Multi-source heterogeneous geographic data processing method based on deep fusion algorithm

Abstract

The invention discloses a multi-source heterogeneous geographic data processing method based on a deep fusion algorithm, and the method comprises the steps: collecting multi-source heterogeneous geographic data, and writing the multi-source heterogeneous geographic data into a unified index database; performing coordinate reference normalization, time sequence completion and integrity-correctness joint verification to obtain a standardized geographic data block; semantic-space joint matching is carried out, and candidate cross-modal entity pairs are generated; constructing a cross-modal attention network, extracting fusion features and calculating a posterior fusion confidence coefficient; selecting a fusion result with the highest confidence coefficient from each candidate cross-modal entity pair as a target geographic entity, and generating a unique fusion identifier; and writing the target geographic entity with the unique fusion identifier into a fusion database to obtain a unified geographic entity data set. According to the method, the problems of low data fusion quality, insufficient credibility in the fusion process and difficulty in cross-modal feature matching in the prior art are solved, and the fusion precision, efficiency and credibility are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of geographic information processing and data fusion, and in particular to a multi-source heterogeneous geographic data processing method based on a deep fusion algorithm. Background Art

[0002] With the rapid development of spatial information technology, the means of collecting geographic data and information sources are becoming increasingly diversified, expanding from single vector maps to multi-source heterogeneous information such as remote sensing raster images, lidar point clouds, points of interest (POIs), drone low-altitude data, and social crowdsourcing data. The fusion and unified processing of these multi-source heterogeneous geographic data has become one of the important topics in current geographic information science research and application. Currently, widely used methods mainly include fusion technologies based on traditional data matching and spatial overlay, as well as intelligent fusion technologies represented by deep learning. However, due to the large modal differences, diverse semantic description methods, and significant differences in spatial resolution and time scale among multi-source heterogeneous geographic data, traditional fusion methods are mostly centered on single-modal data and cannot achieve refined fusion of different modal data in terms of semantics and dimensions. Deep learning methods usually do not fully consider the cross-modal interaction mechanism between data, which limits the accuracy and confidence of the fusion results.

[0003] CN115269947A discloses a data processing method for multi-source heterogeneous geographic information, which is applied to earth resistivity measurement in geological exploration. Multi-channel data acquisition is achieved by switching between repeaters and intelligent electrodes. However, this method focuses on the physical structure design of sensors and measuring equipment, and does not consider the semantic-spatial association matching and fusion between cross-modal geographic data. It cannot meet the high requirements of complex geographic information applications for fusion accuracy and semantic consistency.

[0004] CN116719898A discloses a geographic entity generation method based on multi-source heterogeneous data. Although it takes into account the standardization and fusion processing of multi-source data, it only performs fusion through a preset entity generation model and does not adopt an effective cross-modal attention mechanism. As a result, the accuracy of the fused data is limited. In particular, for data sources with significant feature differences, the unified fusion effect is poor. In addition, the fusion process is not clear enough for the calculation of geographic entity confidence, and the credibility of data fusion cannot be effectively quantified.

[0005] In summary, existing multi-source heterogeneous geographic data fusion technologies generally have the problem of insufficient accuracy in semantic matching and feature fusion of cross-modal data. The present invention aims to solve the problem of how to efficiently and accurately realize deep fusion processing of multi-source heterogeneous geographic data. Summary of the Invention

[0006] The purpose of this section is to summarize some aspects of the embodiments of the present invention and briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section and the abstract of the specification and the title of the invention of this application to avoid blurring the purpose of this section, the abstract of the specification and the title of the invention, and such simplifications or omissions cannot be used to limit the scope of the invention.

[0007] In view of the above existing problems, the present invention is proposed.

[0008] To solve the above technical problems, the present invention provides the following technical solutions: collecting multi-source heterogeneous geographic data and writing them into a unified index library according to the collected time-space metadata;

[0009] Performing coordinate reference normalization, time series completion, and integrity-correctness joint verification on the written multi-source heterogeneous geographic data to obtain a standardized geographic data block;

[0010] Based on the unique identifier of the geographic entity and the spatial topological constraint, the standardized geographic data block is subjected to semantic-spatial joint matching to generate candidate cross-modal entity pairs;

[0011] Constructing a cross-modal attention network including a grid convolution branch, a vector convolution branch, and a text encoding branch, inputting the candidate cross-modal entity pair into the cross-modal attention network, extracting fusion features, and calculating the posterior fusion confidence;

[0012] Selecting the fusion result with the highest confidence in each candidate cross-modal entity pair as the target geographic entity, and generating a unique fusion identifier for the target geographic entity;

[0013] The target geographic entity holding a unique fusion identifier is written into a fusion database, and a unified geographic entity dataset is output.

[0014] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm described in the present invention, the multi-source heterogeneous geographic data at least includes vector feature data, remote sensing raster image data, lidar point cloud data and POI data.

[0015] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm of the present invention, the multi-source heterogeneous geographic data is written into a unified index library according to the acquisition time-space metadata, including:

[0016] The collected multi-source heterogeneous geographic data are stored in an index library using a spatial quadtree index and GeoHash grid encoding method according to the metadata information of the unified timestamp and spatial reference system;

[0017] The spatial quadtree index is used to manage the vector element data and the lidar point cloud data;

[0018] The GeoHash grid code is used to manage the remote sensing raster image data and the POI data.

[0019] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm of the present invention, obtaining the standardized geographic data block includes:

[0020] A dynamic seven-parameter space transformation model is used to normalize data of different coordinate bases;

[0021] The raster pixels with time series missing are filled by bidirectional time-space interpolation method, and the missing vector attributes are adaptively interpolated based on the cosine similarity of word vectors.

[0022] A joint verification rule base consisting of spatial topology rules, attribute logic rules and entity overlap detection rules is established to eliminate data blocks that do not meet the rules and form standardized geographic data blocks.

[0023] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm of the present invention, generating the candidate cross-modal entity pairs includes:

[0024] Call the improved KD-Tree spatial index to retrieve spatial neighbor entities within the first threshold;

[0025] Use SimHash feature vector comparison to calculate the text semantic Hamming distance, and only retain entities with Hamming distance less than the second threshold;

[0026] Entities that meet the first threshold and the second threshold conditions are combined into candidate cross-modal entity pairs, and their initial matching confidence values ​​are recorded.

[0027] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm described in the present invention, a cross-modal attention network including a raster convolution branch, a vector convolution branch, and a text encoding branch is constructed, including:

[0028] The grid convolution branch uses the ResNet-50 network to extract multi-scale texture features layer by layer;

[0029] The vector graph convolution branch uses a graph convolutional network to extract the topological structure and attribute features of vector data;

[0030] The text encoding branch uses the BERT-Base model to extract contextual semantic vectors;

[0031] The grid convolution branch, the vector convolution branch, and the text encoding branch perform feature interaction through a multi-head Transformer attention mechanism and align in a shared latent space to generate a fused feature vector.

[0032] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm described in the present invention, the candidate cross-modal entity pairs are input into the cross-modal attention network, the fusion features are extracted and the posterior fusion confidence is calculated, including:

[0033] Inputting candidate cross-modal entity pairs into the cross-modal attention network respectively to obtain raster data feature vectors, vector data feature vectors and text data feature vectors;

[0034] Using the cross-modal attention mechanism, the attention weights between different data features are calculated one by one, and the cross-modal feature fusion is completed through weighted summation to obtain the cross-modal fusion feature representation;

[0035] Based on the fusion feature representation, the maximum a posteriori probability method is introduced to calculate the posterior confidence of the fusion result;

[0036] The fused feature representation and posterior confidence of each set of candidate cross-modal entity pairs are output for the next step of entity confirmation and fused label generation.

[0037] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm described in the present invention, the calculation of the posterior fusion confidence includes:

[0038] Calculate the conditional probabilities of different geographic entity categories based on the input fusion feature representation based on the trained feature fusion network;

[0039] The prior probability of the geographic entity category and the conditional probability are combined to determine the posterior confidence of the fused entity through the maximum a posteriori probability method.

[0040] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm of the present invention, generating the unique fusion identifier includes:

[0041] Rank the posterior confidence of each candidate cross-modal entity pair;

[0042] The entity combination with the highest posterior confidence is selected as the final confirmed target geographic entity;

[0043] A unique identifier generation algorithm is used to generate a non-repeatable fusion identifier, wherein the unique identifier generation algorithm uses a hash algorithm that combines spatial position coding and time stamp to generate a unique identifier.

[0044] As a preferred solution of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm of the present invention, the unified geographic entity dataset is obtained, including:

[0045] Read the unique fusion identifier, unified coordinate geometry and attribute dictionary of the confirmed target geographic entity, and generate a record to be stored for each target geographic entity based on the result layer data model of the fusion database. The record fields include the key primary key, spatial field, attribute field and version field.

[0046] Perform concurrent conflict detection based on the unique fusion identifier. If there are records with the same name, compare the new record version number Ver. new 、Old record version number Ver old And the new record confidence Conf new , Old record confidence Conf old , when Ver new >Ver old And the confidence level Conf new –Conf old ≥ confidence difference threshold δ conf If yes, overwrite the old record; otherwise, keep the old record and write it into the history table;

[0047] Use the batch COPY / BULK interface to write the fusion result layer data in a serialized transaction. If the write fails, the entire transaction is rolled back; if the write succeeds, the transaction is committed.

[0048] After the transaction is committed, asynchronous index maintenance is triggered, the R*-Tree primary index and GeoHash bucket cache are incrementally updated, and the in-memory Skip-List is refreshed to improve real-time query efficiency;

[0049] Based on the unique fusion identifier, multi-source metadata is aggregated, geometric topology correction is performed, and a unified geographic entity dataset view is generated by segmenting according to business themes and scales.

[0050] Beneficial effects of the present invention: The present invention effectively solves the problems of low data fusion quality, insufficient credibility of the fusion process and difficulty in cross-modal feature matching in the existing technology through data collection, preprocessing, matching, deep fusion, fusion decision-making and final data output. It realizes high-precision, high-efficiency and high-reliability fusion processing of multi-source heterogeneous geographic data as a whole, and significantly improves the fusion accuracy, efficiency and credibility. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. Those skilled in the art can also derive other drawings based on these drawings without inventive effort. Among them:

[0052] Figure 1 This is a flow chart of the multi-source heterogeneous geographic data processing method based on the deep fusion algorithm shown in the present invention. DETAILED DESCRIPTION

[0053] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings. It is obvious that the described embodiments are only part of the embodiments of the present invention, but not all of the embodiments.

[0054] Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in this field without making any creative work should fall within the scope of protection of the present invention.

[0055] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0056] According to an embodiment of the present invention, Figure 1 The flowchart shown is a multi-source heterogeneous geographic data processing method based on a deep fusion algorithm, which specifically includes the following steps:

[0057] S1. Collect multi-source heterogeneous geographic data and write them into a unified index library according to the collected time-space metadata.

[0058] The collected multi-source heterogeneous geographic data are stored in the index library using spatial quadtree indexing and GeoHash (a spatial indexing algorithm that encodes longitude and latitude coordinates into strings) grid encoding according to the metadata information of the unified timestamp and spatial reference system.

[0059] Spatial quadtree index is used to manage vector feature data and lidar point cloud data;

[0060] GeoHash grid coding is used to manage remote sensing raster image data and POI data.

[0061] In an optional embodiment, the method for constructing a spatial quadtree index includes: performing bounding box calculation on vector elements and LiDAR point clouds; and recursively dividing the spatial region to generate quadtree nodes.

[0062] In an optional embodiment, the GeoHash grid encoding method includes: generating a GeoHash string GH according to the geographic coordinates (x, y) with a precision of p bits p = GeoHash(x,y,p); with GH p As the key, write the grid tile ID and POI record into the hash bucket.

[0063] In an optional embodiment, the method of writing to the unified index library includes: first writing to the spatial quadtree layer, obtaining the node path Path QT ; Then write the GeoHash bucket layer to get the bucket key GH p ; Form a joint primary key Key = T utc ||SRID||Path QT ||GH p Insert data into the NoSQL-Spatial engine in BULK mode (large block data transfer mode), achieving concurrent writes in seconds.

[0064] It should be noted that multi-source heterogeneous geographic data includes at least vector feature data, remote sensing raster image data, lidar point cloud data and POI data.

[0065] As an example, vector feature data includes at least road centerlines, building outlines, and administrative boundaries.

[0066] As an example, remote sensing raster image data includes at least multispectral satellite images and drone orthophotos.

[0067] As an example, the lidar point cloud data includes at least airborne LiDAR (laser radar) terrain points and vehicle-mounted LiDAR street view points.

[0068] As an example, POI data (point of interest data, which is geographic spatial information describing a specific location, including location coordinates, name, address, and category attributes) includes at least the name, category, and latitude and longitude coordinates of the point of interest.

[0069] For example, the airborne LiDAR point cloud collected at 2025-06-01 10:30 UTC:

[0070] Timestamp T utc =20250601T103000Z;

[0071] SRID=4490;

[0072] Meta.sensor = "ALS-60";

[0073] Quadtree path / NE / NE / SW.

[0074] For example, let's collect satellite image slices at 2025-06-01 10:32 UTC:

[0075] Image center (118.773°, 32.065°), GeoHash6 66 = "wtw37p";

[0076] Store in bucket wtw37p, record the slice ID and image resolution 0.5m.

[0077] Preferably, through the above steps, the rapid archiving and retrieval of multi-source heterogeneous geographic data such as vectors, rasters, point clouds and POIs are achieved under the framework of unified timestamp + unified coordinate reference + two-level spatial index (quadtree / GeoHash), which effectively avoids the data island phenomenon, reduces the time consumption of data search and matching during subsequent data fusion, and improves the efficiency and real-time performance of data management and fusion process.

[0078] S2. Perform coordinate datum normalization, time series completion, and integrity-correctness joint verification on the written multi-source heterogeneous geographic data to obtain a standardized geographic data block.

[0079] A dynamic seven-parameter space transformation model is used to normalize data of different coordinate bases;

[0080] The raster pixels with time series missing are filled by bidirectional time-space interpolation method, and the missing vector attributes are adaptively interpolated based on the cosine similarity of word vectors.

[0081] A joint verification rule base consisting of spatial topology rules, attribute logic rules and entity overlap detection rules is established to eliminate data blocks that do not meet the rules and form standardized geographic data blocks.

[0082] For example, the source coordinates (X s ,Y s ,Z s ) and the target benchmark ITRF-2014, applying the dynamic Helmert seven-parameter model:

[0083]

[0084] Among them, ΔX, ΔY, ΔZ are translation parameters, R x ,R y ,R z is the micro-angle rotation parameter, s is the scale parameter, X s,Y s ,Z s is the rectangular coordinate of the source coordinate system, X t ,Y t ,Z t is the rectangular coordinate of the target coordinate system.

[0085] In an optional embodiment, when the grid pixel value V(i,j,t) is missing at time t, the most recent valid previous frame time t is defined. f and the post-frame time t b :

[0086]

[0087] According to the linear time weight coefficient, the grid pixel padding value is obtained:

[0088]

[0089] Missing attributes for vector features From the set of similar elements A={a k} to retrieve the one with the largest cosine similarity:

[0090]

[0091] in, is the corresponding word vector, with a * As a missing fill value, it can achieve semantically consistent attribute completion.

[0092] Furthermore, define the joint verification rule base set

[0093] in, It is a subset of spatial topological rules (such as polygons have no self-intersections and road centerlines cannot cross water bodies). is a subset of attribute logic rules (such as building height is greater than 0, point cloud intensity I∈[0,255]), It is a subset of entity overlap detection rules (e.g. the overlap ratio of similar surface features shall not exceed 5%).

[0094] The rule base is solidified in the form of JSON-Schema+SQL CHECK and is automatically called through triggers.

[0095] Furthermore, the joint verification rule is executed on each of the normalized and padded data blocks:

[0096] If the data does not meet the verification rules, it is marked as illegal and discarded, otherwise it is marked as a legal record and entered into the standardized buffer area;

[0097] Standardize the records in the buffer area according to StdKey=T utc||SRID=4490||GridID is renumbered to form a standardized geographic data block. It should be noted that the standardized geographic data block includes three types of entities: unified coordinate vector elements + semantically complete attributes, continuous multi-temporal raster slices with no missing pixels, and strongly consistent coordinate point clouds + POI records.

[0098] It should be noted that by performing coordinate reference normalization, time series completion and completeness-correctness joint verification on the collected multi-source heterogeneous geographic data, the unified spatial expression of data from different coordinate systems is achieved, and the time series missing or attribute incomplete problems generated during the data collection process are effectively repaired, fundamentally ensuring the integrity and accuracy of the data. This refined preprocessing significantly improves the quality of data before fusion, reduces the possibility of incorrect matching or low-quality fusion results in the subsequent fusion process, and ensures the reliability of the fusion process.

[0099] S3. Based on the unique identifier of geographic entities and spatial topological constraints, perform semantic-spatial joint matching on the standardized geographic data blocks to generate candidate cross-modal entity pairs.

[0100] Establish an improved KD-Tree index for each standardized entity;

[0101] Taking the entity center point as the retrieval center, search for spatial neighbors based on the first threshold (e.g., 50 meters);

[0102] If no entity is found within the distance range, no pairing will be established for the entity;

[0103] Furthermore, for the searched spatial neighbor entities, the SimHash fingerprints of the semantic tags are calculated respectively;

[0104] Compare the Hamming distances of the two fingerprints and retain only the entity pairs whose Hamming distance is less than a second threshold (e.g., 12 bits) to form candidate entity pairs that meet the dual conditions of "spatial proximity + semantic similarity";

[0105] For each candidate entity pair, an initial matching confidence value is assigned based on the principle that the closer the spatial distance, the smaller the semantic distance;

[0106] In an optional embodiment, the confidence is a weighted sum of the spatial distance contribution and the semantic similarity contribution, with the weight parameter β d and β h Defined as β d =0.4,β h =0.6;

[0107] Sort the candidate pairs corresponding to the same source entity from high to low according to the initial confidence value, and retain the top 5 groups;

[0108] Record the source entity, candidate entity and its initial confidence value to form a list of candidate cross-modal entity pairs as input for subsequent deep fusion processing.

[0109] It should be noted that by accurately corresponding and preliminarily fusing different modal data in terms of spatial position and semantic attributes, the problem of large modal differences and difficult matching between heterogeneous data is effectively solved, and the fusibility and matching accuracy between multi-source heterogeneous data are significantly improved, providing high-quality candidate inputs for deep fusion, thereby improving the overall fusion accuracy.

[0110] S4. Construct a cross-modal attention network consisting of a grid convolution branch, a vector convolution branch, and a text encoding branch. Input the candidate cross-modal entity pairs into the cross-modal attention network, extract fusion features, and calculate the posterior fusion confidence. Note that:

[0111] The grid convolution branch uses the ResNet-50 network to extract multi-scale texture tensors from the remote sensing slices of the candidate entity pairs step by step;

[0112] The vector convolution branch uses a graph convolutional network to perform graph embedding encoding on vector topologies such as roads and buildings.

[0113] The text encoding branch uses the BERT-Base model to generate contextual semantic vectors for entity labels and description texts;

[0114] The above three types of features are fed into the multi-head Transformer attention layer to adaptively learn the complementary associations between the modalities, complete feature alignment in the shared latent space, and output the aligned sequence attention weight curve;

[0115] Global pooling is performed on the alignment weight curve in the normalized time dimension to obtain a three-way attention function for raster, vector, and text. The three-way features are then aggregated into a single fused feature representation through a cross-modal weighted gating module.

[0116] Using the maximum a posteriori probability paradigm, the fused features are fed into the trained discriminant model, the entity category prior and conditional likelihood are combined to infer the fusion confidence, and a binary tuple of fused feature vector + confidence is output for the next step of entity confirmation and fusion label generation.

[0117] In an optional embodiment, the candidate cross-modal entity pairs are input into the cross-modal attention network, fusion features are extracted, and posterior fusion confidence is calculated, including:

[0118] The candidate cross-modal entity pairs are input into the cross-modal attention network respectively to obtain the raster data feature vector, vector data feature vector and text data feature vector;

[0119] Using the cross-modal attention mechanism, the attention weights between different data features are calculated one by one, and the cross-modal feature fusion is completed through weighted summation to obtain the cross-modal fusion feature representation;

[0120] Based on the fusion feature representation, the maximum a posteriori probability method is introduced to calculate the posterior confidence of the fusion result:

[0121] Calculate the conditional probabilities of different geographic entity categories based on the input fusion feature representation based on the trained feature fusion network;

[0122] Combining the prior probability and conditional probability of geographic entity categories, the posterior confidence of the fused entity is determined by the maximum a posteriori probability method;

[0123] Finally, the fused feature representation and posterior confidence of each set of candidate cross-modal entity pairs are output for the next step of entity confirmation and fused identity generation.

[0124] As an example, the trained discriminative model is used to calculate the cross-modal attention output and posterior confidence, which is mathematically expressed as:

[0125]

[0126] Among them, C ij is the posterior fusion confidence of the candidate entity pair (i, j), σ(·) is the Sigmoid normalization function, κ is the texture difference penalty factor, and a r (τ), a v (τ), a t (τ) are the grid, vector, and text attention weight vectors at the normalized time index τ, N(·) is the zero-mean Gaussian normalization function, L is the number of Transformer layers, is the l-th layer grid, vector, and text deep embedding vector, ReLU(·) is the linear rectification function, ξ is the text sparse adjustment coefficient, and ||·||1 represent the two-norm and the one-norm respectively;

[0127] C ij ∈(0,1), when C ij When the value is close to 1, it means that the three modalities of raster, vector, and text are highly consistent, the spatial difference is minimal, and the semantics are highly matched, and the fusion reliability is the highest;

[0128] When C ij When the value is close to 0, it means that the three modal features are significantly different or the attention cannot be aligned, and the fusion reliability is the lowest.

[0129] It should be noted that by constructing a cross-modal attention network consisting of a grid convolution branch, a vector convolution branch and a text encoding branch, and using this network to extract fusion features and calculate the posterior fusion confidence, deep fusion and feature interaction of data in the cross-modal feature space are achieved. The cross-modal attention mechanism can accurately capture the potential correlation and feature complementarity between the data of each modality, making the fused features more comprehensive and accurate. The calculated posterior confidence can effectively quantify the credibility of the fusion results, significantly improving the reliability and interpretability of the fusion decision-making process.

[0130] S5. Select the fusion result with the highest confidence in each candidate cross-modal entity pair as the target geographic entity, and generate a unique fusion identifier for the target geographic entity. Note that:

[0131] Rank the posterior confidence of each candidate cross-modal entity pair;

[0132] The entity combination with the highest posterior confidence is selected as the final confirmed target geographic entity;

[0133] A unique identifier generation algorithm is used to generate a non-repeatable fusion identifier, wherein the unique identifier generation algorithm uses a hash algorithm that combines spatial position coding and timestamp to generate a unique identifier.

[0134] It should be noted that by selecting the fusion result with the highest fusion confidence in each candidate cross-modal entity pair as the target geographic entity and generating a unique fusion identifier, automatic screening and efficient confirmation of the fused entity are achieved. This step can automatically eliminate ambiguity or incorrect matches that may occur during the fusion process, improve the accuracy of the fused data, avoid the subjective problems brought about by traditional manual judgment or single rule selection methods, and ensure the consistency and accuracy of the fusion results.

[0135] S6. Write the target geographic entity with a unique fusion identifier into the fusion database, and output a unified geographic entity dataset.

[0136] Read the unique fusion identifier, unified coordinate geometry and attribute dictionary of the confirmed target geographic entity, and generate a record to be stored for each target geographic entity based on the result layer data model of the fusion database. The record fields include the key primary key, spatial field, attribute field and version field.

[0137] Perform concurrent conflict detection based on the unique fusion identifier. If there are records with the same name, compare the new record version number Ver. new 、Old record version number Ver old And the new record confidence Conf new , Old record confidence Conf old, when Ver new >Ver old And the confidence level Conf new –Conf old ≥ confidence difference threshold δ conf If yes, overwrite the old record; otherwise, keep the old record and write it into the history table;

[0138] Use the batch COPY / BULK interface to write the fusion result layer data in a serialized transaction. If the write fails, the entire transaction is rolled back; if the write succeeds, the transaction is committed.

[0139] After a transaction is committed, asynchronous index maintenance is triggered, incrementally updating the R*-Tree primary index and GeoHash bucket cache, and refreshing the in-memory Skip-List to improve real-time query efficiency.

[0140] Aggregate multi-source metadata based on unique fusion identifiers, perform geometric topology corrections, and generate a unified geographic entity dataset view by segmentation based on business themes and scales.

[0141] It should be noted that by writing the target geographic entity with a unique fusion identifier into the fusion database and outputting a unified geographic entity dataset, unified storage and efficient sharing of multi-source heterogeneous data are achieved, effectively reducing the complexity of downstream applications calling fusion data, significantly improving the shareability and ease of use of the unified dataset, and providing high-quality unified data support for data analysis and business applications of geographic information systems.

[0142] It should be further explained that, in the acquisition phase, the processing method provided in this embodiment constructs a unified index library based on the acquisition time-space metadata through the unified acquisition and organization of multi-source heterogeneous geographic data such as vector feature data, remote sensing raster image data, lidar point cloud data and POI data, which significantly improves the retrieval efficiency and management convenience of multi-source data, provides an efficient data foundation for subsequent cross-modal data matching and fusion, avoids the data island phenomenon, and significantly reduces the time and computing resource costs of subsequent fusion calculations.

[0143] Secondly, the present invention realizes the standardized processing of data through a combined method of coordinate reference normalization, time series completion and completeness-correctness joint verification. Through the comprehensive application of a dynamic seven-parameter space conversion model, a bidirectional time-space interpolation algorithm and an attribute logic verification rule library, it solves the problems of data reference confusion, missing data and logical errors in the existing technology, improves the intrinsic quality and consistency of the data, and significantly reduces the problems of incorrect matching or poor fusion quality in the fusion stage.

[0144] Based on the unique identification of geographic entities and spatial topological constraints, the present invention performs semantic-spatial joint matching of standardized geographic data blocks. Through improved KD-Tree spatial retrieval and SimHash semantic matching technology, it effectively achieves accurate semantic spatial correspondence between data of different modalities, solving the problem of insufficient matching accuracy caused by large modal differences in traditional technologies. Through precise preliminary matching, it provides higher-quality candidate entity pairs, ensuring the accuracy and efficiency of subsequent deep fusion.

[0145] It should be noted again that an embodiment of the present invention constructs a cross-modal attention network, which realizes deep interaction and precise fusion of features between cross-modal data through the deep fusion of grid convolutional neural networks, graph convolutional networks and text encoding models. The cross-modal attention mechanism effectively captures the intrinsic correlation features and complementary information between different data modalities through the allocation of adaptive weights, and significantly improves the expressive power of fusion features. In addition, the fusion confidence calculated by the maximum a posteriori probability method provides a quantitative reliability basis for the fusion decision, effectively solving the problem that the fusion results in the existing technology are difficult to objectively evaluate, making the fusion process more scientific, credible and well interpretable.

[0146] In terms of fusion result confirmation, the embodiment of the present invention automatically selects the fusion result with the highest confidence among the candidate entity pairs as the final target geographic entity based on the calculated posterior confidence, and automatically generates a globally unique fusion identifier, avoiding the subjectivity and uncertainty problems caused by traditional manual intervention or single rule judgment methods, and greatly improving the consistency, objectivity and accuracy of the fusion results.

[0147] Ultimately, the present invention achieves unified storage management and efficient shared calling of heterogeneous data sources by writing the confirmed target geographic entities into the fusion database and outputting a unified geographic entity dataset. By adopting the OGC standard service interface to open data resources to external systems, the shareability, usability and interoperability of the fused data are significantly improved, and the practical application value of multi-source heterogeneous geographic data is enhanced.

[0148] The aforementioned method for extracting feature vectors and generating fusion identifiers can be performed using methods and means in the prior art and will not be described in detail in this example.

[0149] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A multi-source heterogeneous geographic data processing method based on a deep fusion algorithm, characterized in that: include: Collect multi-source heterogeneous geographic data and write them into a unified index library according to the collected time-space metadata; Performing coordinate reference normalization, time series completion, and integrity-correctness joint verification on the written multi-source heterogeneous geographic data to obtain a standardized geographic data block; Based on the unique identifier of the geographic entity and the spatial topological constraint, the standardized geographic data block is subjected to semantic-spatial joint matching to generate candidate cross-modal entity pairs; Constructing a cross-modal attention network including a grid convolution branch, a vector convolution branch, and a text encoding branch, inputting the candidate cross-modal entity pair into the cross-modal attention network, extracting fusion features, and calculating the posterior fusion confidence; Selecting the fusion result with the highest confidence in each candidate cross-modal entity pair as the target geographic entity, and generating a unique fusion identifier for the target geographic entity; The target geographic entity holding a unique fusion identifier is written into a fusion database, and a unified geographic entity dataset is output.

2. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 1 is characterized in that: The multi-source heterogeneous geographic data includes at least vector feature data, remote sensing raster image data, lidar point cloud data and POI data.

3. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 2 is characterized in that: Writing the multi-source heterogeneous geographic data into a unified index library according to the acquisition time-space metadata includes: The collected multi-source heterogeneous geographic data are stored in an index library using a spatial quadtree index and GeoHash grid encoding method according to the metadata information of the unified timestamp and spatial reference system; The spatial quadtree index is used to manage the vector element data and the lidar point cloud data; The GeoHash grid code is used to manage the remote sensing raster image data and the POI data.

4. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 3 is characterized in that: Obtaining the standardized geographic data block includes: A dynamic seven-parameter space transformation model is used to normalize data of different coordinate bases; The raster pixels with time series missing are filled by bidirectional time-space interpolation method, and the missing vector attributes are adaptively interpolated based on the cosine similarity of word vectors. A joint verification rule base consisting of spatial topology rules, attribute logic rules and entity overlap detection rules is established to eliminate data blocks that do not meet the rules and form standardized geographic data blocks.

5. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 3 is characterized in that: Generating the candidate cross-modal entity pairs includes: Call the improved KD-Tree spatial index to retrieve spatial neighbor entities within the first threshold; Use SimHash feature vector comparison to calculate the text semantic Hamming distance, and only retain entities with Hamming distance less than the second threshold; Entities that meet the first threshold and the second threshold conditions are combined into candidate cross-modal entity pairs, and their initial matching confidence values ​​are recorded.

6. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 1 is characterized in that: Construct a cross-modal attention network consisting of a grid convolution branch, a vector convolution branch, and a text encoding branch, including: The grid convolution branch uses the ResNet-50 network to extract multi-scale texture features layer by layer; The vector graph convolution branch uses a graph convolutional network to extract the topological structure and attribute features of vector data; The text encoding branch uses the BERT-Base model to extract contextual semantic vectors; The grid convolution branch, the vector convolution branch, and the text encoding branch perform feature interaction through a multi-head Transformer attention mechanism and align in a shared latent space to generate a fused feature vector.

7. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 6 is characterized in that: The candidate cross-modal entity pairs are input into the cross-modal attention network, fusion features are extracted and posterior fusion confidence is calculated, including: Inputting candidate cross-modal entity pairs into the cross-modal attention network respectively to obtain raster data feature vectors, vector data feature vectors and text data feature vectors; Using the cross-modal attention mechanism, the attention weights between different data features are calculated one by one, and the cross-modal feature fusion is completed through weighted summation to obtain the cross-modal fusion feature representation; Based on the fusion feature representation, the maximum a posteriori probability method is introduced to calculate the posterior confidence of the fusion result; The fused feature representation and posterior confidence of each set of candidate cross-modal entity pairs are output for the next step of entity confirmation and fused label generation.

8. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 7 is characterized in that: The calculation of the posterior fusion confidence includes: Calculate the conditional probabilities of different geographic entity categories based on the input fusion feature representation based on the trained feature fusion network; The prior probability of the geographic entity category and the conditional probability are combined to determine the posterior confidence of the fused entity through the maximum a posteriori probability method.

9. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 7 is characterized in that: Generating the unique fusion identifier includes: Rank the posterior confidence of each candidate cross-modal entity pair; The entity combination with the highest posterior confidence is selected as the final confirmed target geographic entity; A unique identifier generation algorithm is used to generate a non-repeatable fusion identifier, wherein the unique identifier generation algorithm uses a hash algorithm that combines spatial position coding and time stamp to generate a unique identifier.

10. The multi-source heterogeneous geographic data processing method based on the deep fusion algorithm according to claim 1 is characterized in that: The unified geographic entity dataset obtained includes: Read the unique fusion identifier, unified coordinate geometry and attribute dictionary of the confirmed target geographic entity, and generate a record to be stored for each target geographic entity based on the result layer data model of the fusion database. The record fields include the key primary key, spatial field, attribute field and version field. Perform concurrent conflict detection based on the unique fusion identifier. If there are records with the same name, compare the new record version number Ver. new 、Old record version number Ver old And the new record confidence Conf new , Old record confidence Conf old , when Ver new >Ver old And the confidence level Conf new –Conf old ≥ confidence difference threshold δ conf If yes, overwrite the old record; otherwise, keep the old record and write it into the history table; Use the batch COPY / BULK interface to write the fusion result layer data in a serialized transaction. If the write fails, the entire transaction is rolled back; if the write succeeds, the transaction is committed. After the transaction is committed, asynchronous index maintenance is triggered, the R*-Tree primary index and GeoHash bucket cache are incrementally updated, and the in-memory Skip-List is refreshed to improve real-time query efficiency; Based on the unique fusion identifier, multi-source metadata is aggregated, geometric topology correction is performed, and a unified geographic entity dataset view is generated by segmenting according to business themes and scales.

Images