantic granularity. How to effectively integrate features from different sources is crucial.

[0082] The feature adaptive fusion strategy based on the self-attention mechanism can automatically adjust the importance of different features according to task requirements and realize selective fusion of features.

[0083] Specifically, it includes: cascading feature vectors from different sources such as multi-channel remote sensing image features, multi-dimensional point cloud features, and multi-resolution semantic features to form an initial feature set; calculating the correlation matrix between different features through a multi-head self-attention mechanism to explore the intrinsic connections between features; adaptively generating a weight coefficient for each feature based on the correlation matrix to highlight significant features that are closely related to the task; multiplying the initial feature set by the weight coefficient to achieve weighted fusion of features and obtain a fused feature set with comprehensive information and low redundancy.

[0084] The fused feature set contains rich information from multi-source heterogeneous data, but there may be complex nonlinear relationships between different features.

[0085] By using nonlinear feature fusion methods, we can further explore the deep semantic associations between features.

[0086] Specifically, it includes: inputting the fused feature set into the multi-layer perceptron neural network, and performing nonlinear transformation and combination on the features through nonlinear activation functions such as ReLU; in the hidden layer, automatically learning the high-order interactions between different features through the fully connected weight matrix; in the output layer, the converted deep features are fused again to obtain semantically rich and expressive fused features.

[0087] The nonlinear feature fusion process automatically models the complex mapping relationship between multi-source heterogeneous features, making the fused features more compact and consistent in the semantic space.

[0088] For multiple subtasks of three-dimensional geographic scene reconstruction, such as terrain modeling, building extraction, road network extraction, etc., traditional methods usually design and optimize each subtask independently, ignoring the association constraints between subtasks.

[0089] By adopting an end-to-end multi-task joint modeling method, multiple subtasks can be optimized simultaneously under a unified framework, making full use of the complementary information between subtasks and improving the overall performance of scene understanding.

[0090] Specifically, it includes: sharing the multi-source heterogeneous geographic information feature sets as input into the multi-task network; in the shared encoding layer, using the common feature extractor to learn the basic feature representation of the scene; in the task-specific decoding layer, designing a specific decoder structure for each subtask to generate task-specific outputs; between tasks, introducing an associated loss function to model the logical constraint relationship between different task outputs; and by jointly optimizing the loss functions of multiple tasks, improving the prediction performance of each subtask at the same time.

[0091] Utilizing a multi-task deep learning model, we simultaneously generate terrain models, building models, and road network models for three-dimensional geographic scenes, comprehensively describing the surface morphology and object distribution of the scene.

[0092] Specific outputs include: terrain model, which extracts terrain attributes such as elevation, slope, and aspect, depicts the surface undulations of the scene, and generates a high-resolution digital surface model; building model, which extracts information such as the spatial position, geometric shape, and semantic category of the building, generates vectorized three-dimensional building instances, and reflects the distribution of artificial facilities in the scene; road network model, which extracts information such as the spatial direction, width, and type of the road, generates a topologically connected road network vector diagram, and depicts the traffic network structure of the scene.

[0093] The three-dimensional model generated by multiple tasks comprehensively reflects the spatial structure and landform composition of the geographic scene, laying the foundation for subsequent scene applications.

[0094] In order to deepen the semantic connotation of three-dimensional geographic scene models, it is necessary to build a geographic information semantic ontology library to form a hierarchical structure and logical relationship between geographic concepts.

[0095] Specifically, it includes: defining the category system of geographical concepts based on geographical theories and industry norms, and organizing them into an ontology hierarchy; defining the attribute characteristics and external relationship constraints of each geographical concept to form semantic connections between ontology concepts; using ontology description languages ​​such as OWL to formally represent ontology hierarchies, concept definitions, relationship constraints, etc., and construct a computer-readable ontology knowledge base; using ontology mapping methods to associate multi-source heterogeneous geographic data with ontology concepts, and realizing semantic annotation and knowledge organization of geographic data.

[0096] The geographic information semantic ontology library forms a semantic association network between geographic concepts, providing rich semantic support for three-dimensional scene models.

[0097] By using the geographic information semantic ontology library, the feature instances in the three-dimensional geographic scene model are semantically mapped with the ontology concepts to achieve a deep association between the scene model and geographic knowledge.

[0098] Specifically, it includes: abstracting three-dimensional model elements such as terrain models, building models, and road network models into instances of ontology concepts, such as mapping building models into instances of concepts such as "house" and "school"; based on the attribute definition of concepts in the ontology, extracting corresponding attribute values ​​from model elements to complete the attribute association of concept instances; based on the relationship constraints between concepts in the ontology, establishing semantic connections between model instances, such as establishing an "adjacent" relationship between building instances and road instances; on the basis of semantic associations, using the ontology reasoning mechanism to discover implicit geographic concept associations and further expand the semantic network of three-dimensional scene models.

[0099] Ultimately, the three-dimensional geographic scene model with pure geometric meaning is transformed into a geographic knowledge expression with rich semantics, which enhances the intelligence and usability of the model.

[0100] Through the above technical solution, we comprehensively utilize multi-source heterogeneous geographic information data, and with the help of deep learning algorithms and knowledge reasoning technology, we construct a three-dimensional geographic scene model that integrates multi-level information such as terrain, landforms, roads, etc., to achieve a comprehensive and three-dimensional expression of the geographical environment of the surveying and mapping area.

[0101] This solution aggregates the advantages of multi-source geographic data and combines a variety of artificial intelligence technologies. It can significantly improve the level of automation and refinement of geographic scene reconstruction, and provide a solid foundation for the intelligent application of geographic information in various industries.

[0102] In an optional implementation, a multi-objective optimization model for mapping area situation deduction is constructed, including: taking mapping target data, mapping accuracy data, mapping time data and mapping cost data as inputs of the multi-objective optimization model, establishing mapping coverage sub-objective function, mapping efficiency sub-objective function, mapping cost sub-objective function and mapping accuracy sub-objective function; wherein the independent variable of the mapping coverage sub-objective function is the mapping status of each grid, and the output is a mapping coverage evaluation index; the independent variable of the mapping accuracy sub-objective function is the accuracy parameters of each mapping sensor and the parameters of the fusion algorithm, and the output is a mapping accuracy evaluation index; the independent variable of the mapping cost sub-objective function is the input of each cost factor, and the output is a mapping cost evaluation index; the independent variable of the mapping efficiency sub-objective function is the type and quantity of mapping equipment selected for each work unit, and the output is a mapping efficiency evaluation index; collect dynamic preference information of mapping task requirements, construct a pairwise comparison matrix between different optimization targets of the dynamic preference information, and obtain a dynamic judgment matrix, wherein the element values ​​of the dynamic judgment matrix represent two The dynamic quantitative evaluation value of the relative importance of the optimization target is obtained; the spectral decomposition method is used to calculate the maximum eigenvalue of the dynamic judgment matrix and its corresponding eigenvector, a fuzzy consistency evaluation model of the dynamic judgment matrix is ​​constructed, and the dynamic judgment matrix is ​​solved based on the fuzzy consistency evaluation model to obtain the maximum fuzzy eigenvalue and fuzzy eigenvector that are compatible with the dynamic preference information; the fuzzy eigenvector obtained by the solution is normalized to obtain the initial weight vector, and a subjective and objective combined weighting model is constructed to optimize and correct the initial weight vector to obtain the combined weight vector; the fuzzy consistency ratio of the dynamic judgment matrix is ​​calculated, and whether the matrix conforms to the logical consistency is determined by setting the fuzzy consistency ratio threshold; when the test fails, the feedback mechanism is used to adjust the element value of the judgment matrix, and the combined weight vector is recalculated and a consistency test is performed; when the test passes, the combined weight vector is used as the dynamic weighting coefficient of the surveying and mapping coverage sub-objective function, the surveying and mapping efficiency sub-objective function, the surveying and mapping cost sub-objective function and the surveying and mapping accuracy sub-objective function to construct the optimization target of the multi-objective optimization model.

[0103] Exemplarily, surveying target data, surveying accuracy data, surveying time data and surveying cost data are used as inputs of the multi-objective optimization model.

[0104] Among them, the surveying and mapping target data includes the expected surveying and mapping status of each grid in the survey area; the surveying and mapping accuracy data includes the accuracy parameters of various sensors and the parameter settings of the data fusion algorithm; the surveying and mapping time data includes the time schedule and progress requirements of the surveying and mapping operations; the surveying and mapping cost data includes the budget constraints of various cost elements such as personnel, equipment, and materials.

[0105] Based on the model input, four sub-objective functions are established, namely, surveying and mapping coverage sub-objective function, surveying and mapping efficiency sub-objective function, surveying and mapping cost sub-objective function and surveying and mapping accuracy sub-objective function.

[0106] The independent variable of the survey coverage sub-objective function is the survey status of each grid, which takes a value of 0 or 1, indicating whether the grid is covered by survey.

[0107] The output of the sub-objective function is the overall surveying and mapping coverage evaluation index of the survey area, which comprehensively reflects the coverage integrity and balance of surveying and mapping operations in different areas.

[0108] The independent variables of the sub-objective function of surveying and mapping accuracy are the internal and external orientation elements, resolution and other accuracy parameters of each surveying and mapping sensor, as well as the weight, threshold and other parameters of the multi-sensor fusion algorithm. The output of the sub-objective function is the overall surveying and mapping accuracy evaluation index of the survey area, which comprehensively evaluates the data accuracy level under different sensor combination schemes and fusion strategies.

[0109] The independent variables of the surveying and mapping cost sub-objective function are the input amounts of various cost factors such as labor costs, equipment costs, and material costs.

[0110] The output of the sub-objective function is the total cost evaluation index of the surveying and mapping operation, which quantitatively describes the economic efficiency under different resource allocation schemes.

[0111] The independent variables of the surveying and mapping efficiency sub-objective function are the type and quantity of surveying and mapping equipment selected for each work unit.

[0112] The work unit can be a sub-area of ​​the survey area or a phased task.

[0113] The output of the sub-objective function is the surveying and mapping efficiency evaluation index, which measures the operation speed and timeliness under different equipment scheduling schemes.

[0114] Collect dynamic preference information of decision makers on surveying and mapping task requirements, and judge the relative importance of different optimization goals based on factors such as task priority, accuracy requirements, and funding constraints.

[0115] Construct a pairwise comparison matrix of optimization objectives to form a dynamic judgment matrix.

[0116] The element values ​​in the matrix represent the dynamic quantitative evaluation values ​​of the relative importance of the two optimization objectives in the row and column, usually ranging from 1 to 9 and its reciprocal.

[0117] The spectral decomposition method is used to calculate the maximum eigenvalue of the dynamic judgment matrix and its corresponding eigenvector.

[0118] A fuzzy consistency evaluation model of dynamic judgment matrix is ​​constructed, and the element values ​​of the judgment matrix are expressed as fuzzy numbers, taking into account the uncertainty of the judgment.

[0119] The dynamic judgment matrix is ​​solved based on the fuzzy consistency evaluation model to obtain the maximum fuzzy eigenvalue and fuzzy eigenvector that are compatible with the current preference information.

[0120] The solved fuzzy feature vector is normalized to obtain the initial weight vector, which quantitatively represents the importance of each sub-goal.

[0121] However, the initial weights given solely based on subjective judgment may be biased and need further optimization and correction.

[0122] A subjective and objective combined weighting model is constructed. On the basis of the initial weights, objective data constraints are introduced, such as the value range of each sub-objective function, sensitivity analysis, etc., and the weight vector is adaptively adjusted to obtain a combined weight vector.

[0123] Calculate the fuzzy consistency ratio of the dynamic judgment matrix to measure the degree of logical consistency of the judgment matrix.

[0124] Set the fuzzy consistency ratio threshold, usually 0.1.

[0125] When the consistency ratio exceeds the threshold, the judgment matrix has a large logical contradiction and cannot be used directly.

[0126] At this time, a feedback mechanism is used to require decision makers to readjust the values ​​of certain elements of the judgment matrix until it passes the consistency test.

[0127] The combined weight vector that passes the consistency test is used as the dynamic weighting coefficient of the four sub-objective functions of surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost and surveying and mapping accuracy. The weighted sum of multiple sub-objective functions is obtained to obtain the overall optimization goal of the multi-objective optimization model.

[0128] The optimization goal comprehensively balances the importance of each sub-goal under different demand preferences, and is a trade-off and quantification of the overall performance of the surveying and mapping task.

[0129] According to the actual requirements of the surveying and mapping tasks and resource limitations, the constraints of the multi-objective optimization model are set.

[0130] Common constraints include: the surveying and mapping coverage must reach a predetermined threshold; the surveying and mapping accuracy must meet the specification requirements; the surveying and mapping costs must not exceed the budget limit; the surveying and mapping efficiency must be completed within the specified time limit; the number of different types of surveying and mapping equipment is limited by available resources; certain key areas or targets must be surveyed first; the sensor working mode, flight route, etc. must meet specific rules.

[0131] According to the optimization objectives and constraints, the mathematical form of the multi-objective optimization model is constructed. The model can be solved by using typical multi-objective optimization algorithms such as weighted summation method, ε-constraint method, genetic algorithm, etc.

[0132] During the solution process, by adjusting the model parameters and searching the feasible domain of the optimization variables, we can find the solution set that meets the constraints and achieves the optimal optimization goal.

[0133] Each solution vector in the solution set represents an optimal configuration plan for surveying and mapping operations, including the values ​​of decision variables such as the surveying and mapping status of each grid, the type and quantity of selected sensors, and the investment level of each cost factor.

[0134] The above-mentioned technical solution can dynamically adapt to changes in surveying and mapping demand preferences by constructing and solving a multi-objective optimization model for surveying and mapping area situation deduction, balance multiple performance objectives such as surveying and mapping coverage, accuracy, cost, and efficiency, and combine subjective and objective factors to optimize and dynamically optimize surveying and mapping schemes, providing a theoretical basis and technical support for scientific decision-making and precise implementation of surveying and mapping tasks, and has good practicality and promotion value.

[0135] In an optional implementation, establishing a surveying and mapping coverage sub-objective function, a surveying and mapping efficiency sub-objective function, a surveying and mapping cost sub-objective function, and a surveying and mapping accuracy sub-objective function includes: The area to be measured is divided into grids of equal size by using the rasterization method, and the side length of each grid is adaptively determined according to the geometric characteristics of the surveying and mapping target and the required spatial resolution. The surveying and mapping status of each grid is statistically analyzed to form a surveying and mapping status matrix, and the key ground object elements in the area to be measured are extracted from remote sensing images or laser point cloud data using ground object element extraction technology. The surveying and mapping completeness rate of each type of ground object element is calculated, and the surveying and mapping coverage sub-objective function is constructed with the ground object element surveying and mapping completeness rate being greater than a given threshold as a constraint condition; the surveying and mapping task is decomposed into multiple work units from top to bottom by using the hierarchical decomposition method, and the standard working hours required for each work unit to complete are determined according to the workload and complexity of the work. At the same time, the completion efficiency of each unit of work is calculated based on the equipment performance parameters and work experience for different surveying and mapping equipment, and the surveying and mapping time data is used as a constraint condition to construct a sub-objective function. A surveying and mapping efficiency sub-objective function is constructed; various cost elements related to the surveying and mapping tasks are systematically classified and aggregated by adopting cost accounting methods, cost estimation models are selected for different types of cost elements, and a total cost calculation model is constructed to obtain the budget cost for completing the entire surveying and mapping task. The cost sensitivity analysis technology is used to identify key cost driving factors, and the surveying and mapping cost sub-objective function is constructed with the surveying and mapping cost data as a constraint condition; the accuracy parameters of different surveying and mapping sensors are used as optimization variables, and the functional relationship between the surveying and mapping accuracy evaluation index and the sensor accuracy parameters is established. The differential evolution algorithm is used to search for the optimal combination of surveying and mapping sensor accuracy parameters. After obtaining the optimal accuracy parameters of each surveying and mapping sensor, the multi-source data fusion technology is used to fuse the surveying and mapping data obtained by different sensors, and the surveying and mapping accuracy sub-objective function is constructed with the surveying and mapping accuracy data as a constraint condition.

[0136] Exemplarily, a rasterization method is used to divide the area to be measured into grids of equal size.

[0137] The side length of the grid is adaptively determined based on the geometric characteristics of the surveying and mapping target (such as the size, shape, distribution density, etc. of the target) and the required spatial resolution (i.e. the degree of detail of the surveying and mapping results).

[0138] The selection of grid side length must take into account both surveying efficiency and surveying accuracy. A grid that is too large will reduce the spatial resolution of the survey, while a grid that is too small will increase the workload of the survey.

[0139] The surveying and mapping status of each grid is counted to form a surveying and mapping status matrix.

[0140] The number of rows and columns of the matrix are the number of rows and columns after rasterization, and the matrix elements are 0 or 1, indicating whether the corresponding grid is covered by the survey.

[0141] Initially, all elements of the mapping state matrix are 0, indicating that all grids have not been mapped. As the mapping operation proceeds, the matrix elements of the grids covered by the mapping are set to 1.

[0142] Using the ground feature extraction technology, key ground features in the area to be measured, such as buildings, roads, water systems, etc., are extracted from remote sensing images or laser point cloud data. For each type of ground feature, the number of grids in which it appears is counted, and the number of grids actually covered by the survey is calculated, thereby obtaining the surveying and mapping completeness rate of this type of ground feature, that is, the proportion of the number of grids covered by the survey to the total number of grids.

[0143] Taking the mapping completeness rate of ground features greater than a given threshold as a constraint, a sub-objective function of mapping coverage is constructed.

[0144] The size of the threshold is determined according to the requirements of the surveying and mapping task. For example, the threshold of important land features can be set to 100%, and the threshold of ordinary land features can be appropriately relaxed.

[0145] The independent variable of the sub-objective function is the surveying and mapping status of each grid, and the objective function value is the arithmetic average of the surveying and mapping status of all grids, reflecting the overall surveying and mapping coverage level of the survey area.

[0146] The hierarchical decomposition method is used to decompose the entire surveying and mapping task into multiple work units from top to bottom.

[0147] The work unit can be a sub-area of ​​the survey area, or a stage task in the surveying and mapping process, such as data collection, data preprocessing, feature extraction, etc.

[0148] Each work unit has clear work content, completion standards and time limit requirements.

[0149] For each work unit, estimate the standard working hours required to complete the unit based on its workload and complexity.

[0150] The workload reflects the amount of work, such as the size of the surveying area, the number of target elements, etc.; the workload complexity reflects the difficulty of the work, such as the complexity of the terrain conditions, the degree of detail of the target elements, etc.

[0151] Standard working hours are the time required to complete a unit of work under certain working conditions and resource inputs.

[0152] For different surveying and mapping equipment, such as UAVs, total stations, GPS, etc., the completion efficiency of unit workload is calculated based on their performance parameters (such as speed, range, accuracy, etc.) and operating experience.

[0153] Completion efficiency refers to the amount of work completed by the equipment in a unit of time, such as the area surveyed per hour, the amount of data collected per day, etc.

[0154] Taking the time data of surveying and mapping tasks (such as total construction period, milestone nodes, etc.) as constraints, the sub-objective function of surveying and mapping efficiency is constructed.

[0155] The independent variables of the sub-objective function are the type and quantity of surveying and mapping equipment selected for each work unit. The objective function value is the weighted sum of the completion time of all work units, and the weight is determined according to the importance of the work unit.

[0156] The smaller the objective function value, the higher the overall surveying and mapping efficiency.

[0157] The cost accounting method is used to systematically classify and aggregate various cost elements related to surveying and mapping tasks.

[0158] Cost elements can be divided into direct costs and indirect costs.

[0159] Direct costs are the costs specifically used for the surveying and mapping task, such as labor costs, equipment costs, material costs, etc.; indirect costs are public costs serving multiple tasks, such as management fees, water and electricity costs, etc., which need to be allocated to the task according to a certain proportion.

[0160] For different types of cost elements, choose the appropriate cost estimation model.

[0161] For example, labor costs can be estimated based on the number of surveying and mapping personnel, working hours and standard wage rates; equipment costs can be estimated based on the type, number of equipment, usage time and rental unit price; material costs can be estimated based on the type, quantity and purchase unit price of materials; and management fees can be estimated as a certain percentage of the total cost of the surveying and mapping task.

[0162] Based on the estimation of cost elements, a total cost calculation model is constructed to obtain the budget cost for completing the entire surveying and mapping task.

[0163] The total cost model dynamically estimates the total cost by adding up the estimated values ​​of each cost element and taking into account the mutual influence and risk factors between cost elements, such as inflation and exchange rate fluctuations.

[0164] Cost sensitivity analysis technology is used to identify key cost driving factors that have a greater impact on total cost, such as operation scale, technical difficulty, construction period constraints, etc. Through sensitivity analysis, the optimization direction and adjustment range of cost factors are determined to provide a basis for cost control.

[0165] Taking the cost data of surveying and mapping tasks (such as total budget, single item budget, etc.) as constraints, the sub-objective function of surveying and mapping cost is constructed.

[0166] The independent variable of the sub-objective function is the input of each cost element, and the objective function value is the total cost estimate. On the premise of meeting the task requirements and standards, the smaller the objective function value, the higher the degree of cost optimization.

[0167] Taking the accuracy parameters of different surveying and mapping sensors (such as cameras, lidar, synthetic aperture radar, etc.) as optimization variables, a functional relationship between surveying and mapping accuracy evaluation indicators and sensor accuracy parameters is established.

[0168] The evaluation indicators of surveying and mapping accuracy can be mean error, relative accuracy, etc., which reflect the accuracy of surveying and mapping results.

[0169] Sensor accuracy parameters include measurement accuracy, positioning accuracy, stability, etc., which determine the surveying and mapping performance of the sensor.

[0170] The differential evolution algorithm is used to search for the optimal parameter combination within the range of sensor accuracy parameters.

[0171] The differential evolution algorithm simulates the biological evolution process, continuously iteratively updates the solution population, and finally converges to the global optimal solution.

[0172] During the iteration process, the algorithm adjusts the solution mutation, crossover and selection strategies according to the accuracy function value of the current solution to balance the quality and diversity of the solution.

[0173] After obtaining the optimal accuracy parameters of each surveying and mapping sensor, multi-source data fusion technology is used to fuse the surveying and mapping data obtained by different sensors.

[0174] Data fusion can be performed at the decision layer, feature layer or data layer, such as weighted average method, Kalman filter method, wavelet transform method, etc.

[0175] The fused mapping data combines the advantages of multiple sensors, and its accuracy level is higher than that of any single sensor.

[0176] Taking the accuracy data of surveying and mapping tasks (such as accuracy index, accuracy level, etc.) as constraints, a surveying and mapping accuracy sub-objective function is constructed.

[0177] The independent variables of the sub-objective function are the accuracy parameters of each surveying and mapping sensor and the parameters of the data fusion algorithm, and the objective function value is the surveying and mapping accuracy evaluation index after fusion.

[0178] On the premise of meeting task requirements and specifications, the larger the objective function value, the higher the surveying and mapping accuracy.

[0179] The above technical solution constructs sub-objective functions related to the situation deduction of the surveying and mapping area from four aspects: surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost, and surveying and mapping accuracy.

[0180] Each sub-objective function establishes a quantitative relationship between the optimization goal and the influencing factors, and gives the corresponding solution method and optimization strategy.

[0181] The four sub-objective functions coordinate with each other and jointly support the establishment and solution of the multi-objective optimization model, providing a scientific basis for surveying and mapping task decision-making.

[0182] In an optional implementation, in the three-dimensional geographic scene model, determining an optimal surveying and mapping equipment layout solution set for a surveying and mapping area based on a simulation optimization algorithm includes: In the three-dimensional geographic scene model, multiple surveying and mapping equipment layout plans are initialized according to the geographic environment characteristics of the surveying and mapping area and the surveying and mapping task requirements. Each surveying and mapping equipment layout plan is composed of different types and quantities of surveying and mapping equipment, wherein the surveying and mapping equipment includes drones, lidars, and panoramic cameras. Each device is composed of three parts: sensor parameters, platform parameters, and environmental parameters. The constructed surveying and mapping equipment simulation model is used to simulate the operation process of each surveying and mapping equipment layout plan, and the surveying and mapping data collection volume, resolution, and measurement error indicators under different surveying and mapping equipment layout plans are obtained through sensor data solution, platform trajectory planning, and noise interference analysis. Each indicator is matched with the corresponding surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost, and surveying and mapping accuracy sub-objective functions, and the sub-objective satisfaction value of each surveying and mapping equipment layout plan is calculated in the form of fuzzy membership. Each surveying and mapping equipment layout plan is represented as a particle, and the position vector of the particle is composed of the type, quantity, and The particle speed vector represents the adjustment direction and amplitude of the layout scheme; the fitness function in the form of weighted sum is constructed with the sub-goal satisfaction value, the weight coefficient reflects the importance of different sub-goals, and the optimization goal is to maximize the fitness function value; the entropy weight method is used to determine the initial weight of each sub-goal, and the entropy value is calculated according to the discrete degree of the sub-goal satisfaction value, and the weight is normalized using the inverse of the entropy value; in the particle swarm optimization iteration process, the external archive is updated according to the fitness value of the current particle, and the crowding distance of the particles in the archive is calculated, the particle with the largest crowding distance is taken as the global optimal particle, and the particle with the smallest crowding distance is deleted from the archive; when the particle speed and position are updated, the search is performed according to the particle's own historical optimal position, the position of the global optimal particle and the crowding distance; through multiple rounds of particle position iterative update, a set of optimal solutions consisting of non-dominated solutions is obtained as the optimal surveying and mapping equipment layout scheme set for the surveying and mapping area.

[0183] Exemplarily, in a three-dimensional geographic scene model, multiple surveying and mapping equipment layout plans to be evaluated are initialized and generated based on the geographical environment characteristics of the surveying and mapping area (such as terrain, landforms, objects, etc.) and the surveying and mapping task requirements (such as surveying and mapping scope, surveying and mapping accuracy, surveying and mapping time limit, etc.).

[0184] Each layout plan consists of different types and quantities of surveying and mapping equipment to meet the needs of the surveying and mapping tasks.

[0185] Surveying and mapping equipment mainly includes three categories: drones, lidars, and panoramic cameras. Drones are used to obtain aerial remote sensing images of the surveying and mapping area, lidars are used to obtain three-dimensional point cloud data of the surveying and mapping area, and panoramic cameras are used to obtain panoramic real-life images of the surveying and mapping area.

[0186] Each surveying and mapping device consists of three parts: sensor parameters, platform parameters and environmental parameters.

[0187] Sensor parameters reflect the performance indicators of the equipment, such as resolution, accuracy, field of view, etc.; platform parameters reflect the maneuverability of the equipment, such as speed, range, load, etc.; environmental parameters reflect the adaptability of the equipment, such as wind speed, temperature, light, etc.

[0188] For each type of surveying and mapping equipment, a corresponding mathematical simulation model is constructed. The input of the model is the equipment parameters and environmental parameters, and the output is the quantitative indicators of the equipment working process.

[0189] The simulation model comprehensively considers the physical characteristics, working mechanism and environmental impact of the equipment, and can simulate the operation process of the equipment more realistically.

[0190] The constructed surveying and mapping equipment simulation model is used to simulate the operation process of each surveying and mapping equipment layout plan.

[0191] The operation process mainly includes three stages: data collection, transmission and processing of surveying and mapping data. Among them, data collection is realized through sensor data solution, such as the geometric optical model of camera imaging, the flight time method of laser radar ranging, etc.; data transmission is realized through platform trajectory planning, such as drone route planning, laser radar scanning mode, etc.; data processing is realized through noise interference analysis, such as signal distortion correction, error compensation, etc.

[0192] Through simulation, the surveying and mapping data collection volume, resolution, measurement error and other operational indicators under different surveying and mapping equipment layout schemes are obtained.

[0193] The collection volume reflects the ability to obtain data per unit time, the resolution reflects the ability to express spatial details of the data, and the measurement error reflects the accuracy of the data.

[0194] All indicators can be quantitatively calculated to provide a basis for subsequent program evaluation.

[0195] The operating indicators of each layout plan are matched with the corresponding four sub-objective functions of surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost and surveying and mapping accuracy.

[0196] The sub-objective functions reflect the different requirements of surveying and mapping tasks, among which the coverage function describes the completeness of the surveying and mapping scope, the efficiency function describes the speed of the surveying and mapping progress, the cost function describes the level of surveying and mapping investment, and the accuracy function describes the quality of surveying and mapping.

[0197] The satisfaction of each surveying and mapping equipment layout plan with respect to the four sub-goals is calculated in the form of fuzzy membership.

[0198] Fuzzy membership is the degree to which a solution belongs to a sub-goal optimal solution set, and its value is between 0 and 1. The larger the membership, the more the solution meets the requirements of the sub-goal.

[0199] The determination of the membership function requires comprehensive consideration of factors such as the size of the indicator value, its changing trend, and its ideal interval.

[0200] Based on the sub-goal satisfaction, the fitness function of the surveying and mapping equipment layout plan is constructed.

[0201] The fitness function takes the form of weighted summation, which linearly combines the satisfaction of the four sub-goals, and the weight coefficient reflects the importance of different sub-goals.

[0202] The optimization goal is to maximize the fitness function value, that is, to find the layout plan with the best comprehensive performance while taking into account multiple sub-goals.

[0203] The entropy weight method is used to determine the weight coefficient of each sub-goal.

[0204] The entropy weight method is an objective weighting method that measures the importance of indicators according to the degree of dispersion of indicator values.

[0205] The greater the degree of dispersion, the higher the discrimination of the indicator, the greater its effect on program evaluation, and the greater the weight coefficient.

[0206] First, according to the sub-goal satisfaction of each plan, the entropy value of each sub-goal is calculated.

[0207] The entropy value reflects the degree of disorder of sub-goal satisfaction. The larger the entropy value, the more uniform the distribution of satisfaction and the lower the discrimination.

[0208] Then, the weights are normalized using the inverse of the entropy value to obtain the initial weight coefficient.

[0209] During the optimization iteration process, the weight coefficient can be dynamically adjusted according to the solution needs.

[0210] Each mapping equipment layout plan is represented as a particle.

[0211] The particle position vector is composed of the type, quantity and working parameters of each surveying and mapping equipment, reflecting the configuration of the scheme.

[0212] The particle's velocity vector represents the adjustment direction and amplitude of the deployment plan, guiding the particle to search in the solution space.

[0213] During the particle swarm optimization iteration process, the external archive is updated according to the fitness value of the current particle.

[0214] The external archive is used to save the non-dominated solutions found so far, that is, solutions that are not dominated by any other solution.

[0215] The non-dominated solution is not inferior to other solutions in each sub-objective and is the optimal compromise for multi-objective optimization.

[0216] At the same time, the crowding distance of the particles in the archive is calculated. The crowding distance indicates the density of a particle with other particles in the target space. The larger the distance, the sparser the solution in the area where the particle is located, and the greater the contribution to the diversity of the solution set.

[0217] The particle with the largest crowding distance is taken as the global optimal particle, and the particle with the smallest crowding distance is deleted from the archive to maintain the archive size within a reasonable range.

[0218] When updating the particle speed and position, the particle's own historical optimal position, the position of the global optimal particle, and the crowding distance are comprehensively considered.

[0219] The historical optimal position indicates the optimal solution found by the particle in the past search, the position of the global optimal particle indicates the optimal solution in the current archive, and the crowding distance indicates the distribution of the solution.

[0220] All three provide useful reference information for particle motion.

[0221] By setting the calculation formula for speed and position update, the particles can continuously iteratively search in the solution space and gradually approach the global optimal solution set.

[0222] The update formula includes parameters such as learning factor, inertia weight, and constraint conditions, which are used to balance the global exploration ability and local development ability of particles and improve the convergence speed and accuracy of the algorithm.

[0223] Through multiple rounds of iterative updates of particle positions, the performance of the surveying and mapping equipment deployment plan is continuously improved.

[0224] The algorithm terminates when the number of iterations reaches the preset upper limit or the fitness value of the particles in the archive tends to be stable.

[0225] At this point, the set of non-dominated solutions saved in the external archive is the optimal surveying and mapping equipment layout plan set for the surveying and mapping area.

[0226] This solution comprehensively weighs multiple objectives such as surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost and surveying and mapping accuracy, and can provide diverse decision-making references for actual surveying and mapping projects.

[0227] Decision makers can select the most appropriate layout plan from the plan set according to actual conditions and implement it, so as to achieve the purpose of multi-objective optimization and maximization of comprehensive benefits.

[0228] In an optional implementation, multi-temporal remote sensing image data and geographic national survey data are integrated, and a long short-term memory neural network algorithm is used to predict the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future based on the current three-dimensional situation information, including: performing radiation calibration and geometric correction on high-resolution visible light images and synthetic aperture radar image data acquired at multiple time points, using a scale-invariant feature transformation algorithm to perform registration between remote sensing image data of different temporal phases, and fusing multi-temporal remote sensing image data into a unified time series image sequence through principal component analysis or wavelet transform; using a semantic segmentation algorithm based on a convolutional neural network to extract land feature elements from land use status maps, topographic maps, and place name and address maps in the geographic national survey data; The multi-level semantic features of the image are extracted through multi-scale convolution kernels, and the conditional random field model is used to realize pixel-level object classification. At the same time, the topological relationship and spatial distribution law between the ground object elements are learned by using a bidirectional recursive neural network, and the interpreted ground object semantic information is spatially associated with the time series image sequence to construct a multi-temporal three-dimensional ground object distribution model. A three-dimensional situation evolution prediction model based on a bidirectional long short-term memory neural network is constructed, and the three-dimensional situation information of each time point is represented as a multi-dimensional feature vector including object type, geometric shape, spatial position and semantic attributes. The historical three-dimensional situation feature vector sequence is input into the three-dimensional situation evolution prediction model, and the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future is iteratively predicted.

[0229] Exemplarily, the present invention proposes a method that integrates multi-temporal remote sensing image data and geographic national conditions census data, adopts a long short-term memory neural network algorithm, and predicts the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future based on the current three-dimensional situation information.

[0230] Image data of the survey area are obtained at multiple time points through high-resolution visible light remote sensing platforms and synthetic aperture radar systems.

[0231] Visible light images can truly reflect the surface information of objects such as color and texture, while synthetic aperture radar images can penetrate clouds, fog, vegetation, etc. to obtain geometric structure information of objects. The two types of data complement each other and help to fully depict the surface conditions of the surveyed area.

[0232] At the same time, geographical national conditions census data related to the surveying and mapping area are collected, mainly including land use status maps, topographic maps and place name and address maps.

[0233] The current land use map reflects the type and scope of land resource utilization, the topographic map reflects the undulating shape and elevation information of the surface, and the place name and address map reflects the spatial distribution and attribute information of settlements.

[0234] The three types of data are interrelated and together constitute the basic geographic information of the surveying and mapping area.

[0235] The acquired multi-temporal remote sensing image data are subjected to radiation calibration and geometric correction to eliminate radiation distortion and geometric distortion in the imaging process.

[0236] Radiometric calibration converts image grayscale into physical reflectivity by establishing a quantitative relationship between sensor response and ground object radiance, thereby improving the radiometric consistency of the image.

[0237] Geometric correction corrects the image geometry to a unified spatial reference system by establishing a mapping relationship between image coordinates and ground coordinates, thereby improving the position accuracy of the image.

[0238] The scale-invariant feature transformation algorithm is used to align remote sensing image data of different phases.

[0239] The algorithm extracts scale-invariant feature points in images, such as corner points and spots, and establishes corresponding relationships between feature points to achieve automatic spatial matching of images of different phases.

[0240] The time-series images after registration completely overlap in geometry, laying the foundation for subsequent time-series analysis.

[0241] Multi-temporal remote sensing image data are fused into a unified time series image sequence through principal component analysis or wavelet transform.

[0242] The principal component analysis method projects the image data of multiple phases into a new orthogonal feature space through linear transformation, and extracts the first few principal components with the largest contribution rate to achieve data dimensionality reduction and feature enhancement.

[0243] The wavelet transform method achieves scale isomorphism and edge preservation of data by performing multi-scale and multi-directional wavelet decomposition on image data and fusing high-frequency details and low-frequency contours at different scales.

[0244] The fused time-series images achieve integrated expression in spectrum, space and time, facilitating multi-dimensional information mining.

[0245] A semantic segmentation algorithm based on convolutional neural network is used to extract geographical features and annotate their attributes in the national geographic census data.

[0246] Convolutional neural network is a deep learning model that simulates the human brain's visual system. It automatically extracts multi-level semantic features of images through layer-by-layer convolution and pooling operations.

[0247] During the extraction process, the algorithm's adaptability to complex objects is enhanced by setting convolution kernels of different scales and directions.

[0248] Based on feature extraction, the conditional random field model is used to achieve pixel-level object classification.

[0249] Conditional random field is a probabilistic graph model that establishes Markov attribute dependencies between pixels and considers the spatial correlation between pixels and surrounding areas to ultimately obtain smooth and continuous land object classification results.

[0250] The classification process makes full use of the multi-dimensional characteristics of the objects such as spectrum, texture, shape and spatial context information, which greatly improves the accuracy of object recognition.

[0251] While performing semantic segmentation, a bidirectional recursive neural network is used to learn the topological relationship and spatial distribution patterns between ground features.

[0252] Recursive neural network is a deep learning model suitable for processing sequence data. It can model long-term dependencies between data by recursively transferring hidden layer states in the time dimension.

[0253] In the spatial dimension, the spatial relationships such as adjacency and inclusion between land features are modeled through a two-way transmission mechanism to characterize the spatial combination patterns of different land features.

[0254] The semantic information of land objects obtained by interpreting the geographical national conditions is spatially associated with the time-series remote sensing image sequence to construct a multi-temporal three-dimensional land object distribution model.

[0255] Spatial association is achieved through coordinate transformation and attribute linking between image and map data, mapping two-dimensional map elements to three-dimensional image space and inheriting corresponding semantic attributes.

[0256] In the association process, a fast indexing method based on geometric hashing is used to reduce the computational complexity of spatial search.

[0257] The three-dimensional feature distribution model is organized using spatial index structures such as quadtrees and octrees. Each leaf node stores attribute information such as the type, boundary, elevation, and topology of the features in the area.

[0258] Based on the attribute information, the geometric shape of the object is reconstructed in three dimensions through triangular patches, voxels, etc., generating a multi-level, multi-scale three-dimensional model of the object.

[0259] The multi-temporal feature distribution model comprehensively records the distribution status of features in different periods, and reveals the temporal and spatial evolution of features through time series analysis, providing an important basis for the prediction of future trends.

[0260] Construct a three-dimensional situation evolution prediction model based on a bidirectional long short-term memory neural network.

[0261] Long short-term memory neural network is an improved variant of recurrent neural network. By introducing gating units and memory units, it can better capture long-term dependencies in time series and alleviate problems such as gradient disappearance and gradient explosion.

[0262] In the prediction model, the three-dimensional situation information at each time point is represented as a multi-dimensional feature vector including the type of feature, geometric shape, spatial position and semantic attributes.

[0263] Among them, the type of feature reflects the nature and function of the feature, the geometric shape reflects the size and appearance of the feature, the spatial position reflects the latitude, longitude and elevation of the feature, and the semantic attributes reflect the business meaning and application value of the feature.

[0264] The feature vector quantifies multiple attribute dimensions of the features in a normalized form, making it easier to express and calculate in the model.

[0265] Through the sliding window method, the historical three-dimensional situation feature vector sequence is input into the bidirectional long short-term memory neural network in sequence.

[0266] The forward network and the backward network extract features from two directions of the time series respectively, and obtain the encoded representation of the historical situation in the hidden layer state of the last time step.

[0267] The encoded representation is mapped to the feature space of the future situation through a fully connected layer, and the sustained softmax function is used to obtain the probability distribution of the three-dimensional situation evolution in each future time period.

[0268] A large amount of historical three-dimensional situation data is used to train the prediction model, and the model parameters are optimized through supervised learning.

[0269] The historical data is divided into samples of fixed length according to the time series, each sample includes the input historical situation sequence and the output future situation sequence.

[0270] The cross entropy loss function is used to measure the difference between the predicted situation and the actual situation, and the stochastic gradient descent method is used to update the parameters.

[0271] During the training process, the generalization ability and robustness of the model are improved by setting strategies such as dropout and L2 regularization.

[0272] After the model training is completed, the currently acquired three-dimensional situation information is input into the prediction model to iteratively predict the three-dimensional situation evolution of the surveying area in different time periods in the future.

[0273] The prediction results include the change trends and probability distribution of the type, geometry, spatial location and semantic attributes of local features in each time period.

[0274] Through 3D visualization engines such as trois.js and unity3D, the prediction results are intuitively displayed in the form of dynamic scenes, revealing the future distribution pattern of land objects, spatial topological relationships, development trends, etc. in the surveying and mapping area, providing auxiliary decision-making support for applications such as urban planning, resource management, and disaster prevention and control.

[0275] The present invention utilizes multi-source spatiotemporal big data and intelligent technologies such as deep learning and three-dimensional reconstruction to achieve dynamic monitoring of complex geographical environments and accurate prediction of future trends.

[0276] The prediction results are intuitive and reliable, which greatly expands the spatiotemporal analysis capabilities of basic surveying and mapping.

[0277] In an optional embodiment, a three-dimensional situation evolution prediction model based on a bidirectional long short-term memory neural network is constructed, including: using a normalization or standardization method to perform feature scaling on the continuous variables in the multidimensional feature vector, and using a unique hot encoding or embedding method to perform feature conversion on the discrete variables in the multidimensional feature vector to obtain a preprocessed three-dimensional situation feature vector; determining the number of input layer nodes of the bidirectional long short-term memory neural network, the number of hidden units of the bidirectional long short-term memory layer, and the number of stacked layers according to the dimension of the preprocessed three-dimensional situation feature vector; in the bidirectional long short-term memory layer, using an input gating unit, a forgetting gating unit, and an output gating unit to output key feature information in the preprocessed three-dimensional situation feature vector; storing the key feature information in a memory unit, and concatenating the forward hidden state and the backward hidden state of the bidirectional long short-term memory layer at each time step to form a context feature representation; in the attention layer, calculating the importance weights of the context feature representations at different historical moments in parallel through multiple attention heads to determine the importance weights related to the current Predict the most relevant historical context features, calculate the similarity between the query vector and the key-value vector using dot product attention or additive attention, generate an attention weight distribution, perform weighted summation of the attention weight distribution and the corresponding value vector, and obtain an aggregated attention context feature vector; in the fully connected layer, concatenate the context feature representation output by the bidirectional long short-term memory layer and the attention context feature vector output by the attention layer into a complete feature representation vector, and perform nonlinear transformation and dimensional conversion on the complete feature representation vector through one or more fully connected layers to extract high-level semantic feature information; according to the type of the three-dimensional situation evolution prediction task, select an activation function to activate the high-level semantic feature information; in the output layer, according to the type of the three-dimensional situation evolution prediction task, use the activated high-level semantic feature information output by the fully connected layer to generate an output feature vector representing the future three-dimensional situation state, add an additional decoder structure to the output layer, and construct a three-dimensional situation evolution prediction model based on a bidirectional long short-term memory neural network.

[0278] Exemplarily, the present invention proposes a method for constructing a three-dimensional situation evolution prediction model based on a bidirectional long short-term memory neural network.

[0279] This method utilizes the advantages of bidirectional long short-term memory neural network in processing time series data, combines the attention mechanism and decoder structure, and can effectively capture the evolution law of historical three-dimensional situation and accurately predict the three-dimensional situation status in different time periods in the future.

[0280] The specific technical solutions are as follows: Normalization or standardization methods are used to perform feature scaling on continuous variables in multidimensional feature vectors.

[0281] Normalization eliminates the dimension effect by mapping the value range of continuous variables to the interval [0,1], making features of different scales comparable.

[0282] Standardization converts the feature distribution into a standard normal distribution with a mean of 0 and a variance of 1 by subtracting the mean of the continuous variable and dividing it by the standard deviation, thereby improving the convergence speed and stability of the model.

[0283] The discrete variables in the multidimensional feature vector are transformed using one-hot encoding or embedding methods.

[0284] One-hot encoding maps each value of a discrete variable into a binary vector, in which only one element is 1 and the rest are 0, encoding the category information of the discrete variable.

[0285] Embedding maps discrete variables into a low-dimensional dense vector. Through the learned embedding matrix, the discrete variables are converted into continuous vector representations, revealing the semantic correlation between different categories.

[0286] After normalization / standardization and one-hot encoding / embedding processing, the preprocessed three-dimensional situation feature vector is obtained. The preprocessed feature vector has a unified scale and format, which is convenient for expression and calculation in neural networks.

[0287] According to the dimension of the preprocessed three-dimensional situation feature vector, the number of input layer nodes of the bidirectional long short-term memory neural network, the number of hidden units in the bidirectional long short-term memory layer and the number of stacked layers are determined.

[0288] The number of nodes in the input layer is consistent with the dimension of the feature vector and is used to receive input data.

[0289] The number of hidden units in the bidirectional long short-term memory layer determines the feature extraction capability of the network and is usually set to 2 to 4 times the input dimension.

[0290] The number of stacked layers determines the depth and abstraction level of the network. By increasing the number of stacked layers, a more complex feature hierarchy can be constructed.

[0291] In the bidirectional long short-term memory layer, the input gating unit, the forgetting gating unit and the output gating unit are used to output the key feature information in the preprocessed three-dimensional situation feature vector.

[0292] Among them, the input gating unit determines which information in the current input features needs to be updated to the memory unit, the forgetting gating unit determines which information in the historical memory unit needs to be forgotten, and the output gating unit determines which information in the memory unit needs to be output to the hidden state.

[0293] The gating unit uses the sigmoid activation function to achieve soft switching between 0 and 1, precisely controlling the flow of information.

[0294] Storing key feature information in memory units enables long-term memory and transmission of historical information.

[0295] The memory unit realizes the cyclic accumulation of information through a linear self-loop structure, and realizes the selective forgetting and updating of memory through the gating unit.

[0296] The bidirectional LSTM layer recursively transfers hidden states in both the forward and backward directions, capturing the forward and backward dependencies of the time series, respectively.

[0297] The forward hidden state and backward hidden state of the bidirectional long short-term memory layer at each time step are concatenated to form a complete contextual feature representation.

[0298] In the attention layer, multiple attention heads are used to calculate the importance weights of contextual feature representations at different historical moments in parallel to determine the historical contextual features most relevant to the current prediction.

[0299] The number of attention heads determines the parallelism and diversity of attention calculations, and is usually set to 4 to 8. Each attention head independently calculates the attention weight distribution and captures different aspects of correlation information.

[0300] The dot product attention or additive attention method is used to calculate the similarity between the query vector and the key-value vector to generate the attention weight distribution.

[0301] Among them, the query vector represents the hidden state at the current moment, and the key-value vector represents the hidden state at the historical moment.

[0302] The dot product attention obtains the normalized attention weight by calculating the inner product of the query vector and the key-value vector and dividing it by the square root of the key-value vector dimension.

[0303] Additive attention concatenates the query vector and the key-value vector and passes them through a fully connected layer and a tanh activation function to obtain the attention weight after nonlinear transformation.

[0304] The attention weight distribution is weighted and summed with the corresponding value vector to obtain the aggregated attention context feature vector.

[0305] The attention weight distribution reflects the contribution of different historical moments to the current prediction, and the value vector represents the hidden state of the historical moment.

[0306] Through weighted summation, historical context features are aggregated according to importance to obtain a context representation that is closely related to the current prediction.

[0307] In the fully connected layer, the contextual feature representation output by the bidirectional long short-term memory layer and the attention context feature vector output by the attention layer are concatenated into a complete feature representation vector.

[0308] The concatenation operation combines the two features to obtain a feature representation that takes into account both global dependency and local importance.

[0309] The complete feature representation vector is transformed nonlinearly and dimensionally through one or more fully connected layers to extract high-level semantic feature information.

[0310] The fully connected layer maps the input features to a higher-dimensional feature space through matrix multiplication and nonlinear activation functions, and mines the nonlinear interactions between features.

[0311] The number of layers and nodes in the fully connected layer is set according to the complexity of the task and the scale of the data. A pyramid structure is usually adopted to reduce the number of nodes layer by layer.

[0312] According to the type of three-dimensional situation evolution prediction task, select the appropriate activation function to activate the high-level semantic feature information.

[0313] For classification tasks, the softmax function is usually used to convert features into probability distribution.

[0314] For regression tasks, linear functions or sigmoid functions are usually used to convert features into continuous values.

[0315] The choice of activation function determines the form and interpretation of the output layer.

[0316] In the output layer, according to the type of three-dimensional situation evolution prediction task, the activated high-level semantic feature information output by the fully connected layer is used to generate an output feature vector representing the future three-dimensional situation state.

[0317] The dimension and meaning of the output feature vector correspond to the definition of the task and represent the model's prediction results for the future three-dimensional situation.

[0318] An additional decoder structure is added to the output layer to further process and interpret the output feature vector.

[0319] The decoder structure can be customized according to the requirements of the task, such as using a recurrent neural network to generate the output sequence and a convolutional neural network to reconstruct the output image.

[0320] The introduction of the decoder structure enhances the expressiveness of the model and makes the prediction results more accurate and explainable.

[0321] The constructed three-dimensional situation evolution prediction model based on bidirectional long short-term memory neural network is applied to practical tasks.

[0322] The model is trained using historical three-dimensional situation data, and the model parameters are optimized through supervised learning.

[0323] The training data includes historical three-dimensional situation feature vector sequences and corresponding future three-dimensional situation state labels.

[0324] During the training process, the cross entropy loss function or the mean square error loss function is used to measure the difference between the predicted results and the true labels, and the back propagation algorithm and gradient descent optimization method are used to minimize the loss function and update the model parameters.

[0325] At the same time, strategies such as dropout regularization and early stopping are used to control the complexity of the model and improve the generalization ability of the model.

[0326] After the model training is completed, the currently acquired three-dimensional situation feature vector sequence is input into the model. After the forward propagation process, the prediction result representing the future three-dimensional situation state is obtained in the output layer.

[0327] The prediction results can be the three-dimensional situation classification probability distribution in different time periods in the future, or the three-dimensional situation regression value in different time periods in the future.

[0328] Figure 2 FIG. 1 is a schematic diagram of the structure of a surveying and mapping area situation deduction system under multi-objective optimization according to an embodiment of the present invention. Figure 2 As shown, the system comprises: The first unit is used to obtain multi-source heterogeneous geographic information data of the surveying and mapping area, and the multi-source heterogeneous geographic information data includes remote sensing image data collected by different types of sensors, lidar point cloud data and geographic national conditions census data; a deep learning algorithm is used to extract features and semantic segment the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying and mapping area, and the three-dimensional geographic scene model includes a high-precision terrain model, a semantic building model and a multi-level road network model; the second unit is used to construct a multi-objective optimization model for situation deduction of the surveying and mapping area, and the multi-objective optimization model takes surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost and surveying and mapping accuracy as optimization objectives, including multiple sub-objective functions, and the weight of each sub-objective function is determined by hierarchical analysis method; in the three-dimensional geographic scene model, the optimal surveying and mapping equipment layout of the surveying and mapping area is determined based on the simulation optimization algorithm. a case set; based on each solution in the optimal surveying and mapping equipment layout solution set, an optimal surveying and mapping equipment layout solution is obtained; in the three-dimensional geographic scene model, the surveying and mapping operation process under the optimal surveying and mapping equipment layout solution is simulated, three-dimensional point cloud data consistent with the surveying and mapping operation process is generated, and the current three-dimensional situation information of the surveying and mapping area is obtained; a third unit is used to fuse multi-temporal remote sensing image data and geographic national conditions census data, and use a long short-term memory neural network algorithm to predict the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future based on the current three-dimensional situation information; the three-dimensional situation evolution information is input into the multi-objective optimization model, the fitness value of each surveying and mapping equipment layout solution is recalculated, and the optimal surveying and mapping equipment layout solution is updated; based on the updated optimal surveying and mapping equipment layout solution, virtual reality technology is used to generate three-dimensional visualization scenes of the surveying and mapping area in different time periods in the future.

[0329] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, including: A processor; a memory for storing processor-executable instructions; wherein the processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0330] A fourth aspect of the embodiments of the present invention is: A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0331] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0332] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for deducing the situation of a surveying and mapping area under multi-objective optimization, characterized in that: include: Acquire multi-source heterogeneous geographic information data of the surveying and mapping area, wherein the multi-source heterogeneous geographic information data includes remote sensing image data collected by different types of sensors, laser radar point cloud data, and geographic national conditions census data; use a deep learning algorithm to perform feature extraction and semantic segmentation on the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying and mapping area, wherein the three-dimensional geographic scene model includes a high-precision terrain model, a semantic building model, and a multi-level road network model; A multi-objective optimization model for mapping area situation deduction is constructed, wherein the multi-objective optimization model takes mapping coverage, mapping efficiency, mapping cost and mapping accuracy as optimization objectives, and includes multiple sub-objective functions, and the weight of each sub-objective function is determined by using a hierarchical analysis method; in the three-dimensional geographic scene model, an optimal mapping equipment layout scheme set for the mapping area is determined based on a simulation optimization algorithm; based on each scheme in the optimal mapping equipment layout scheme set, an optimal mapping equipment layout scheme is obtained; In the three-dimensional geographic scene model, the surveying and mapping operation process under the optimal surveying and mapping equipment layout scheme is simulated to generate three-dimensional point cloud data consistent with the surveying and mapping operation process, and obtain the current three-dimensional situation information of the surveying and mapping area; By integrating multi-temporal remote sensing image data and geographic national survey data, and using a long short-term memory neural network algorithm, the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future is predicted based on the current three-dimensional situation information; the three-dimensional situation evolution information is input into the multi-objective optimization model, the fitness value of each surveying and mapping equipment layout plan is recalculated, and the optimal surveying and mapping equipment layout plan is updated; based on the updated optimal surveying and mapping equipment layout plan, virtual reality technology is used to generate three-dimensional visualization scenes of the surveying and mapping area in different time periods in the future.

2. The method according to claim 1, characterized in that A deep learning algorithm is used to perform feature extraction and semantic segmentation on the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying area, including: The remote sensing image data is processed by joint radiometric calibration, multi-view geometric correction and multi-scale image fusion to generate a multi-source remote sensing image dataset with consistent spectral response and spatial geometric matching; a convolutional neural network algorithm based on an attention mechanism is used to extract the texture, spectral and spatial features of the multi-source remote sensing image dataset to generate a multi-channel remote sensing image feature map; the lidar point cloud data is processed by adaptive noise filtering, multi-scale ground point extraction and point cloud segmentation based on graph cuts to obtain lidar point cloud subsets of different types of ground objects; a deep learning algorithm based on a graph attention network is used to extract the geometric features, topological relationship features and semantic attribute features of the lidar point cloud subset to generate a multi-dimensional point cloud feature vector; The geographic national conditions census data is processed by spatial coordinate conversion, attribute data standardization and multi-level topological reconstruction to generate a geographic national conditions element set that is consistent with the spatial reference of the multi-source remote sensing image data set and the lidar point cloud subset; a deep learning algorithm based on a gated graph convolutional network is used to extract the semantic features, spatial relationship features and hierarchical structure features of the geographic national conditions element set to generate a multi-resolution semantic feature map; a feature adaptive fusion strategy based on a self-attention mechanism is used to nonlinearly fuse the multi-channel dynamic remote sensing image feature map, the multi-dimensional point cloud feature vector and the multi-resolution semantic feature map to generate a fused multi-source heterogeneous geographic information feature set; The fused multi-source heterogeneous geographic information feature set is input into a multi-task deep learning model to generate a high-precision terrain model, a semantic building model and a multi-level road network model of the surveying area; a geographic information semantic ontology library is constructed, and the high-precision terrain model, the semantic building model and the multi-level road network model are semantically mapped and knowledge associated with the geographic information semantic ontology library to obtain a three-dimensional geographic scene model of the surveying area.

3. The method according to claim 1, characterized in that Construct a multi-objective optimization model for mapping area situation deduction, including: The surveying and mapping target data, surveying and mapping accuracy data, surveying and mapping time data and surveying and mapping cost data are used as the input of the multi-objective optimization model to establish the surveying and mapping coverage rate sub-objective function, the surveying and mapping efficiency sub-objective function, the surveying and mapping cost sub-objective function and the surveying and mapping accuracy sub-objective function; among them, the independent variable of the surveying and mapping coverage rate sub-objective function is the surveying and mapping status of each grid, and the output is the surveying and mapping coverage rate evaluation index; the independent variable of the surveying and mapping accuracy sub-objective function is the accuracy parameters of each surveying and mapping sensor and the parameters of the fusion algorithm, and the output is the surveying and mapping accuracy evaluation index; the independent variable of the surveying and mapping cost sub-objective function is the input amount of each cost factor, and the output is the surveying and mapping cost evaluation index; the independent variable of the surveying and mapping efficiency sub-objective function is the type and quantity of surveying and mapping equipment selected for each work unit, and the output is the surveying and mapping efficiency evaluation index; The dynamic preference information of surveying and mapping task requirements is collected, and a pairwise comparison matrix between different optimization objectives of the dynamic preference information is constructed to obtain a dynamic judgment matrix, wherein the element values ​​of the dynamic judgment matrix represent the dynamic quantitative evaluation values ​​of the relative importance of the two optimization objectives; the maximum eigenvalue of the dynamic judgment matrix and its corresponding eigenvector are calculated by the spectral decomposition method, and a fuzzy consistency evaluation model of the dynamic judgment matrix is ​​constructed; the dynamic judgment matrix is ​​solved based on the fuzzy consistency evaluation model to obtain the maximum fuzzy eigenvalue and fuzzy eigenvector that are compatible with the dynamic preference information; the solved fuzzy eigenvector is normalized to obtain an initial weight vector, and a subjective and objective combined weighting model is constructed to optimize and correct the initial weight vector to obtain a combined weight vector; The fuzzy consistency ratio of the dynamic judgment matrix is ​​calculated, and whether the matrix conforms to the logical consistency is judged by setting the fuzzy consistency ratio threshold; when the test fails, the feedback mechanism is used to adjust the element values ​​of the judgment matrix, and the combined weight vector is recalculated and a consistency test is performed; when the test passes, the combined weight vector is used as the dynamic weighting coefficient of the surveying and mapping coverage sub-objective function, the surveying and mapping efficiency sub-objective function, the surveying and mapping cost sub-objective function and the surveying and mapping accuracy sub-objective function to construct the optimization target of the multi-objective optimization model.

4. The method according to claim 3, characterized in that ,Establish the sub-objective function of surveying and mapping coverage, the sub-objective function of surveying and mapping efficiency, the sub-objective function of surveying and mapping cost and the sub-objective function of surveying and mapping accuracy, including: The area to be measured is divided into grids of equal size by using a rasterization method. The side length of each grid is adaptively determined according to the geometric characteristics of the surveying and mapping target and the required spatial resolution. The surveying and mapping status of each grid is statistically analyzed to form a surveying and mapping status matrix. The key ground object elements in the area to be measured are extracted from remote sensing images or laser point cloud data using ground object element extraction technology. The surveying and mapping completeness rate of each type of ground object element is calculated. The sub-objective function of surveying and mapping coverage rate is constructed with the constraint condition that the surveying and mapping completeness rate of ground object elements is greater than a given threshold. By adopting the hierarchical decomposition method, the surveying and mapping tasks are decomposed into multiple work units from top to bottom. The standard working hours required for each work unit are determined according to the workload and complexity of the work. At the same time, the completion efficiency of the unit workload is calculated based on the equipment performance parameters and work experience for different surveying and mapping equipment, and the surveying and mapping efficiency sub-objective function is constructed with the surveying and mapping time data as the constraint condition; by adopting the cost accounting method, various cost elements related to the surveying and mapping tasks are systematically classified and aggregated, and cost estimation models are selected for different types of cost elements. The total cost calculation model is constructed to obtain the budget cost for completing the entire surveying and mapping task, and the cost sensitivity analysis technology is used to identify the key cost driving factors, and the surveying and mapping cost sub-objective function is constructed with the surveying and mapping cost data as the constraint condition; Taking the accuracy parameters of different surveying and mapping sensors as optimization variables, the functional relationship between surveying and mapping accuracy evaluation indicators and sensor accuracy parameters is established. The differential evolution algorithm is used to search for the optimal combination of surveying and mapping sensor accuracy parameters. After obtaining the optimal accuracy parameters of each surveying and mapping sensor, the multi-source data fusion technology is used to fuse the surveying and mapping data obtained by different sensors. The surveying and mapping accuracy sub-objective function is constructed with the surveying and mapping accuracy data as the constraint condition.

5. The method according to claim 1, characterized in that In the three-dimensional geographic scene model, the optimal surveying and mapping equipment layout solution set of the surveying and mapping area is determined based on the simulation optimization algorithm, including: In the three-dimensional geographic scene model, according to the geographic environment characteristics of the surveying and mapping area and the requirements of the surveying and mapping task, multiple surveying and mapping equipment layout plans are initialized, each surveying and mapping equipment layout plan is composed of different types and quantities of surveying and mapping equipment, wherein the surveying and mapping equipment includes drones, lidars, and panoramic cameras, and each device is composed of three parts: sensor parameters, platform parameters, and environmental parameters; the constructed surveying and mapping equipment simulation model is used to simulate the operation process of each surveying and mapping equipment layout plan, and the surveying and mapping data collection volume, resolution, and measurement error indicators under different surveying and mapping equipment layout plans are obtained through sensor data solution, platform trajectory planning, and noise interference analysis; each indicator is matched with the corresponding surveying and mapping coverage, surveying and mapping efficiency, surveying and mapping cost, and surveying and mapping accuracy sub-objective functions, and the sub-objective satisfaction value of each surveying and mapping equipment layout plan is calculated in the form of fuzzy membership; Each surveying and mapping equipment layout plan is represented as a particle. The particle's position vector is composed of the type, quantity and working parameters of each surveying and mapping equipment. The particle's velocity vector represents the adjustment direction and amplitude of the layout plan. A fitness function in the form of weighted summation is constructed based on the sub-goal satisfaction values. The weight coefficient reflects the importance of different sub-goals. The optimization goal is to maximize the fitness function value. The entropy weight method is used to determine the initial weight of each sub-goal. The entropy value is calculated according to the discrete degree of the sub-goal satisfaction value, and the weight is normalized using the inverse of the entropy value. In the iterative process of particle swarm optimization, the external archive is updated according to the fitness value of the current particle, and the crowding distance of the particles in the archive is calculated. The particle with the largest crowding distance is taken as the global optimal particle, and the particle with the smallest crowding distance is deleted from the archive; when the particle speed and position are updated, the search is performed based on the particle's own historical optimal position, the position of the global optimal particle and the crowding distance; through multiple rounds of iterative updates of particle positions, a set of optimal solutions composed of non-dominated solutions is obtained as the optimal surveying and mapping equipment layout plan set for the surveying and mapping area.

6. The method according to claim 1, characterized in that By integrating multi-temporal remote sensing image data and geographic national survey data, and using the long short-term memory neural network algorithm, the 3D situation evolution information of the surveying and mapping area in different time periods in the future is predicted based on the current 3D situation information, including: Radiometric calibration and geometric correction are performed on high-resolution visible light images and synthetic aperture radar image data acquired at multiple time points. Scale-invariant feature transformation algorithm is used to align remote sensing image data of different temporal phases, and multi-temporal remote sensing image data are fused into a unified time series image sequence through principal component analysis or wavelet transform. The semantic segmentation algorithm based on convolutional neural network is used to extract and annotate the land features of the land use status map, topographic map and place name address map in the national geographic survey data. The multi-level semantic features of the image are extracted through multi-scale convolution kernels, and the conditional random field model is used to achieve pixel-level classification of land features. At the same time, the topological relationship and spatial distribution law between land features are learned by using a bidirectional recurrent neural network, and the interpreted land feature semantic information is spatially associated with the time series image sequence to construct a multi-temporal three-dimensional land feature distribution model. A three-dimensional situation evolution prediction model based on a bidirectional long short-term memory neural network is constructed, and the three-dimensional situation information at each time point is represented as a multidimensional feature vector including object type, geometric shape, spatial position and semantic attributes. The historical three-dimensional situation feature vector sequence is input into the three-dimensional situation evolution prediction model, and the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future is iteratively predicted.

7. The method according to claim 6, characterized in that Construct a three-dimensional situation evolution prediction model based on a bidirectional long short-term memory neural network, including: The continuous variables in the multidimensional feature vector are normalized or standardized to perform feature scaling, and the discrete variables in the multidimensional feature vector are converted by one-hot encoding or embedding methods to obtain a preprocessed three-dimensional situation feature vector; the number of input layer nodes, the number of hidden units in the bidirectional long short-term memory layer, and the number of stacked layers of the bidirectional long short-term memory neural network are determined according to the dimension of the preprocessed three-dimensional situation feature vector; in the bidirectional long short-term memory layer, the key feature information in the preprocessed three-dimensional situation feature vector is output by using an input gating unit, a forgetting gating unit, and an output gating unit; The key feature information is stored in the memory unit, and the forward hidden state and the backward hidden state of the bidirectional long short-term memory layer at each time step are concatenated to form a context feature representation; in the attention layer, the importance weights of the context feature representations at different historical moments are calculated in parallel by multiple attention heads, the historical context features most relevant to the current prediction are determined, and the similarity between the query vector and the key value vector is calculated by using the dot product attention or additive attention method to generate an attention weight distribution, and the attention weight distribution is weighted and summed with the corresponding value vector to obtain an aggregated attention context feature vector; In the fully connected layer, the context feature representation output by the bidirectional long short-term memory layer and the attention context feature vector output by the attention layer are spliced ​​into a complete feature representation vector, and the complete feature representation vector is subjected to nonlinear transformation and dimensional conversion through one or more fully connected layers to extract high-level semantic feature information; according to the type of the three-dimensional situation evolution prediction task, an activation function is selected to activate the high-level semantic feature information; in the output layer, according to the type of the three-dimensional situation evolution prediction task, the activated high-level semantic feature information output by the fully connected layer is used to generate an output feature vector representing the future three-dimensional situation state, and an additional decoder structure is added to the output layer to construct a three-dimensional situation evolution prediction model based on the bidirectional long short-term memory neural network.

8. A surveying and mapping area situation deduction system under multi-objective optimization, used to implement the method described in any one of claims 1 to 7, characterized in that: include: The first unit is used to obtain multi-source heterogeneous geographic information data of the surveying and mapping area, wherein the multi-source heterogeneous geographic information data includes remote sensing image data collected by different types of sensors, laser radar point cloud data and geographic national conditions census data; use a deep learning algorithm to perform feature extraction and semantic segmentation on the multi-source heterogeneous geographic information data to generate a three-dimensional geographic scene model of the surveying and mapping area, wherein the three-dimensional geographic scene model includes a high-precision terrain model, a semantic building model and a multi-level road network model; The second unit is used to construct a multi-objective optimization model for mapping area situation deduction, wherein the multi-objective optimization model takes mapping coverage, mapping efficiency, mapping cost and mapping accuracy as optimization objectives, and includes multiple sub-objective functions, and the weight of each sub-objective function is determined by the hierarchical analysis method; in the three-dimensional geographic scene model, the optimal mapping equipment layout plan set of the mapping area is determined based on the simulation optimization algorithm; based on each plan in the optimal mapping equipment layout plan set, the optimal mapping equipment layout plan is obtained; In the three-dimensional geographic scene model, the surveying and mapping operation process under the optimal surveying and mapping equipment layout scheme is simulated to generate three-dimensional point cloud data consistent with the surveying and mapping operation process, and obtain the current three-dimensional situation information of the surveying and mapping area; The third unit is used to integrate multi-temporal remote sensing image data and geographic national conditions census data, and adopt a long short-term memory neural network algorithm to predict the three-dimensional situation evolution information of the surveying and mapping area in different time periods in the future based on the current three-dimensional situation information; input the three-dimensional situation evolution information into the multi-objective optimization model, recalculate the fitness value of each surveying and mapping equipment layout plan, and update the optimal surveying and mapping equipment layout plan; based on the updated optimal surveying and mapping equipment layout plan, use virtual reality technology to generate three-dimensional visualization scenes of the surveying and mapping area in different time periods in the future.

9. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Citation Information

Patent Citations

  • Method for classifying remote sensing images

    CN102938073A

  • Forest stand factor extraction method fusing unmanned aerial vehicle image and ground-based radar point cloud

    CN112200846A

  • Pointeff method for urban object classification with lidar point cloud data

    US20240020970A1

  • Consilence of data-mining

    WO2007147166A2

  • Three-dimensional lidar point cloud semantic segmentation method and apparatus based on deep learning

    WO2024130776A1

Cited By

  • Multi-element three-dimensional data seamless rapid fusion method and system, product and medium

    CN119251443A

  • Block chain-based intelligent spraying management method and system for ore wharf

    CN119295021A

  • Arterial aneurysm 3D point cloud automatic segmentation method combined with geometric topology analysis

    CN119379721A

  • Object situation awareness method, device and equipment and storage medium

    CN119537780A

  • Land space planning surveying and mapping method and system based on dynamic remote sensing technology

    CN119758329A