Index construction method, device and equipment

Abstract

The invention provides an index construction method, device and equipment. The method comprises the following steps: acquiring vector data and raster data; obtaining fusion features of the vector data and the raster data according to the vector data and the raster data; determining a target index parameter according to the fusion feature and a swarm intelligence optimization algorithm; and constructing an index of the vector data and an index of the raster data according to the target index parameter. According to the method provided by the embodiment of the invention, the suitability of the index parameters and the data features is effectively improved, the problems of split management, poor suitability and low efficiency in the related technology are solved, and the finally constructed index can effectively consider the requirements of fine granularity and large-scale retrieval.

Description

Technical Field

[0001] This invention relates to the fields of GIS information and artificial intelligence technology, and in particular to an index construction method, apparatus and device. Background Technology

[0002] In related technologies, geographic information data management often employs a parallel approach of using vector data structures and raster data structures. Therefore, how to ensure efficient multi-source data management while fully meeting diverse retrieval needs is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0003] This invention provides an index construction method, apparatus, and device that effectively improves the adaptability of index parameters and data features, solves the problems of fragmented management, poor adaptability, and low efficiency in related technologies, and enables the final constructed index to effectively meet both fine-grained and wide-ranging retrieval needs.

[0004] This invention provides an index construction method, comprising the following steps.

[0005] Acquire vector and raster data; Based on the vector data and raster data, the fusion features of the vector data and raster data are obtained; Based on the fusion features and swarm intelligence optimization algorithm, the target index parameters are determined; Based on the target index parameters, construct the indexes for the vector data and the raster data.

[0006] According to an index construction method provided by the present invention, obtaining the fusion feature of the vector data and raster data based on the vector data and raster data includes: Multi-channel information is extracted from the vector data and raster data; the multi-channel information includes vector domain information, floor detail information, and satellite image texture information. The fusion feature is obtained based on the multi-channel information.

[0007] According to an index construction method provided by the present invention, obtaining the fusion feature based on the multi-channel information includes: The multi-channel information is input into the basic convolution module to obtain initial feature information; the basic convolution module assigns different convolution kernel weights to each channel information. The initial feature information is input into a multi-layer convolutional module to obtain the feature information output by each convolutional layer, and the first feature information is then used as the input. The feature information output by the convolutional layers and the features from layer 1 to layer 2 The feature information output from the -1 layer convolution is fused to obtain the fused feature; It is a positive integer greater than or equal to 2.

[0008] According to an index construction method provided by the present invention, before extracting multi-channel information from the vector data and raster data, the method further includes: The vector data and raster data are projected onto the same coordinate system and aligned and spatially divided.

[0009] According to an index construction method provided by the present invention, determining the target index parameters based on the fusion features and the swarm intelligence optimization algorithm includes: Based on the fusion features, construct the individual vector and fitness function in the swarm intelligence optimization algorithm; The target index parameters are determined based on the individual vectors and the fitness function.

[0010] According to an index construction method provided by the present invention, the step of constructing individual vectors and fitness functions in a swarm intelligence optimization algorithm based on the fusion features includes: Based on the scale of vector data, the scale of raster data, the vector domain features in the fusion features and the raster domain features in the fusion features, individual vectors in the swarm intelligence optimization algorithm are constructed. Based on the query efficiency, storage overhead, and feature discrimination information in the fused features of the index parameters, a fitness function is constructed in the swarm intelligence optimization algorithm; the feature discrimination information is used to characterize the difference between vector domain features and raster domain features in the fused features.

[0011] According to an index construction method provided by the present invention, the method further includes: The feature gradient vector is determined based on the fused features; the feature gradient vector is used to update the individual vectors in the swarm intelligence optimization algorithm.

[0012] According to an index construction method provided by the present invention, constructing an index for the vector data and an index for the raster data based on the target index parameters includes: Based on the target index parameters, construct an R-tree type index for the vector data; Based on the target index parameters, construct a pyramid-type index for the raster data.

[0013] According to an index construction method provided by the present invention, the method further includes: Establish a cross-reference matrix between the R-tree type index and the pyramid type index.

[0014] According to an index construction method provided by the present invention, after constructing the index of the vector data and the index of the raster data based on the target index parameters, the method further includes: Based on the user's query request, determine the user's query scope and query precision; The target index path is determined based on the user's query range and query precision. Based on the target index path, perform queries for vector data and raster data.

[0015] This invention also provides an index construction method, comprising the following modules: The acquisition module is used to acquire vector data and raster data; The fusion module is used to obtain the fusion features of the vector data and the raster data based on the vector data and the raster data; The determination module is used to determine the target index parameters based on the fusion features and the swarm intelligence optimization algorithm; A construction module is used to construct the index of the vector data and the index of the raster data based on the target index parameters.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements an index construction method as described above.

[0017] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the index construction method as described above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the index construction method as described above.

[0019] The index construction method, apparatus, and device provided by this invention fuse vector data and raster data to obtain fused features that combine the fine-grained geometric details of vector data with the large-scale texture information of raster data. This effectively breaks down the barriers between vector and raster data, enabling efficient collaborative management of both fine-grained vector and large-scale raster data. Furthermore, target index parameters are determined based on the fused features and a swarm intelligence optimization algorithm, effectively improving the adaptability of index parameters to data features. This provides a technical solution that balances performance and adaptability for multi-source spatial data indexing, solving the problems of fragmented management, poor adaptability, and low efficiency in related technologies. The resulting index effectively addresses both fine-grained and large-scale retrieval needs. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts illustrating the index construction method provided by this invention.

[0022] Figure 2 This is the second flowchart of the index construction method provided by the present invention.

[0023] Figure 3 This is the third flowchart of the index construction method provided by the present invention.

[0024] Figure 4 This is a schematic diagram of the index building device provided by the present invention.

[0025] Figure 5 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0027] The following is combined Figures 1-5 The present invention describes the index construction method, apparatus, and device.

[0028] To facilitate a clearer understanding of the technical solutions of the various embodiments of this application, some technical content related to the various embodiments of this application will be introduced first.

[0029] In related technologies, geographic information data management often employs a parallel approach of vector and raster data structures. In practice, this is mostly achieved through common coordinate transformation rules to unify disparate coordinate formats into a universal projection system. To facilitate retrieval and storage, some systems initially divide the data based on geographic boundaries or grid dimensions, ensuring rapid location of regions during subsequent data queries or fusion. These methods typically integrate hardware sensor networks to acquire raw external data, importing it online or offline into a central or distributed storage environment, and then using database or file management systems designed for high-concurrency applications for unified maintenance.

[0030] In the data processing and feature extraction stages, many solutions employ different semantic analysis processes for vector levels and image textures. The vector portion typically relies on tree-like or minimum bounding rectangle-based index structures to improve feature retrieval speed, while rapid rendering of large-scale imagery utilizes hierarchical scaling pyramid tiling techniques, progressively splitting the same original map across several resolution levels. In some richly imaged or topographically complex scenes, high-coverage remote sensing data is also combined to enhance details. Programs often perform geometric correction and blur removal during image preprocessing, or associate image sections through attribute tables at the vector level, thereby achieving comprehensive access to different scales and levels of precision.

[0031] In terms of index optimization, early systems relied solely on simple tree-like or grid-based strategies for feature localization, leveraging conventional database query mechanisms to improve search efficiency. As data volumes have increased, some teams have introduced high-performance server clusters or distributed computing frameworks. In a multi-node environment, massive vector features and large-format imagery are divided into smaller data blocks, and load balancing or block scheduling techniques are used to place the most frequently accessed data on high-speed nodes, ensuring query throughput and response speed. Simultaneously, the application of graphics processing units in geographic information systems has become more widespread. Parallelization methods are used to accelerate projection calculations, region rendering, or feature extraction processes, minimizing the computational burden of processing large-scale scenes while meeting real-time application requirements.

[0032] In terms of distributed caching and adaptive scheduling, some systems perform statistical analysis on query requests to predict potential access hotspots and proactively load vector or tile data of important regions into a high-bandwidth storage layer to reduce the number of times remote resources are repeatedly read.

[0033] In terms of multi-source data fusion, while existing technologies can achieve preliminary unification of different coordinate systems and data levels, they often lack deep correlation methods between fine-grained vector features and large-view raster imagery, easily leading to query bottlenecks under high concurrency and cross-domain access. Some solutions rely on traditional tree structures or pyramid tile storage, which, while enabling fast access to single features or images, may struggle to handle flexible switching between multi-scale data when the system needs to simultaneously retrieve large-scale textures and floor-level details. For the dynamic scheduling of distributed computing nodes, existing methods are mostly based on simple load balancing or caching strategies, lacking adaptive identification of differences in data features, easily resulting in inefficient allocation of computing resources. Hierarchical or multi-level index structures may also exhibit overlapping layers and data redundancy in the face of ever-expanding vector building data and high-resolution imagery. With the continuous growth of data volume and application demands, the existing systems' support for real-time performance and high accuracy remains insufficient.

[0034] Figure 1 This is one of the flowcharts illustrating the index construction method provided by this invention, such as... Figure 1 As shown, the method includes the following: Step 101: Obtain vector data and raster data.

[0035] Specifically, raw vector data and large-scale raster images can be acquired by deploying high-precision sensors, such as GNSS / GPS receivers, IMU inertial measurement units, lidar, or high-resolution cameras.

[0036] Step 102: Obtain the fusion features of vector data and raster data based on the vector data and raster data.

[0037] Specifically, in this embodiment of the application, by fusing vector data and raster data, a fused feature is obtained that combines "fine-grained geometric details of vector data" and "large-scale texture information of raster data". This breaks down the barriers between vector data and raster data, achieves deep collaboration of dual-domain features, and takes into account both fine-grained details and large-scale global retrieval needs.

[0038] Step 103: Determine the target index parameters based on the fusion features and the swarm intelligence optimization algorithm.

[0039] Specifically, traditional index parameters rely on manual setting, such as using the same index parameters for densely built-up areas and open areas, leading to low retrieval efficiency or loss of details, and failing to adapt to different data characteristics. In contrast, this application determines the target index parameters based on fusion features and swarm intelligence optimization algorithms, which can achieve matching between index parameters and data features, solve the shortcomings of empirical index parameter setting, and enable the final index to effectively balance fine-grained and broad-based retrieval needs.

[0040] Step 104: Construct the indexes for vector data and raster data based on the target index parameters.

[0041] Specifically, in this embodiment, dedicated indexes for vector and raster are constructed according to the target index parameters. For example, a vector tree index is constructed to support floor-level precision queries, and a raster pyramid index is constructed to support province-level large-scale rendering, thereby effectively balancing fine-grained and large-scale retrieval needs.

[0042] The method described in the above embodiments, by fusing vector data and raster data, obtains a fused feature that combines the fine-grained geometric details of vector data with the large-scale texture information of raster data. This effectively breaks down the barriers between vector and raster data, enabling efficient collaborative management of both fine-grained vector and large-scale raster data. Furthermore, based on the fused feature and a swarm intelligence optimization algorithm, target index parameters are determined, effectively improving the adaptability of index parameters to data features. This provides a technical solution for multi-source spatial data indexing that balances performance and adaptability, solving the problems of fragmented management, poor adaptability, and low efficiency in related technologies. The resulting index effectively addresses both fine-grained and large-scale retrieval needs.

[0043] In some embodiments, a fusion feature of vector data and raster data is obtained based on vector data and raster data, including: Multi-channel information is extracted from vector and raster data; the multi-channel information includes vector domain information, floor detail information, and satellite image texture information. Based on the multi-channel information, the fusion features are obtained.

[0044] Specifically, this application extracts vector domain information from vector data and raster data respectively. Floor details and satellite image texture information ,in This represents the position coordinates or index identifiers within the partition, thus forming a multi-channel input tensor. The three types of data are merged along the channel dimension to generate the multi-channel input. : in It can characterize the distribution of building shapes based on vector elements. It can represent floor level height or corresponding detailed attributes. It can characterize image texture information.

[0045] By extracting and fusing multi-channel information containing vector domain information, floor detail information, and satellite image texture information from vector and raster data, the limitations of single-domain / single-dimensional features can be effectively overcome, achieving full-dimensional coverage and efficient collaborative management of vector and raster data, effectively balancing the needs of high-precision local data and large-scale global data.

[0046] In some embodiments, the fusion features are obtained based on multi-channel information, including: Multi-channel information is input into the basic convolutional module to obtain initial feature information; the basic convolutional module assigns different convolutional kernel weights to each channel of information; The initial feature information is input into a multi-layer convolutional module to obtain the feature information output by each convolutional layer, and the first feature information is then used as the input. The feature information output by the convolutional layers and the features from layer 1 to layer 2 The feature information output from the -1 layer convolution is fused to obtain the fused feature; It is a positive integer greater than or equal to 2.

[0047] Specifically, this application extracts vector domain information from vector data and raster data respectively. Floor details and satellite image texture information The three types of data are merged along the channel dimension to generate a multi-channel input. Next, a basic convolutional module capable of handling features from different channels is constructed. An independent convolutional kernel weight matrix is ​​assigned to each channel, with the kernel size and stride matched to floor-level precision and the large viewport texture range. Standard activation functions are employed. The convolutional output is non-linearly mapped to ensure sufficient expressive power in both spatial coordinate and channel dimensions. This basic convolutional module ensures the extraction of spatial morphology and texture details in local regions, providing initial feature mappings for subsequent feature fusion.

[0048] Optionally, after completing the basic convolution, this embodiment constructs multi-level convolutions and utilizes a skip connection mechanism to achieve a comprehensive expression of high-level semantics and low-level geometric details, based on the requirements of multi-level feature extraction. Optionally, the first... The feature mapping of the output of the convolutional layer is , and the first The feature maps of the layers are fused in a skip-step manner to generate the fused product. .

[0049] The fusion form is given by the following formula: in Indicates the first The output obtained directly from the convolution operation. The weight coefficients representing cross-layer fusion are introduced by using adjustable parameters in multi-level feature fusion and skip connections. This enables dynamic weighting of the importance of features at different resolutions, meaning that multi-level fusion can fully preserve details such as building outlines and the global distribution of large-scale textures.

[0050] Optionally, the generated multi-level fusion feature maps can be dimensionally compressed to output fusion feature maps for subsequent index parameter selection and optimization. The fusion feature maps contain both floor-level details provided by high-precision vectors and overall texture and background information provided by a large-scale raster, ensuring that they can quickly distinguish feature differences between different partitions when combined with the BFO optimization algorithm in the future.

[0051] It should be noted that this application introduces a multi-channel processing approach for feature extraction, integrating vector shape descriptions, floor level precision elements, and raster texture information into a multi-dimensional input. In the initial convolutional module, an independent convolutional kernel is configured for each channel, allowing specialized operators to capture both floor-level geometric details and large-area image textures. Subsequently, through multi-level networks and skip-style fusion, feature maps of different resolutions are coordinated and combined to generate fused features that possess both fine-grained and global representation capabilities. These fused features not only reflect the precise shape of local building outlines but also the overall distribution of the large-scale geographical background, thus providing a discriminative basis for multi-scale data during the parameter search and index optimization stages.

[0052] The method described above achieves cross-level feature collaboration through multi-level convolution and the use of skip connection mechanisms, breaking through the limitations of traditional single-level convolution. It effectively integrates low-level geometric details with high-level semantic information, taking into account both fine-grained and large-scale requirements, and realizing efficient collaborative management of multi-source geographic information data within the same system.

[0053] In some embodiments, before extracting multi-channel information from vector data and raster data, the method further includes: Vector and raster data are projected onto the same coordinate system and aligned and spatially partitioned.

[0054] Specifically, before fusing vector and raster data, this application requires projecting both data onto the same coordinate system and performing alignment checks and spatial partitioning. In other words, this application integrates the two major data domains of vector and raster data through multi-stage processing to achieve unified management of high-precision floor information and large-scale image data within the same system. First, vector and image data from different sources are standardized through coordinate projection and alignment mechanisms. This process corrects the original data scattered across various coordinate systems to a unified reference system, ensuring alignment accuracy based on cumulative distribution analysis and threshold comparison, providing a consistent geographical benchmark for multi-channel encoding and index fusion. Administrative or grid-based strategies are used to spatially partition the data, enabling more efficient parallel or block scheduling in the data feature extraction and index construction stages, reducing the burden of repeated scanning of large-scale data.

[0055] For example, this application obtains the original vector dataset. With raster datasets Then, record each and The coordinate system identifier. For and Extract the minimum and maximum values ​​from all coordinates in the array, and let... These represent the minimum and maximum latitude and longitude values ​​of all data, forming basic statistical parameters that provide boundary references for subsequent coordinate unification. Based on the recorded coordinate system identifiers, and Projected onto a unified coordinate reference system .make Let the original coordinates be denoted as , and let . Representing a unified coordinate system The coordinates are determined by the projection transformation matrix. , used to Switch to The transformation is expressed by the following formula: in Indicates the mapping parameters. For and Applying the above matrix transformations respectively, we obtain and This ensures that all data are consistently represented in the same coordinate system.

[0056] Optionally, alignment checking is used to measure how well all data matches in a unified coordinate system. Let Represents the projection matrix Point The mapping makes Indicates in The reference positioning is below. An error accumulation function is introduced. The overall coordinate alignment deviation is characterized by double integrals: in This represents the L2 norm distance. (The rest of the text appears to be a list of terms or tags, possibly related to distance or frequency Perform threshold comparison, if If the value exceeds the preset upper limit, update the projection matrix. Or adjust the data precision until The requirements are met. This process ensures that the vectors and raster maintain an accurate mapping relationship during subsequent spatial partitioning.

[0057] Optionally, after completing the alignment check, based on and The coordinate range is used to partition the space. Let... Represents the set of administrative division boundaries, or defines a partitioning function in grid mode. Calculate the partition number to which each element or cell belongs. It can be defined by the following formula: in and Indicates that the mesh cell is in Step size in coordinate system Indicates the number of horizontal grid lines. This indicates a round-down operation. It represents the coordinate values ​​of an element or pixel in a unified coordinate system.

[0058] The method described in the above embodiments, by unifying the coordinate system, alignment verification, and spatial segmentation of vector and raster data, can effectively ensure that vector and raster data are correlated and without misalignment, providing a reliable data foundation for subsequent accurate extraction of multi-channel information and effective generation of fusion features, and avoiding distortion of fusion features.

[0059] In some embodiments, the target index parameters are determined based on fusion features and swarm intelligence optimization algorithms, including: Construct individual vectors and fitness functions in swarm intelligence optimization algorithms based on fusion characteristics; The target index parameters are determined based on the individual vectors and the fitness function.

[0060] Specifically, after feature fusion, this application maps the fused features to the multidimensional search space of the BFO algorithm, iteratively searching for the optimal combination of index structure parameters in the high-dimensional search space. The initial attributes of each individual are determined by referencing the scale factors of the vector and raster domains and combining them with the feature scalars output by the convolutional network to form the individual's initial search vector. During the search, the individual's fitness is jointly measured by query efficiency, storage overhead, and feature discriminative power, and adaptive search is performed in the multidimensional parameter space using global replication and local feeding strategies. This global-local iteration, combined with gradient information from multi-channel output, ensures that each update round takes into account both floor details and large-scale texture, resulting in an index configuration that combines high query efficiency with low storage burden.

[0061] For example, a vector field scaling factor is defined in the embodiments of this application. and raster size factor These two factors are respectively compared with the feature scalar output of the multi-channel convolutional encoder. and By combining the results, an individual-level initialization matrix is ​​formed. .make Indicates the dimension of the index structure to be searched, let This represents the initial set of attributes for each individual. Constructing the BFO individual set. , For the number of individuals. individual The initial attribute vector λ s It can be calculated using the following formula: in and This represents the hierarchical mapping coefficient, which is adjusted... and It can control the degree of influence of the vector domain and the raster domain on individual initialization.

[0062] Optionally, the output fused feature map is mapped to the fitness function of each individual, allowing the individual to reference feature weights from floor details and image texture during the search. Define the index structure parameter vector. ,make This represents a composite metric for actual query efficiency and storage overhead. This represents the feature discrimination metric extracted from the multi-channel convolutional encoder, corresponding to the difference between vector floor levels and large-scale textures. A fitness function is then constructed after synthesis. : in and As weighting coefficients, through Balancing query efficiency with multi-channel feature discrimination, the BFO algorithm can combine floor precision and large-scale texture requirements when filtering index structures, thus achieving a quantitative evaluation of overall performance.

[0063] Optionally, BFO individual set Its index parameter vector The fused features are mapped to the evaluation function to form an initial evaluation of query efficiency and storage overhead. Let... This represents a query efficiency metric, denoted as . This represents the storage overhead metric, establishing an initial comprehensive evaluation and preparing for subsequent iterations. For each individual... Extract the index parameter vector from the previous iteration. This serves as the initial value, laying the foundation for this round of evaluation.

[0064] Feature mapping output from a multi-channel convolutional encoder Embedded into the fitness evaluation process of the BFO algorithm, it enables the algorithm to distinguish between differences between floor deformation and large-scale raster texture. A fitness function is established. for The impact of multi-channel features on index applicability is characterized by the following formula: in and They represent the first The contribution of each index parameter to query efficiency and storage overhead. and These are the weighting coefficients for both. Representation of feature map exist Gradient information at the location, This represents the feature mapping function for the current index parameter.

[0065] Based on this, the perception of differences in floor and texture during the search can simultaneously incorporate both broad and fine-grained requirements into the fitness evaluation, ensuring the balanced performance of the index structure in multi-level scenarios.

[0066] After defining the fitness function, a multidimensional search strategy is constructed. Let the individual... exist The position in the dimensional search space is Let the current generation be .

[0067] A multi-source displacement operator is constructed by combining a global replication strategy with a local feeding strategy. This enables adaptive movement of BFO individuals in a high-dimensional parameter space. The update rules perform a global search and a local fine-tuning sequentially to avoid getting stuck in extreme values ​​through a simple local search.

[0068] After constructing the individual vectors and fitness functions in the swarm intelligence optimization algorithm, and combining the feature judgment of the multi-channel convolutional encoder, the BFO individuals can take into account both fine-grained floors and large-scale grids in high-dimensional space, and enhance the global gradient search of the index structure in complex scenarios.

[0069] The method described above integrates the fine-grained geometric information and global texture features of both types of data through multi-channel convolution and skip connections, generating the fusion features of vector and raster data. The individual vectors and fitness functions of the swarm intelligence optimization algorithm (BFO) are constructed based on these fusion features, ensuring that parameter search always revolves around the true features of vector and raster data, avoiding blind exploration and improving the compatibility between index parameters and data from the source. This application, by first constructing individual vectors and fitness functions and then iteratively searching for target parameters, leverages the global replication and local feeding strategies of the BFO algorithm to balance the global distribution and local details of vector and raster fusion features in a high-dimensional parameter space. Individual vectors provide the initial search direction, and the fitness function provides a quantitative evaluation standard. The combination of these two allows the algorithm to escape local extrema while accurately converging to the globally optimal parameters that meet multiple needs. The final output target index parameters can simultaneously support floor-level fine-grained retrieval and large-scale raster rapid rendering.

[0070] In some embodiments, it also includes: The feature gradient vector is determined based on the fusion features; the feature gradient vector is used to update the individual vectors in the swarm intelligence optimization algorithm.

[0071] Specifically, after completing the fitness calculation, this application uses the global replication strategy and local mining strategy based on the BFO algorithm to analyze individual... Corrections are made during updates to the index parameter vector space.

[0072] Optionally, during the update process, the representation considers the gradient information of the multi-channel convolutional network, and corrects the individual search direction based on a visualization feedback mechanism. Let Represents an individual The amount of index parameter modification, let This indicates that a multi-channel convolutional network is represented as an individual f s The provided feature gradient vectors have the following update relationships: in This is the fusion coefficient, used to balance the internal update volume of the BFO algorithm. With convolution gradient The degree of coupling. Through this coupling update mechanism, after each iteration, not only is the individual position updated based on local and global search information, but it is also adaptively corrected based on the feature gradient of the multi-channel convolutional network, ensuring consistent learning ability for complex floor data and large-scale raster features.

[0073] The method described in the above embodiments updates the index parameter vector of an individual by combining a global copy strategy and a local feeding strategy during the iteration process of the BFO algorithm; and adaptively corrects the individual search direction based on the feature gradient information provided by the multi-channel convolutional network, so that the individual vector update no longer depends on general rules, but is deeply aligned with the data characteristics, solving the problem of the disconnect between parameter optimization and data features in existing technologies, and fundamentally improving the accuracy of index parameter search.

[0074] In some embodiments, the construction of indexes for vector data and raster data based on target index parameters includes: Based on the target index parameters, construct an R-tree type index for the vector data; Construct a pyramid-type index for the raster data based on the target index parameters.

[0075] Specifically, this application is based on the obtained optimal index parameter vector The structure type of the vector field index is determined by information such as its partition scale and capacity threshold. The tree-based solution is first based on the building. Attributes such as floor number define a minimum bounding rectangle for each building element. .make Indicates the level number of the index, let Indicates the first The minimum number of bounding rectangles within a layer node. To reduce spatial overlap between layer nodes, an overlap metric function is defined. And perform area intersection integrals on each pair of smallest enclosing rectangles in a double summation form: in Indicates an indicator function, when Located in the minimum bounding rectangle Internal time take Otherwise take By minimizing the construction of each level of index node. It can significantly reduce at the floor level. The data overlap between tree layers avoids the problem of excessive splitting in densely populated areas of tall buildings, which is a problem of traditional methods.

[0076] After completing the construction of the vector field index, The raster data is divided into blocks and compressed at multiple levels to form a pyramid-shaped hierarchical index. Let represent the total number of pyramid levels. Indicates the first The scaling factor of the layer corresponds to the number of layers. Layer grid segmentation precision. Using the segmented grid tiles. Construct the first Layer indexing, and recording tiles in the index nodes. The index defines the scope, tile location, and corresponding parent or child layer relationships. It supports coarse-grained retrieval at higher levels and fine-grained detail retrieval at lower levels, avoiding the resource consumption of loading full-resolution data at once. Based on this pyramid index, it can quickly locate tile blocks for high-precision areas at the floor level, and schedule only high-level compressed data for large-scale background information.

[0077] Alternatively, it can also be based on building The mapping relationship between the tile number and the tile number is... The tree and pyramid tile index entries are cross-referenced. Let... Indicates building identification, Represents the tile identifier and defines the cross-index reference function. Used to determine building The pyramid level and tile number corresponding to the partition.

[0078] To improve search speed, Constructing sparse matrices ,in This represents the total number of entries in the building index. This represents the total number of pyramid tiles. (Matrix) Each element is given by the following formula: when At that time, it indicates the building. With tiles Cross-domain relationships exist within multi-level indexes; when A value of 0 indicates that there is no direct retrieval relationship between the two. This is achieved by adjusting the query parameters during the query phase. It performs fast lookups on non-zero elements in the index, enabling joint access to the building index and the grid pyramid index. This allows for precise queries of floor information within a small scope, while also associating high-level compressed tiles in a large scope query, thus balancing global and local retrieval needs.

[0079] After completing the cross-referencing of vector and raster multi-level indexes, the entire index structure is deployed to a distributed storage environment. A block scheduling mechanism is adopted, caching index nodes or tiles in high-frequency areas to high-speed storage nodes and placing low-frequency areas on ordinary storage nodes, achieving coherent hierarchical retrieval management. Compared to existing general-purpose index structures, this solution is more suitable for floor-level and large-scale raster scenarios. Trees provide fine-grained representation of vector data, while pyramid hierarchies enable rapid scaling and aggregation of large-scale raster data, through cross-reference matrices. By establishing a close relationship between the two types of indexes, the retrieval of irrelevant data is significantly reduced, further improving query efficiency and storage utilization.

[0080] The method described in the above embodiments constructs multi-level indexes for both the vector and raster domains based on the obtained optimal index parameter scheme. For the vector domain, a tree-structured scheme is adopted, and excessive splitting of densely populated high-rise areas is reduced by minimizing overlap. In the raster domain, a pyramid hierarchical index is introduced, dividing the high-resolution image into tiles and recording them in the index nodes at different scaling levels. To achieve unified retrieval between buildings and tiles, the system constructs a sparse matrix to identify cross-domain associations. During real-time queries, the corresponding tree nodes or pyramid tile nodes can be directly accessed through this matrix. Finally, caching and scheduling are performed in a distributed storage environment, placing the most frequently accessed tiles or index nodes on high-speed nodes to improve query speed and redundancy fault tolerance.

[0081] In some embodiments, after constructing the indexes for vector data and raster data based on the target index parameters, the method further includes: Based on the user's query request, determine the user's query scope and query precision; Determine the target index path based on the user's query scope and query precision; Based on the target index path, perform queries for vector and raster data.

[0082] Specifically, once the index structure is deployed, the real-time query phase begins. This application determines whether the query prefers fine-grained vector or coarse-grained raster based on the user-input target range, resolution, or floor precision, while also considering metrics such as query area and access overhead for indexing decisions. If "floor priority" is determined, the system quickly assigns the query to the tree index for local detailed retrieval; if "large-scale coarse-grained" is deemed more optimal, the request directly connects to the higher-level tiles of the pyramid index for rapid rendering. Subsequently, if sudden changes or batch requests occur during the query process, the index scheduling strategy is fine-tuned, utilizing gradient information from multi-channel feature maps to balance fine-grained protection of local areas with global storage efficiency. Based on this overall system architecture process, retrieval performance is optimized and improved.

[0083] For example, after constructing indexes for vector data and raster data, this application can receive real-time query requests from users. ,extract target range With resolution requirements .make Define the coverage area of ​​the query on the two-dimensional plane, let This indicates the desired floor level precision or raster resolution. Based on the established multi-level index structure, this is determined first. It falls into either the category of floor precision priority or broad coarse-grained priority, preparing for subsequent index scheduling.

[0084] Optionally, for quantitative description Define the corresponding geographic coverage scale and the function to calculate the query area. : in Indicates an indicator function, when Falling within the query range Time value Otherwise .

[0085] Optionally, an index decision cost function can be introduced to select the most suitable index path. This provides a comprehensive measurement of the access overhead of vector indexes and pyramid indexes. This indicates the overhead of calling the vector index. This indicates the cost of calling the pyramid index, and the optimal index path is determined by the following piecewise function. : in This represents the weighting coefficients for query coverage and floor precision during index scheduling. If Significantly lower than If the vector index is selected, it means that the vector index is more effective at the floor level; otherwise, switch to the pyramid high-level index to complete a wide range of low-precision queries.

[0086] After making the decision, the policy mapping generated by the BFO algorithm's collaborative iteration is invoked to... Further verification and fine-tuning will be conducted at the implementation level. If it is determined to be a vector index path, then it relies on the floor number and building. Attributes from Local retrieval is performed in a tree or grid index; if If the path points to the pyramid index, then higher-level tiles are loaded for large-scale data rendering. The visualized gradient information generated by the BFO algorithm is... Real-time corrections are performed to ensure adaptive performance of index scheduling when receiving multiple batches of queries continuously.

[0087] The method described in the above embodiments receives real-time query requests, extracts the query range and resolution requirements, determines whether the query is floor-level precision priority or large-scale coarse-grained priority, selects the corresponding tree-type multi-level index or pyramid-type hierarchical index, and retrieves related data through the aforementioned cross-index reference matrix, thereby achieving adaptive querying and scheduling for high-precision floor and large-scale raster data.

[0088] For example, such as Figure 2 and Figure 3 As shown in the figure, this application provides an index construction method, which is as follows: First, raw vector and raster data are acquired through a hardware sensor network, while offline data is centrally imported into the main control server via the network. During the system initialization phase, the main control unit performs a uniform projection transformation on all data based on the coordinate system and latitude / longitude range identified in the sensors or data files. Alignment verification is then performed, using L2 distance accumulation or threshold comparison to ensure accurate overlap between vector features and raster imagery within the same coordinate reference system. If the alignment error exceeds a pre-set upper limit, the projection matrix is ​​iteratively updated or the source data precision is adjusted. Subsequently, the system divides the vectors and rasters according to administrative divisions or equal-step grids. Each feature or cell is assigned a corresponding partition number. After spatial partitioning, the data is organized accordingly in the underlying storage, providing controllable and ordered spatial blocks for subsequent processing.

[0089] After acquiring data from each partition, the data enters a multi-channel input, customizing the building shape distribution, floor or attribute details, and texture information such as satellite imagery from the vector domain into a multi-channel input tensor. In the basic convolutional module, the GPU matches convolutional kernels and strides to different channels, extracting floor-level geometry and large-area textures in parallel. Then, through multi-level convolutions and skip connections, details and global distributions are fused in high-dimensional feature mapping, ultimately compressing to form a fused feature map. At the hardware level, high-speed buses or high-bandwidth memory are used for data transmission and caching to minimize I / O bottlenecks and improve feature extraction efficiency. Specifically, in feature extraction, this application introduces a multi-channel processing approach, integrating vector shape descriptions, floor precision elements, and raster texture information into a multi-dimensional input. In the initial convolutional module, an independent convolutional kernel is configured for each channel, allowing specialized operators to capture both floor-level geometric details and large-area image textures. Subsequently, through multi-level networks and skip connections, feature maps of different resolutions are coordinated and combined to generate a fused feature map with both fine and global representation capabilities. The fused feature map not only reflects the precise shape of local building outlines, but also reflects the overall distribution of a large geographical background, thus providing a basis for distinguishing multi-scale data during the parameter search and index optimization stages.

[0090] This application proposes an improved BFO algorithm for adaptive search of the index parameter space. In this process, the initial attributes of individuals are not only based on conventional indicators such as query efficiency and storage overhead, but also incorporate the feature difference in multi-channel convolution output, enabling individuals to balance the different needs of the vector domain and the image domain during evolution. A global replication strategy combined with a local feeding strategy allows the algorithm to maintain its global exploration capability of the high-dimensional parameter space. Furthermore, by introducing multi-channel feature gradients, the search direction is dynamically corrected, allowing the index configuration to continuously approach a balance between fine-grained and wide-ranging access. This feature-search combination avoids the limitations of traditional indexes that rely solely on the number of features or simple statistical information for partitioning, striving to achieve more effective parameter scheduling in complex data scenarios.

[0091] In terms of index construction and deployment, this application employs a multi-level index based on a tree structure at the vector level, combined with precise control of the minimum bounding rectangle between nodes at each level. This helps to rationally split densely populated floor areas and reduce overlap. At the raster level, a pyramid-style organization is used to achieve multi-level tile management, allowing access from high-level overviews to low-level details to rely on the same block system. The two sets of indexes are then connected by a cross-index reference matrix, enabling a bidirectional mapping between building numbers and tile identifiers. When faced with various query types, the system can quickly determine whether to prioritize the vector tree index or the pyramid image index. Adaptive cache scheduling is used in a distributed storage environment to ensure support for high concurrency and large-scale queries. Through this overall design, this application's solution achieves efficient joint management of multi-source data within a single platform, providing more systematic technical support for dynamic access to floor-level precision and large-scale raster scenes.

[0092] It should be noted that this application, based on a multi-channel mechanism, integrates the vector domain geometry and floor level precision, as well as the large-scale image texture of the raster domain, under the same coordinate reference system. Based on this fused feature map, it iteratively searches index parameters using the BFO algorithm. Existing solutions often employ independent tree structures at the vector level or deploy pyramid-shaped slices at the image level, but struggle to simultaneously balance building-level precision and efficient switching between large-area scenes in multi-scale queries. This application addresses this by configuring a minimum bounding rectangle in the vector index, implementing cross-layer parent-child relationships in the pyramid hierarchy, and using a cross-index reference matrix to establish a bidirectional association between building identifiers and tile numbers, thus forming a large-scale indexing system that synchronously manages both vectors and rasters. Furthermore, by adaptively correcting the search direction of the BFO algorithm using the feature gradients output by a multi-channel convolutional network, it can more flexibly balance the multiple requirements of floor level features and wide-area texture data in terms of storage level and query speed, making it more adaptable than conventional indexing strategies based primarily on feature quantity or tile size. Through this joint optimization, this application exhibits structural features that differ from the traditional "tree index-pyramid tile" decentralized management model in real-time query and distributed scheduling scenarios, enabling high-precision floor access and wide-area low-precision retrieval to be completed within the same process system.

[0093] The method described in the above embodiments simultaneously enables rapid retrieval of fine-grained floor information and wide-area raster imagery within the same system, and dynamically schedules and manages multi-level index structures in a distributed environment to address the complex needs of multi-scale queries. This application effectively improves the synergy between data of different resolutions and types, automatically finds the optimal combination of index parameters in large-scale data environments, and adaptively schedules system resources in real-time query and update scenarios, thereby ensuring the efficiency of multi-source data management while fully meeting the needs of multi-type retrieval.

[0094] It should be noted that this application is based on the technical concept of multi-level indexing of vector and raster, which emphasizes the integration and dynamic management of data with different precision and multiple types, and also has potential for expansion beyond the field of geographic information.

[0095] For intelligent transportation systems, multi-channel convolution can be used to fuse and analyze road plane and street view images, providing more refined perception capabilities for autonomous driving and traffic flow scheduling. This allows the system to freely switch between large-scale road network search and local lane-level positioning. Simultaneously, this multi-level indexing for rapid scheduling of data with different behavioral characteristics or varying precision is also beneficial for intelligent logistics warehousing and distribution planning. By unifying the management of complex warehouse structures and map information of external distribution areas, the efficiency of flexible scheduling can be improved.

[0096] In the fields of UAV mapping and disaster emergency response, multi-channel networks can be used to extract structural features or building damage information at disaster sites, while pyramid indexes can provide multi-level tiling when the resolution of aerial imagery is insufficient. By using cross-index references, damaged floors or affected tiles can be quickly located, thus assisting emergency response teams in determining rescue priorities in a short time. The built-in BFO algorithm of the technical solution can still adaptively re-search and optimize index parameters under conditions of frequent updates and variable distribution of on-site data, achieving dynamic adjustment without interrupting the entire system.

[0097] For urban planning or smart city management, the core lies in how to achieve efficient correlation between adjacent areas and geographic features. This solution's vector index and spatial partitioning mechanism can interface with sensor networks or property management systems. Combined with building number mapping, this allows decision-making platforms to accurately query information about specific buildings or floors when dealing with macro-level issues such as public facility layout or population distribution. Furthermore, by integrating multi-scale image data based on pyramid-layered tiling, planners can not only observe the overall layout from a macro perspective but also perform fine-grained visual monitoring of key areas. If applied to indoor navigation or robot localization, this solution can leverage its parallel processing capabilities for multi-level features and establish cross-index correlations between vector elements such as rooms and corridors on each floor and the background images collected by sensors. This allows service robots to more smoothly complete path planning and target recognition in high-density crowds or dynamic environments. In other words, this multi-level indexing method is not limited to outdoor or traditional geographic information systems. Combining deep feature extraction and adaptive parameter optimization, it offers new possibilities for various scenarios requiring refined retrieval and large-scale integration.

[0098] The index building apparatus provided by the present invention is described below. The index building apparatus described below can be referred to in correspondence with the index building method described above. The index building apparatus of the embodiments of this application is as follows: Figure 4As shown, it includes: Acquisition module 410 is used to acquire vector data and raster data; The fusion module 420 is used to obtain the fusion features of vector data and raster data based on vector data and raster data; The determination module 430 is used to determine the target index parameters based on the fusion features and the swarm intelligence optimization algorithm; Module 440 is used to construct indexes for vector data and raster data based on the target index parameters.

[0099] Figure 5 A schematic diagram of the physical structure of an electronic device is provided. This electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540. The processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can invoke logical instructions in the memory 530 to execute an index construction method. This method includes: acquiring vector data and raster data; obtaining fusion features of the vector data and raster data based on the vector data and raster data; determining target index parameters based on the fusion features and a swarm intelligence optimization algorithm; and constructing indexes for the vector data and raster data based on the target index parameters.

[0100] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a 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 to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0101] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the index construction method provided by the above methods. The method includes: acquiring vector data and raster data; obtaining fusion features of the vector data and raster data based on the vector data and raster data; determining target index parameters based on the fusion features and a swarm intelligence optimization algorithm; and constructing an index for the vector data and an index for the raster data based on the target index parameters.

[0102] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the index construction method provided by the above methods, the method comprising: acquiring vector data and raster data; obtaining a fusion feature of the vector data and raster data based on the vector data and raster data; determining target index parameters based on the fusion feature and a swarm intelligence optimization algorithm; and constructing an index for the vector data and an index for the raster data based on the target index parameters.

[0103] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0104] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0105] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; 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; and these 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. An index construction method, characterized in that, include: Acquire vector and raster data; Based on the vector data and raster data, the fusion features of the vector data and raster data are obtained; Based on the fusion features and swarm intelligence optimization algorithm, the target index parameters are determined; Based on the target index parameters, construct the indexes for the vector data and the raster data.

2. The index construction method according to claim 1, characterized in that, The step of obtaining the fusion features of the vector data and raster data based on the vector data and raster data includes: Multi-channel information is extracted from the vector data and raster data; the multi-channel information includes vector domain information, floor detail information, and satellite image texture information. The fusion feature is obtained based on the multi-channel information.

3. The index construction method according to claim 2, characterized in that, The step of obtaining the fusion feature based on the multi-channel information includes: The multi-channel information is input into the basic convolution module to obtain initial feature information; the basic convolution module assigns different convolution kernel weights to each channel information. The initial feature information is input into a multi-layer convolutional module to obtain the feature information output by each convolutional layer, and the first feature information is then used as the input. The feature information output by the convolutional layers and the features from layer 1 to layer 2 The feature information output from the -1 layer convolution is fused to obtain the fused feature; It is a positive integer greater than or equal to 2.

4. The index construction method according to claim 2, characterized in that, Before extracting multi-channel information from the vector data and raster data, the method further includes: The vector data and raster data are projected onto the same coordinate system and aligned and spatially divided.

5. The index construction method according to any one of claims 1-4, characterized in that, The step of determining the target index parameters based on the fusion features and the swarm intelligence optimization algorithm includes: Based on the fusion features, construct the individual vector and fitness function in the swarm intelligence optimization algorithm; The target index parameters are determined based on the individual vectors and the fitness function.

6. The index construction method according to claim 5, characterized in that, The step of constructing the individual vector and fitness function in the swarm intelligence optimization algorithm based on the fusion features includes: Based on the scale of vector data, the scale of raster data, the vector domain features in the fusion features and the raster domain features in the fusion features, individual vectors in the swarm intelligence optimization algorithm are constructed. Based on the query efficiency, storage overhead, and feature discrimination information in the fused features of the index parameters, a fitness function is constructed in the swarm intelligence optimization algorithm; the feature discrimination information is used to characterize the difference between vector domain features and raster domain features in the fused features.

7. The index construction method according to claim 6, characterized in that, The method further includes: The feature gradient vector is determined based on the fused features; the feature gradient vector is used to update the individual vectors in the swarm intelligence optimization algorithm.

8. The index construction method according to any one of claims 1-4, characterized in that, The step of constructing the indexes for the vector data and the raster data based on the target index parameters includes: Based on the target index parameters, construct an R-tree type index for the vector data; Based on the target index parameters, construct a pyramid-type index for the raster data.

9. The index construction method according to claim 8, characterized in that, The method further includes: Establish a cross-reference matrix between the R-tree type index and the pyramid type index.

10. The index construction method according to any one of claims 1-4, characterized in that, After constructing the indexes for the vector data and the raster data based on the target index parameters, the method further includes: Based on the user's query request, determine the user's query scope and query precision; The target index path is determined based on the user's query range and query precision. Based on the target index path, perform queries for vector data and raster data.

11. An index construction apparatus, characterized in that, include: The acquisition module is used to acquire vector data and raster data; The fusion module is used to obtain the fusion features of the vector data and the raster data based on the vector data and the raster data; The determination module is used to determine the target index parameters based on the fusion features and the swarm intelligence optimization algorithm; A construction module is used to construct the index of the vector data and the index of the raster data based on the target index parameters.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the index construction method as described in any one of claims 1 to 10.

13. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the index construction method as described in any one of claims 1 to 10.

14. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the index construction method as described in any one of claims 1 to 10.

Images