Efficient data storage and retrieval method for point cloud digital twinning

Abstract

The invention relates to the technical field of data storage, and particularly provides a point cloud digital twinning-oriented efficient data storage and retrieval method, which comprises the following steps of: converting an original disordered point set into a dynamic voxel unit with probability density by analyzing a spatial distribution entropy value of a point cloud; generating a hierarchical voxel network with entropy weight marks; based on a hierarchical voxel network, extracting a voxel state transition rule across a time sequence, and encoding dynamic change features into a directed weighted graph; map nodes of the directed weighted map represent voxel stability, and a topological map with a space-time coupling characteristic is formed; training an implicit neural network to generate a continuous semantic vector field by using the space-time coupling characteristic of the topological graph; during retrieval, a target feature region is directly positioned through vector field gradient descent and then is reversely mapped to an original voxel level, and discrete retrieval is converted into vector navigation in a continuous space. According to the method, the contradiction between storage compression and retrieval efficiency is unified.

Description

Technical Field

[0001] This invention relates to the field of data storage technology, and in particular to an efficient data storage and retrieval method for point cloud digital twins. Background Technology

[0002] In the digital economy era, data volume is exploding at a compound annual growth rate of 40%. Traditional data storage architectures are like old-fashioned library card catalog systems; when the collection jumps from tens of thousands to hundreds of millions of books, administrators have to spend 90% of their time searching for books rather than serving readers. Currently, various industries face similar dilemmas: financial transaction systems generate hundreds of thousands of records per second, IoT devices generate petabytes of data daily, and medical imaging data grows by 120% annually. This data deluge makes traditional B+ tree index structures like riding a bicycle on a highway; the expansion of index levels leads to an exponential increase in query latency, and storage space waste generally exceeds 35%.

[0003] Prior art 1, Chinese Patent Application No. 202510585078.2, discloses a data quantization query method and system based on spatial raster. The method acquires data of various types and formats through multiple methods and reads it into a system-recognizable format. Subsequently, the data is parsed and reconstructed, completing cleaning and filtering operations, including data translation and parsing, spatial data processing, and JSON data organization. The processed data undergoes precision unification and quantization reconstruction, and is finally stored in a database to generate a query index. While possessing efficient data processing and query capabilities, enabling efficient storage and rapid retrieval of large-scale spatial data, it lacks adaptability to the unique sparsity and local density variations of point clouds, disrupting the inherent detailed topological structure of point clouds or causing redundant storage in low-information areas; it also fails to consider the high dynamism of the time dimension: point cloud digital twins require continuous updates, and their indexes struggle to support rapid incremental updates and spatiotemporal correlation queries.

[0004] Prior art two, Chinese patent application number 202410960502.2, discloses an efficient data storage indexing method, system, and device based on a virtual plane. This method achieves mutual association between edge servers and data items by projecting them onto the same virtual plane. It constructs a Delaunay triangulation and a Voronoi diagram based on the set of coordinate points of the edge servers. The storage location is determined by the location of the data item's coordinate point within the Voronoi diagram region. Greedy routing is performed through Delaunay triangulation to determine the addressing path for the requested data. While this virtual plane-based data indexing mechanism ultimately achieves fast and efficient data storage and retrieval, projecting a 3D point cloud onto a 2D virtual plane results in the loss of crucial height information and 3D spatial topological relationships, failing to accurately reflect the proximity and structural characteristics of the point cloud in real 3D space.

[0005] Prior art three, Chinese patent application number 202410153433.4, discloses an efficient distributed data storage and retrieval system, method, and storage medium, including a retrieval unit for providing retrieval information for distributed content. The input end of the retrieval unit is electrically connected to the output end of a classification retrieval unit, which selects different retrieval function modules according to user needs. Although splitting the data can distribute it evenly across multiple database modules, reducing the load on a single database, and collecting database load information allows for real-time adjustment of the database's balanced load, task scheduling with different weights improves the processing efficiency of distributed data storage, and establishing corresponding feedback information during distributed storage allows for rapid analysis and tracing of the corresponding nodes of the data blocks when the data owner calls for retrieval, thus improving retrieval processing efficiency; however, it focuses on system load and ignores data semantics. It mainly optimizes the physical storage location and access path of the data, solving the scalability and performance bottlenecks at the system level, but does not deeply address the semantic understanding and efficient content retrieval problems of point cloud data itself.

[0006] Current technologies 1, 2, and 3 fail to address the challenge of achieving efficient data management in point cloud digital twin scenarios by coordinating information entropy-driven adaptive storage, spatiotemporally coupled dynamic evolution modeling, and vector navigation retrieval in continuous semantic space. Therefore, this invention provides an efficient data storage and retrieval method for point cloud digital twins. Summary of the Invention

[0007] To achieve the above objectives, the present invention adopts the following technical solution: One aspect of the present invention provides an efficient data storage and retrieval method for point cloud digital twins, comprising the following steps: utilizing the spatiotemporal coupling characteristics of topological maps to train an implicit neural network to generate a continuous semantic vector field; during retrieval, the target feature region is directly located through gradient descent of the vector field, and then back-mapped to the original voxel level, transforming discrete retrieval into vector navigation in continuous space.

[0008] In one alternative implementation, the process of transforming discrete retrieval into vector navigation in continuous space includes the following steps: Based on the node stability and edge weights of the spatiotemporally coupled topological graph, dynamic evolution patterns are extracted, including the mutation trajectory of high-entropy nodes and the gradual change trend of low-entropy nodes; the node state transition rules are encoded into high-dimensional spatiotemporal feature vectors, which are used as inputs to the implicit neural network. The spatiotemporal features of the topological map are fitted using an implicit neural network, and a continuous semantic vector field is output. The gradient direction of the continuous semantic vector field represents the convergence path of the target features, and the field strength reflects the saliency of the region. The training process minimizes the reconstruction error of the map nodes in the vector field. During the retrieval phase, starting from the query location, the search iteratively searches along the negative gradient direction of the semantic vector field until a local extremum is reached. The extremum corresponds to the key feature nodes in the topological graph, and then is back-mapped to the hierarchical voxel network through entropy weight labeling to locate the target region of the original data. Based on the reverse mapping results, the geometric precision is dynamically adjusted in conjunction with the voxel entropy weight coefficient; fine-grained topological structures are returned for high-entropy regions, and aggregated levels are returned for low-entropy regions, achieving an adaptive balance between retrieval precision and efficiency.

[0009] In one alternative implementation, the process of outputting a continuous semantic vector field includes the following steps: The node stability features and edge weight features of the topological graph are processed separately. The node stability features reflect long-term evolutionary patterns, while the edge weight features capture short-term correlation characteristics. The temporal smoothing component and the spatial abrupt change component are obtained through a dual-channel feature extraction architecture. The temporal smoothing component and the spatial abrupt component are dynamically weighted and fused, and the weight coefficients are adaptively adjusted by the node entropy weight label; the fused features are transformed into latent semantic representations in continuous space through multi-layer implicit mapping. Differentiable field functions are constructed in the latent semantic space, with spatial coordinates as inputs and semantic vectors with directionality and magnitude as outputs; the parameters of the differentiable field functions are optimized by the reconstruction constraints of the topological graph node features. By utilizing the spatiotemporal coupling characteristics of the topological map, the evolution continuity of the generated vector field in adjacent time series is verified; local field strength adjustment is performed in regions that do not meet the continuity constraint.

[0010] In one alternative implementation, the process of constructing a differentiable field function in the latent semantic space includes the following steps: Based on the fused features, the latent semantic space is divided into regular grid units, with the center point of each unit storing the semantic feature vector of the corresponding region; the grid resolution is dynamically adjusted by the entropy weight label, with fine-grained division for high-entropy regions and coarse-grained division for low-entropy regions. A locally differentiable field function is defined for each grid cell, and its initial parameters are determined by the weighted average of all semantic features within that cell; the weights are determined by the spatiotemporal correlation of the node stability features. A globally continuous semantic vector field is formed by connecting the local field functions of adjacent grid cells using a differentiable interpolation method; the interpolation weights are adjusted by the feature similarity between grid cells. The node reconstruction error of the topological graph is used as a monitoring signal to iteratively adjust the field function parameters. The optimization process forces the field function to output vector direction and amplitude consistent with the graph at key feature nodes, while maintaining the continuous differentiability of non-node regions.

[0011] In one alternative implementation, the process of locating the target region of the raw data includes the following steps: Using the query location as the initial point, obtain the gradient direction and field strength of the continuous semantic vector field at the initial point; perform adaptive step-size iterations along the negative gradient direction, with the step size dynamically adjusted by the field strength at the current position; after each iteration, detect the rate of change of the gradient magnitude, and determine that the extreme point has been reached when the rate of change is lower than the threshold. Extract the direction and magnitude features of semantic vectors at extreme points; perform similarity measurement with node features in the spatiotemporally coupled topological graph; after successful matching, obtain the entropy weight label sequence and temporal evolution identifier associated with the node; Based on the entropy weight labels, the corresponding voxel hierarchical structure is parsed. High entropy labels trigger fine-grained reconstruction to restore the local topological details of the original point cloud, while low entropy labels activate cross-level aggregation to generate simplified voxel blocks that maintain topological consistency. The deformation continuity of the reconstruction results in adjacent time series is checked, and local gradient backtracking is performed for regions that do not meet the spatiotemporal evolution law; accuracy compensation is achieved by adjusting the extreme point judgment threshold; the output dimension is automatically selected in combination with the query semantic requirements, and the spatial analysis task focuses on geometric accuracy output.

[0012] In one optional implementation, the process of measuring the similarity with node features in a spatiotemporally coupled topological graph includes the following steps: Based on the obtained extreme point coordinates, the three-dimensional direction component and normalized amplitude of the position semantic vector are extracted to form a four-dimensional feature descriptor. The four-dimensional feature descriptor generates direction feature vector and amplitude feature vector respectively through a two-stream encoder. The direction feature vector is expanded using spherical Fourier basis functions, and the amplitude feature vector is transformed by logarithmic transformation to generate a multi-scale energy distribution representation. The node features in the spatiotemporally coupled topological graph are pre-processed with similar dual-stream coding to form a comparable feature space. The matching process adopts a dynamic feature gating mechanism, which measures the cosine similarity between the directional feature vector of the current extreme point and the graph node, and aligns the amplitude feature vector step by step through cross-attention weight allocation. The two similarity scores are fused with learnable parameters to generate a comprehensive matching confidence score. When the overall matching confidence exceeds the adaptive threshold, node association information extraction is triggered; the gradient change trajectory recorded during the iterative convergence process is re-encoded into temporal features and phase-aligned with the temporal evolution identifier of the node for verification; the verified matching result will activate the entropy weight label decoder, which uses the gradient magnitude change rate at the extreme point as a priori constraint to perform confidence-weighted decoding on the label sequence.

[0013] In one alternative implementation, the process of generating simplified but topologically consistent voxel blocks includes the following steps: The system receives a sequence of low-entropy labels as input, which contains a verified topological feature priority indicator. The low-entropy labels trigger a hierarchical merging protocol to divide the original point cloud data into multiple regional units with similar geometric characteristics. Each region unit is assigned a resolution coefficient based on its corresponding entropy weight. The resolution coefficient is negatively correlated with the rate of change of the gradient magnitude at the extreme point. Feature-preserving sampling is performed within the region unit, and the sampling density is dynamically controlled by the resolution coefficient, while the connectivity constraints of the boundary transition region are preserved. The sampled point set is converted into an initial voxel mesh through adaptive spatial partitioning, and the mesh cell size and resolution coefficient maintain a linear mapping relationship. Topological consistency detection is performed on the initial voxel mesh, and the detection criteria are derived from the adjacency relationship templates stored in the successfully matched spatiotemporally coupled topological graph. During the detection process, a compensation mechanism is activated for discontinuous regions, using pre-stored deformation continuity parameters to adjust the local mesh.

[0014] In one alternative implementation, the adjusted voxel block undergoes final verification, with the geometric simplification rate conforming to the specified entropy weighting requirements.

[0015] In one optional implementation, the original disordered point set is transformed into dynamic voxel units with probability density by analyzing the spatial distribution entropy value of the point cloud; each dynamic voxel unit adaptively adjusts its geometric precision according to the local entropy value, retains the detailed topological structure in high-entropy regions, and merges low-entropy regions into sparse hierarchies to generate a hierarchical voxel network with entropy weight labels.

[0016] In one optional implementation, based on a hierarchical voxel network, the voxel state transition patterns across time series are extracted, and the dynamic change features are encoded into a directed weighted graph. The graph nodes of the directed weighted graph represent voxel stability, and the edge weights reflect the deformation correlation strength between adjacent time series, forming a topological graph with spatiotemporal coupling characteristics.

[0017] This invention achieves efficient storage and intelligent retrieval of point cloud digital twin data through a three-stage collaborative optimization process: spatial entropy-driven dynamic voxelization, temporal topology map construction, and semantic vector field indexing. This embodiment dynamically allocates geometric precision based on spatial entropy values, avoiding redundant storage of low-entropy region data while preserving details of high-entropy regions, thus achieving adaptive compression and distribution of storage resources. The hierarchical voxel network structure further reduces the explicit recording of unnecessary topological information. The temporal topology map encodes dynamic changes between discrete frames into continuously weighted relationships, implicitly representing the evolutionary patterns of point clouds, eliminating the redundancy of traditional multi-frame independent storage, and providing associated paths for cross-temporal retrieval. The semantic vector field maps discrete voxel levels through continuous space, transforming traditional spatial partitioning-based retrieval into gradient-driven feature navigation, reducing iterative computation and directly locating the target region; the spatiotemporal coupling characteristics of the topology map ensure the consistency of retrieval results across the time dimension. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of the efficient data storage and retrieval method for point cloud digital twins provided in Embodiment 1 of the present invention; Figure 2 This is a process diagram of generating a hierarchical voxel network with entropy weight labels provided in Embodiment 2 of the present invention; Figure 3 This is a process diagram of forming a topological map with spatiotemporal coupling characteristics provided in Embodiment 3 of the present invention; Figure 4 This is a process diagram of converting discrete retrieval into vector navigation in continuous space, as provided in Embodiment 4 of the present invention; Figure 5 A block diagram of the electronic device provided by the present invention; Figure 6A block diagram of a computer-readable storage medium provided for this invention. Detailed Implementation

[0019] The technical solutions of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Hereinafter, the terms "first," "second," etc., are used for descriptive convenience only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0021] In this invention, unless otherwise explicitly specified and limited, the term "connection" should be interpreted broadly. For example, "connection" can be a fixed mechanical connection, a detachable mechanical connection, or an integral part; or, "connection" can be a direct connection or an indirect connection through an intermediate medium. Furthermore, unless otherwise explicitly specified and limited, the term "coupling" should be interpreted broadly. For example, "coupling" can be a direct electrical connection, such as physical contact and electrical conduction between two components; it can also be understood as an electrical connection between different components in a circuit structure through physical lines capable of transmitting electrical signals, such as copper foil or wires on a printed circuit board (PCB), to transmit electrical signals; or, "coupling" can be an indirect electrical connection between two components through an intermediate medium; or, "coupling" can be an electrical connection between two components in a non-contact manner, such as an electrical connection between two components using capacitive coupling to transmit electrical signals.

[0022] In this embodiment of the invention, directional terms such as "up," "down," "left," and "right" may be defined relative to the orientation of the components shown in the accompanying drawings. It should be understood that these directional terms can be relative concepts, used for relative description and clarification, and can change accordingly depending on the orientation of the components in the accompanying drawings.

[0023] Example 1: like Figure 1 As shown, this embodiment of the invention provides an efficient data storage and retrieval method for point cloud digital twins, comprising the following steps: Step S100: By analyzing the spatial distribution entropy value of the point cloud, the original disordered point set is transformed into dynamic voxel units with probability density; each dynamic voxel unit adaptively adjusts its geometric precision according to the local entropy value, retains the detailed topological structure in high-entropy regions, and merges low-entropy regions into sparse hierarchies to generate a hierarchical voxel network with entropy weight labels. Step S200: Based on the hierarchical voxel network, extract the voxel state transition rules across time series, and encode the dynamic change features into a directed weighted graph; the graph nodes of the directed weighted graph represent voxel stability, and the edge weights reflect the deformation correlation strength between adjacent time series, forming a topological graph with spatiotemporal coupling characteristics. Step S300: Utilize the spatiotemporal coupling characteristics of the topological map to train an implicit neural network to generate a continuous semantic vector field; during retrieval, the target feature region is directly located through gradient descent of the continuous semantic vector field, and then back-mapped to the original voxel level, transforming discrete retrieval into vector navigation in continuous space.

[0024] In the above embodiments, this embodiment achieves efficient storage and intelligent retrieval of point cloud digital twin data through the coordinated optimization of three stages: spatial entropy-driven dynamic voxelization, temporal topology map construction, and semantic vector field indexing. This embodiment dynamically allocates geometric precision through spatial entropy values, avoiding redundant storage of low-entropy region data while preserving details of high-entropy regions, achieving adaptive compression and distribution of storage resources. The structural characteristics of the hierarchical voxel network further reduce the explicit recording of unnecessary topological information. The temporal topology map encodes the dynamic changes between discrete frames into continuous weighted relationships, implicitly representing the evolutionary laws of point clouds, eliminating the redundancy of traditional multi-frame independent storage, and providing associated paths for cross-time retrieval. The semantic vector field maps discrete voxel levels through continuous space, transforming traditional spatial partitioning-based retrieval into gradient-driven feature navigation, reducing iterative calculations and directly locating the target region; the spatiotemporal coupling characteristics of the topology map ensure the consistency of retrieval results in the time dimension.

[0025] In summary, this embodiment resolves the conflict between storage compression and retrieval efficiency through a chain-like transformation of entropy-map-vector field: dynamic voxelization reduces storage overhead, topological maps enhance spatiotemporal continuity, and vector field indexing accelerates retrieval response, ultimately achieving efficient management and precise access to high-density data in point cloud digital twins.

[0026] Example 2: like Figure 2 As shown, based on Example 1, the process of generating a hierarchical voxel network with entropy weight labels in step S100 of this embodiment of the invention includes the following steps: Step S101: Spatial density sampling is performed on the point cloud, the entropy distribution of each local region is calculated, and a probability field reflecting the degree of disorder of the point set is generated; high entropy regions correspond to parts with complex geometric features or significant dynamic changes, while low entropy regions represent smooth structures or static backgrounds. Step S102: Voxel unit generation is driven by entropy field. High-entropy regions are divided into fine voxels to preserve details, while low-entropy regions are merged into sparse large-scale voxels. Each voxel unit is associated with an entropy weight coefficient, which represents the geometric precision level to be recovered during quantization decoding. Step S103: Construct a hierarchical structure from bottom to top based on the voxel entropy weight coefficients: the bottom layer consists of voxels with original precision, and the upper layers aggregate low-entropy voxels through entropy weighting to form a multi-resolution representation; finally, output a hierarchical voxel network with entropy weight labels.

[0027] In the above embodiments, this embodiment achieves intelligent hierarchical compression and multi-scale representation of point cloud data through an entropy-driven dynamic voxelization strategy. By mapping the spatial entropy distribution to voxel granularity control parameters, it intelligently merges low-entropy regions while maintaining the geometric accuracy of high-entropy regions, achieving a compression ratio for non-uniform sampling rates. The entropy weight coefficients of the bottom voxels are propagated to higher layers through weighted aggregation, forming a multi-resolution pyramid with accuracy inheritance. The entropy weight coefficients carried by each voxel unit essentially constitute the accuracy control parameters during decoding, and the reconstruction level can be dynamically selected according to the view distance in real-time rendering.

[0028] In summary, this embodiment uses information entropy as a physical quantity to unify the optimization goals of data compression, multi-scale representation, and real-time rendering across three dimensions.

[0029] Example 3: like Figure 3 As shown, based on Embodiment 1, the process of forming a topological map with spatiotemporal coupling characteristics in step S200 of this embodiment of the invention includes the following steps: Step S201: Based on the entropy weight label of the hierarchical voxel network, statistically analyze the changes in geometric properties of each voxel unit in the continuous time series, including positional offset, volume scaling and topological connectivity changes, and calculate the voxel state transition matrix; the migration mode of the high-entropy region is characterized by high-frequency local perturbation, while the low-entropy region exhibits low-frequency global displacement. Step S202: Project the voxel state transition matrix onto the graph space. Nodes represent the spatiotemporal stability of voxels, determined by the product of entropy weight coefficients and state transition frequencies. Edge weights are determined by the deformation correlation of voxels in adjacent time series, including the consistency of deformation direction, similarity of amplitude, and strength of topological constraints. A preliminary weighted directed graph structure is formed. Step S203: Apply bidirectional constraints to the weighted directed graph. The positive constraint forces short-term abrupt associations between high-entropy nodes, while the negative constraint maintains long-term smooth transitions of low-entropy nodes. Through iterative optimization, the graph simultaneously satisfies local dynamic accuracy and global motion consistency, and finally outputs a topological graph with spatiotemporal coupling characteristics.

[0030] In the above embodiments, this embodiment uses entropy weight labeling to drive state transition modeling and utilizes the hierarchical characteristics of voxel networks to distinguish spatiotemporal evolution patterns. The resulting topological map retains detailed dynamic features while maintaining the overall motion trend, providing a structured representation basis for subsequent semantic vector field training.

[0031] Example 4: like Figure 4 As shown, based on Embodiment 1, the process of converting discrete retrieval into vector navigation in continuous space in step S300 of this embodiment of the invention includes the following steps: Step S301: Based on the node stability and edge weights of the spatiotemporally coupled topological graph, extract the dynamic evolution pattern, including the mutation trajectory of high-entropy nodes and the gradual change trend of low-entropy nodes; encode the node state transition law into a high-dimensional spatiotemporal feature vector as the input of the implicit neural network; Step S302: Use an implicit neural network to fit the spatiotemporal features of the topological map and output a continuous semantic vector field; the gradient direction of the continuous semantic vector field represents the convergence path of the target features, and the field strength reflects the regional saliency; the training process minimizes the reconstruction error of the map nodes in the vector field; Step S303: In the retrieval phase, starting from the query location, iteratively search along the negative gradient direction of the semantic vector field until a local extremum is reached; the extremum corresponds to the key feature nodes in the topological graph, and then is back-mapped to the hierarchical voxel network through entropy weight labeling to locate the target region of the original data. Step S304: Based on the reverse mapping results, dynamically adjust the geometric precision by combining the voxel entropy weight coefficients; return fine-grained topology structures for high-entropy regions and aggregate levels for low-entropy regions, achieving an adaptive balance between retrieval precision and efficiency.

[0032] In the above embodiments, this embodiment drives the generation of semantic vector fields through topological graphs, uses gradient navigation to replace traditional discrete retrieval, optimizes the query path in continuous space, and maintains the precision control of the hierarchical voxel network through entropy weight mapping, forming an end-to-end spatiotemporal feature retrieval framework.

[0033] Example 5: Based on Example 4, the process of outputting a continuous semantic vector field in step S302 of this embodiment of the invention includes the following steps: Step S3021: Separate the node stability features and edge weight features of the topological graph. The node stability features reflect the long-term evolution law, while the edge weight features capture the short-term correlation characteristics. The temporal smoothing component and the spatial abrupt change component are obtained respectively through a dual-channel feature extraction architecture. Step S3022: Dynamically weight and fuse the temporal smoothing component and the spatial abrupt component, with the weight coefficients adaptively adjusted by the node entropy weight label; the fused features are converted into latent semantic representations in continuous space through multi-layer implicit mapping; Step S3023: Construct a differentiable field function in the latent semantic space, with spatial coordinates as input and semantic vectors with directionality and amplitude as output; the parameters of the differentiable field function are optimized by the reconstruction constraints of the topological graph node features; Step S3024: Utilize the spatiotemporal coupling characteristics of the topological map to verify the evolution continuity of the generated vector field in adjacent time series; adjust the local field strength for regions that do not meet the continuity constraint.

[0034] In the above embodiments, this embodiment processes spatiotemporal features through a decoupling-fusion strategy and uses entropy weighting to guide feature fusion. The resulting semantic vector field not only maintains the key features of the topological map but also possesses continuous differentiable mathematical properties, providing optimization space for subsequent gradient navigation.

[0035] Example 6: Based on Example 5, the process of constructing a differentiable field function in the latent semantic space in step S3023 of this embodiment of the invention includes the following steps: Step S30231: Based on the fused features, the latent semantic space is divided into regular grid units, and the center point of each unit stores the semantic feature vector of the corresponding region; the grid resolution is dynamically adjusted by the entropy weight label, with fine-grained division for high-entropy regions and coarse-grained division for low-entropy regions. Step S30232: Define a locally differentiable field function for each grid cell, whose initial parameters are determined by the weighted average of all semantic features within the cell; the weights are determined by the spatiotemporal correlation of the node stability features. Step S30233: Connect the local field functions of adjacent grid cells using a differentiable interpolation method to form a globally continuous semantic vector field; the interpolation weights are adjusted by the feature similarity between grid cells; Step S30234: Use the node reconstruction error of the topological map as a supervision signal to iteratively adjust the field function parameters; the optimization process forces the field function to output the same vector direction and amplitude as the map at key feature nodes, while maintaining the continuous differentiability of non-node regions.

[0036] In the above embodiments, this embodiment achieves local controllability of the field function through a structured latent semantic space, uses entropy weighting to guide dynamic resolution partitioning, and the finally generated field function not only accurately matches the key features of the topological map, but also has the mathematical property of global continuous differentiability, providing an optimization basis for gradient navigation.

[0037] Example 7: Based on Example 4, the process of locating the target region of the original data in step S303 of this embodiment of the invention includes the following steps: Step S3031: Using the query location as the initial point, obtain the gradient direction and field strength of the continuous semantic vector field at the initial point; perform adaptive step-size iteration along the negative gradient direction, with the step size dynamically adjusted by the field strength at the current position; detect the rate of change of the gradient magnitude after each iteration, and determine that the extreme point has been reached when the rate of change is lower than the threshold. Step S3032: Extract the direction and magnitude features of the semantic vector at the extreme point; perform similarity measurement with the node features in the spatiotemporal coupled topological graph; after successful matching, obtain the entropy weight label sequence and temporal evolution identifier associated with the node; Step S3033: Based on the entropy weight labels, the corresponding voxel hierarchical structure is parsed. High entropy labels trigger fine-grained reconstruction to restore the local topological details of the original point cloud. Low entropy labels activate cross-level aggregation to generate simplified voxel blocks that maintain topological consistency. Step S3034: Check the deformation continuity of the reconstruction results in adjacent time series, and perform local gradient backtracking for regions that do not meet the spatiotemporal evolution law; achieve accuracy compensation by adjusting the extreme point judgment threshold; automatically select the output dimension based on the query semantic requirements, with the spatial analysis task focusing on geometric accuracy output.

[0038] In the above embodiments, this embodiment achieves precise intelligent positioning of the target region of the original data through the synergistic effect of gradient navigation and hierarchical mapping; it deeply integrates the geometric guidance capability of continuous semantic vector fields with the semantic representation capability of discrete topological maps to form a closed-loop optimized retrieval mechanism. In the initial stage, the gradient trajectory generated based on the query position not only follows the physical constraints of field strength changes, but also balances search efficiency and stability through dynamic step size adjustment; after reaching the extreme point, the correspondence between continuous space and discrete topology is established through feature matching, so that the directional characteristics of semantic vectors are transformed into interpretable entropy weight labels; the conversion process retains the multi-scale characteristics of the original data, ensuring that the microstructural features of high-entropy regions and the macroscopic evolution laws of low-entropy regions can be accurately captured. The subsequent hierarchical reconstruction is not a simple data restoration, but rather an adaptive precision allocation driven by entropy weight labels, forming a smooth transition between fine-grained reconstruction and cross-layer aggregation. The spatiotemporal consistency verification step further enhances the reliability of the output results. By eliminating abrupt noise in the temporal dimension through the gradient backtracking mechanism, the positioning results can simultaneously meet the requirements of spatial accuracy and temporal coherence. The final output target area has dynamic optimization characteristics and can automatically adjust the presentation weight of geometric details and evolution trends according to the task type, achieving dual protection of retrieval accuracy and semantic adaptability. In summary, this embodiment employs a three-stage collaborative mechanism based on continuous field gradients for navigation, entropy weighting as a conversion bridge, and spatiotemporal consistency as a verification standard. This mechanism enables the target area localization process to possess physical rationality, semantic interpretability, and cross-scale adaptability.

[0039] Example 8: Based on Example 7, the process of measuring the similarity between the node features in the spatiotemporally coupled topological graph in step S3032 of this embodiment of the invention includes the following steps: Step S30321: Based on the obtained extreme point coordinates, extract the three-dimensional direction component and normalized amplitude of the position semantic vector to form a four-dimensional feature descriptor; the four-dimensional feature descriptor generates direction feature vector and amplitude feature vector respectively through a two-stream encoder, wherein the direction feature vector is expanded using a spherical Fourier basis function, and the amplitude feature vector is transformed by logarithmic transformation to generate a multi-scale energy distribution representation. Step S30322: The node features in the spatiotemporally coupled topological graph are pre-processed with similar dual-stream coding to form a comparable feature space; the matching process adopts a dynamic feature gating mechanism, which measures the cosine similarity between the directional feature vector of the current extreme point and the graph node, and simultaneously aligns the amplitude feature vector step by step through cross-attention weight allocation; after the two similarity scores are fused with learnable parameters, a comprehensive matching confidence score is generated. Step S30323: When the comprehensive matching confidence exceeds the adaptive threshold, node association information extraction is triggered; the gradient change trajectory recorded during the iterative convergence process is re-encoded into temporal features and phase-aligned with the temporal evolution identifier of the node for verification; the verified matching result will activate the entropy weighted label decoder, which uses the gradient magnitude change rate at the extreme point as a priori constraint to perform confidence-weighted decoding on the label sequence.

[0040] In the above embodiments, the entire matching process of this embodiment forms a closed-loop feedback: the gradient characteristics of extreme points affect the feature encoding method, the encoding result determines the matching strategy, and the matching quality feedback adjusts the gradient iteration parameters; the coupling mechanism ensures the coordinated optimization of feature extraction and matching, so that the output of step S3032 naturally connects to the entropy weight analysis requirements of step S3033.

[0041] Example 9: Based on Example 7, the process of generating simplified but topologically consistent voxel blocks in step S3033 of this embodiment of the invention includes the following steps: Step S30331: Receive the output low-entropy label sequence as input. The low-entropy label sequence contains verified topological feature priority indicators. The low-entropy labels trigger a hierarchical merging protocol to divide the original point cloud data into multiple regional units with similar geometric characteristics. Step S30332: Each region unit is assigned a resolution coefficient according to its corresponding entropy weight. The resolution coefficient is negatively correlated with the rate of change of the gradient magnitude at the extreme point. Feature-preserving sampling is performed inside the region unit. The sampling density is dynamically controlled by the resolution coefficient, while the connectivity constraints of the boundary transition region are preserved. Step S30333: The sampled point set is converted into an initial voxel mesh through adaptive spatial partitioning, and the mesh cell size and resolution coefficient maintain a linear mapping relationship; a topological consistency check is performed on the initial voxel mesh, and the check criteria are derived from the adjacency relationship template stored in the successfully matched spatiotemporal coupled topology map; Step S30334: The compensation mechanism is activated for discontinuous regions found during the detection process, and the local mesh is adjusted using the pre-stored deformation continuity parameters; the adjusted voxel block is then finally verified, and the geometric simplification rate is consistent with the specified entropy weight label requirements.

[0042] In the above embodiments, this embodiment establishes a deterministic mapping from entropy weight labels to simplified voxels. Through a dual mechanism of dynamic resolution adjustment and topological constraint detection, a balance is achieved between reducing geometric complexity and preserving structural features. The output voxel block carries the core topological attributes of the original data, providing standardized input for subsequent spatiotemporal analysis.

[0043] Figure 5A block diagram of an exemplary electronic device suitable for implementing embodiments of the present invention is shown.

[0044] Electronic devices may include a central processing unit / microprocessor / main control chip, etc.; a storage medium coupled to the central processing unit / microprocessor / main control chip, etc., and storing computer-executable instructions therein for performing the steps of various methods of embodiments of the present invention when executed by a processor.

[0045] The central processing unit / microprocessor / main control chip, etc., may include, but are not limited to, one or more processors or microprocessors.

[0046] Storage media may include, but are not limited to, random access memory (RAM), read-only memory (ROM), flash memory, EPROM memory, EEPROM memory, registers, and computer storage media (such as hard disks, floppy disks, solid-state drives, removable disks, CD-ROMs, DVD-ROMs, Blu-ray discs, etc.).

[0047] In addition, the electronic device may include (but is not limited to) a data bus, an input / output bus / external bus / device bus, a display, and input / output devices (e.g., keyboard, mouse, speaker, etc.).

[0048] The central processing unit / microprocessor / main control chip, etc., can communicate with external devices via the I / O bus through a wired or wireless network (not shown).

[0049] The storage medium may also store at least one computer-executable instruction for performing the steps of various functions and / or methods in the embodiments described herein when run by a central processing unit / microprocessor / main control chip, etc.

[0050] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0051] Figure 6 A schematic diagram of a computer-readable storage medium according to an embodiment of the present invention is shown.

[0052] like Figure 6As shown, instructions, such as computer-readable instructions, are stored on a non-transitory computer-readable storage medium. When the computer-readable instructions are executed by a processor, the various methods described above can be performed. The non-transitory computer-readable storage medium includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-transitory non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, the non-transitory computer-readable storage medium can be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.

[0053] In the several embodiments provided by this invention, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0054] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0055] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0056] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for executing all or part of the steps of the methods of the various embodiments of this invention through a computer device (which may be a personal computer, server, or network device, etc.). The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, optical disks, and other media capable of storing program code.

[0057] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for efficient data storage and retrieval of point cloud digital twins, characterized in that, The process includes the following steps: utilizing the spatiotemporal coupling characteristics of topological maps, training an implicit neural network to generate a continuous semantic vector field; during retrieval, directly locating the target feature region through gradient descent of the vector field, and then back-mapping it back to the original voxel level, thus transforming discrete retrieval into vector navigation in continuous space.

2. The efficient data storage and retrieval method for point cloud digital twins as described in claim 1, characterized in that, The process of transforming discrete retrieval into vector navigation in continuous space includes the following steps: Based on the node stability and edge weights of the spatiotemporally coupled topological graph, dynamic evolution patterns are extracted, including the abrupt change trajectory of high-entropy nodes and the gradual change trend of low-entropy nodes. The node state transition patterns are encoded into high-dimensional spatiotemporal feature vectors, which are then used as input to the implicit neural network. The spatiotemporal features of the topological map are fitted using an implicit neural network, and a continuous semantic vector field is output. The gradient direction of the continuous semantic vector field represents the convergence path of the target features, and the field strength reflects the regional saliency. The training process minimizes the reconstruction error of the graph nodes in the vector field; During the retrieval phase, starting from the query location, the search iteratively searches along the negative gradient direction of the semantic vector field until a local extremum is reached. The extreme points correspond to key feature nodes in the topological graph, and then are back-mapped to the hierarchical voxel network through entropy weight labeling to locate the target region of the original data. Based on the reverse mapping results, the geometric precision is dynamically adjusted in conjunction with the voxel entropy weight coefficient; fine-grained topological structures are returned for high-entropy regions, and aggregated levels are returned for low-entropy regions, achieving an adaptive balance between retrieval precision and efficiency.

3. The efficient data storage and retrieval method for point cloud digital twins as described in claim 2, characterized in that, The process of outputting a continuous semantic vector field includes the following steps: The node stability features and edge weight features of the topological graph are processed separately. The node stability features reflect long-term evolutionary patterns, while the edge weight features capture short-term correlation characteristics. The temporal smoothing component and the spatial abrupt change component are obtained through a dual-channel feature extraction architecture. The temporal smoothing component and the spatial abrupt component are dynamically weighted and fused, and the weight coefficients are adaptively adjusted by the node entropy weight label; the fused features are transformed into latent semantic representations in continuous space through multi-layer implicit mapping. Differentiable field functions are constructed in the latent semantic space, with spatial coordinates as inputs and semantic vectors with directionality and magnitude as outputs; the parameters of the differentiable field functions are optimized by the reconstruction constraints of the topological graph node features. By utilizing the spatiotemporal coupling characteristics of the topological map, the evolution continuity of the generated vector field in adjacent time series is verified; local field strength adjustment is performed in regions that do not meet the continuity constraint.

4. The efficient data storage and retrieval method for point cloud digital twins as described in claim 3, characterized in that, The process of constructing differentiable field functions in the latent semantic space includes the following steps: Based on the fused features, the latent semantic space is divided into regular grid units, with the center point of each unit storing the semantic feature vector of the corresponding region; the grid resolution is dynamically adjusted by the entropy weight label, with fine-grained division for high-entropy regions and coarse-grained division for low-entropy regions. A locally differentiable field function is defined for each grid cell, and its initial parameters are determined by the weighted average of all semantic features within that cell; the weights are determined by the spatiotemporal correlation of the node stability features. A globally continuous semantic vector field is formed by connecting the local field functions of adjacent grid cells using a differentiable interpolation method; the interpolation weights are adjusted by the feature similarity between grid cells. The node reconstruction error of the topological graph is used as a monitoring signal to iteratively adjust the field function parameters. The optimization process forces the field function to output vector direction and amplitude consistent with the graph at key feature nodes, while maintaining the continuous differentiability of non-node regions.

5. The efficient data storage and retrieval method for point cloud digital twins as described in claim 2, characterized in that, The process of locating the target region of the raw data includes the following steps: Using the query location as the initial point, obtain the gradient direction and field strength of the continuous semantic vector field at the initial point; perform adaptive step-size iteration along the negative gradient direction, with the step size dynamically adjusted by the field strength at the current position; After each iteration, the rate of change of the gradient magnitude is detected, and when the rate of change is lower than the threshold, it is determined that the extreme point has been reached. Extract the direction and magnitude features of the semantic vector at extreme points; Similarity measurement is performed on node features in the spatiotemporally coupled topological graph; After a successful match, obtain the entropy weight label sequence and temporal evolution identifier associated with that node; Based on the entropy weight labels, the corresponding voxel hierarchical structure is parsed. High entropy labels trigger fine-grained reconstruction to restore the local topological details of the original point cloud, while low entropy labels activate cross-level aggregation to generate simplified voxel blocks that maintain topological consistency. The deformation continuity of the reconstruction results in adjacent time series is checked, and local gradient backtracking is performed for regions that do not meet the spatiotemporal evolution law; accuracy compensation is achieved by adjusting the extreme point judgment threshold; the output dimension is automatically selected in combination with the query semantic requirements, and the spatial analysis task focuses on geometric accuracy output.

6. The efficient data storage and retrieval method for point cloud digital twins as described in claim 5, characterized in that, The process of measuring the similarity between the node features in the spatiotemporally coupled topological graph includes the following steps: Based on the obtained extreme point coordinates, the three-dimensional direction component and normalized amplitude of the position semantic vector are extracted to form a four-dimensional feature descriptor. The four-dimensional feature descriptor generates direction feature vector and amplitude feature vector respectively through a two-stream encoder. The direction feature vector is expanded using spherical Fourier basis functions, and the amplitude feature vector is transformed by logarithmic transformation to generate a multi-scale energy distribution representation. The node features in the spatiotemporal coupled topology graph are pre-processed with similar dual-stream coding to form a comparable feature space; the matching process adopts a dynamic feature gating mechanism, which measures the cosine similarity between the directional feature vector of the current extreme point and the graph node, and aligns the amplitude feature vector step by step through cross-attention weight allocation; The two similarity scores are fused using learnable parameters to generate a comprehensive matching confidence score. When the overall matching confidence exceeds the adaptive threshold, node association information extraction is triggered; the gradient change trajectory recorded during the iterative convergence process is re-encoded into temporal features and phase-aligned with the temporal evolution identifier of the node for verification. The verified matching results will activate the entropy-weighted label decoder, which uses the rate of change of gradient magnitude at extreme points as a priori constraint to perform confidence-weighted decoding on the label sequence.

7. The efficient data storage and retrieval method for point cloud digital twins as described in claim 5, characterized in that, The process of generating simplified but topologically consistent voxel blocks includes the following steps: The system receives a sequence of low-entropy labels as input, which contains a verified topological feature priority indicator. The low-entropy labels trigger a hierarchical merging protocol to divide the original point cloud data into multiple regional units with similar geometric characteristics. Each region unit is assigned a resolution coefficient based on its corresponding entropy weight. The resolution coefficient is negatively correlated with the rate of change of the gradient magnitude at the extreme point. Feature-preserving sampling is performed within the region unit, and the sampling density is dynamically controlled by the resolution coefficient, while the connectivity constraints of the boundary transition region are preserved. The sampled point set is converted into an initial voxel mesh through adaptive spatial partitioning, and the mesh cell size and resolution coefficient maintain a linear mapping relationship. Topological consistency detection is performed on the initial voxel mesh, and the detection criteria are derived from the adjacency relationship templates stored in the successfully matched spatiotemporally coupled topological graph. During the detection process, a compensation mechanism is activated for discontinuous regions, using pre-stored deformation continuity parameters to adjust the local mesh.

8. The efficient data storage and retrieval method for point cloud digital twins as described in claim 7, characterized in that, The adjusted voxel blocks were then subjected to final verification, and the geometric simplification rate was found to be consistent with the specified entropy weighting requirements.

9. The efficient data storage and retrieval method for point cloud digital twins as described in claim 1, characterized in that, By analyzing the spatial distribution entropy of point clouds, the original disordered point set is transformed into dynamic voxel units with probability density. Each dynamic voxel unit adaptively adjusts its geometric precision according to the local entropy value. High-entropy regions retain detailed topological structures, while low-entropy regions are merged into sparse hierarchies, generating a hierarchical voxel network with entropy weight labels.

10. The efficient data storage and retrieval method for point cloud digital twins as described in claim 1, characterized in that, Based on a hierarchical voxel network, the voxel state transition patterns across time series are extracted, and the dynamic change features are encoded into a directed weighted graph. The graph nodes of the directed weighted graph represent voxel stability, and the edge weights reflect the deformation correlation strength between adjacent time series, forming a topological graph with spatiotemporal coupling characteristics.

Images