Space-time multi-source data intelligent fusion system and method

Abstract

The invention relates to the technical field of big data processing, in particular to a space-time multi-source data intelligent fusion system and method, and the method comprises the following steps: S1, collecting multi-source heterogeneous data, extracting time, space and semantic information, and constructing a cross-period semantic evolution graph of the same region; s2, identifying a modal fault section based on an evolution graph, and generating a multi-modal interpolation candidate set of structure compensation under multiple time scales; and S3, fusing the candidate set and the original data to form a space-time fusion path, performing semantic and time confidence correction on boundary uncertain nodes, and outputting a continuous fusion map. According to the method, the dynamic semantic evolution graph is constructed, the modal interpolation candidate set is generated, and the semantic belief propagation mechanism is introduced, so that structured completion of heterogeneous spatio-temporal data and semantic re-estimation of boundary uncertain nodes are realized, and the continuity and consistency of a multi-source fusion result are remarkably improved.

Description

A spatiotemporal multi-source data intelligent fusion system and method Technical Field

[0001] This invention relates to the field of big data processing technology, and in particular to a spatiotemporal multi-source data intelligent fusion system and method. Background Technology

[0002] With the rapid development of remote sensing observation, IoT monitoring, and multi-source heterogeneous databases, massive spatiotemporal data has been widely applied in urban management, environmental monitoring, traffic scheduling, and other fields. Different types of data sources, such as satellite imagery, video surveillance, meteorological sensors, and text records, typically exhibit significant differences in acquisition frequency, data modality, spatial precision, and semantic granularity. To enhance the collaborative analysis capabilities of cross-modal data, researchers have proposed a graph-based representation method, attempting to reveal the semantic change trends of spatial units along the time axis by constructing spatiotemporal semantic evolution paths, thereby assisting in subsequent intelligent decision-making and predictive modeling.

[0003] However, existing multi-source spatiotemporal fusion technologies have the following shortcomings: First, modal gaps and temporal discontinuities often exist between multi-source data, resulting in incomplete structures in the constructed maps during certain time periods, making it difficult to support continuous semantic evolution analysis; Second, for nodes in spatial boundary regions, existing methods lack effective uncertainty identification mechanisms, which easily lead to boundary ambiguity and label drift; Third, most current semantic fusion strategies lack dynamic confidence update mechanisms, making it impossible to dynamically correct the labels and temporal positions of uncertain nodes according to contextual changes, thereby reducing the consistency and interpretability of the fused map. Summary of the Invention

[0004] This invention provides an intelligent spatiotemporal multi-source data fusion system and method. By constructing a dynamic semantic evolution map, it identifies modal fault regions and generates a multimodal interpolation candidate set, fusing the original multi-source data to form a spatiotemporal fusion path. Furthermore, it introduces a spatial boundary overlap metric and a semantic confidence propagation mechanism to identify uncertain boundary nodes and perform confidence correction operations on labels and temporal locations, effectively improving the continuity, consistency, and semantic accuracy of the fused map in boundary regions, and enhancing the stability and intelligence of the multi-source data fusion system in complex dynamic environments.

[0005] A spatiotemporal multi-source data intelligent fusion method includes the following steps: S1, collecting multi-source data from multiple heterogeneous data sources, extracting their timestamps, spatial coordinates, and original semantic labels, and generating a dynamic semantic evolution map describing the semantic change path of the same spatial region at different time periods based on semantic fingerprint matching and historical state trajectory reasoning; S2, identifying modal fault segments in the current multi-source data based on the time span and modal missing status between nodes in the dynamic semantic evolution map, and constructing prediction windows at multiple time scales by combining the spatial-semantic correlation of adjacent nodes, and generating a multimodal interpolation candidate set for the fault area through structural compensation reasoning; S3, fusing the multimodal interpolation candidate set with the multi-source data into a spatiotemporal multi-source fusion path, identifying uncertain nodes located in the spatial boundary region in the spatiotemporal multi-source fusion path based on boundary overlap measurement and semantic confidence propagation mechanism, re-estimating the confidence of their semantic labels and time positions and correcting the boundary, and outputting a continuous fusion map with enhanced confidence consistency.

[0006] Optionally, S1 includes: S11, collecting multi-source data from different modal data sources, including remote sensing images, environmental sensors, traffic flow, social media event streams, and geographic tag data, and uniformly structuring all multi-source data into tuple form, represented as: ;in, For the first 1 data record For timestamps, For spatial coordinates, For data modalities, For native semantic labels; S12, discretize all multi-source data according to spatial range and time window to generate standardized spatial units and time slice grid indexes, denoted as ,in, For the first The spatial unit of the first An index unit consisting of time slices Encoding of spatial units, For time segments; S13, for each semantic tag set within Generate semantic fingerprint vector S14 is used to capture the comprehensive semantic state of the region-time period; S15 combines the same spatial unit. different time slices semantic fingerprint vector Construct a sequence of nodes and calculate the degree of semantic drift between adjacent nodes. S15, based on the degree of semantic drift Compared with the preset semantic drift threshold By comparing and identifying nodes with significant semantic changes, if And mark it as a semantic evolution trajectory node. Ultimately, a set of semantic drift indicators is generated to identify unstable regions. and dynamic semantic evolution graph ,in, , These are evolutionary edges in chronological order.

[0007] Optionally, S2 includes: S21, in the dynamic semantic evolution graph, traversing the semantic evolution trajectory nodes under each spatial unit, analyzing the time span and modal data integrity between adjacent nodes, if any modality is missing in the corresponding time slice, and the time interval between its adjacent nodes exceeds the set sampling frequency tolerance threshold, then the time period where the node is located is marked as a modal fault segment; S22, for the identified modal fault segments, constructing multiple sets of interpolation prediction windows at multiple time scales, using time series interpolation, spatial neighborhood regression and semantic similarity propagation algorithms to perform structural compensation reasoning on the missing modal information, generating a set of candidate modal values, each candidate modal value is simultaneously associated with its prediction source, and finally outputting a multimodal interpolation candidate set.

[0008] Optionally, S21 includes: S211, for each spatial unit in the dynamic semantic evolution graph Extract its time slices semantic evolution trajectory nodes And count each mode In the corresponding time slice Does the data contain valid data? Construct a modality coverage matrix. S212, for the same spatial unit Statistically analyze the time span of adjacent time slice nodes in its semantic evolution trajectory. S213, Set the sampling frequency tolerance threshold for each mode. If a mode exists satisfy and Then the corresponding time slice Or the preceding and following window segments are marked as modal fault segments. .

[0009] Optionally, S22 includes: S221, for the identified modal fault segments, using the semantic evolution trajectory nodes corresponding to the fault nodes. Centered on time, a set of timescales is constructed along the time dimension. Including short timescale windows and long-term timescale windows S222, within each short-term and long-term time-scale window, for the missing modality Candidate time series interpolation elements are generated from the temporal, spatial, and semantic dimensions, respectively. Candidates for Spatial Neighborhood Regression and semantic similarity propagation candidates S223: Candidate modal values ​​from different interpolation paths are organized into a candidate set, and a prediction source label is attached to each candidate value to form a multimodal interpolation candidate set. .

[0010] Optionally, S3 includes: S31, mapping the generated multimodal imputation candidate set and multi-source data together to a dynamic semantic evolution graph, and reconstructing the connection relationship between nodes within each spatial unit based on temporal order and spatial adjacency to form a spatiotemporal multi-source fusion path; S32, traversing all nodes in the spatiotemporal multi-source fusion path, and identifying uncertain nodes located in the boundary ambiguity zone or sparse data distribution area based on the spatial boundary overlap metric function and semantic uncertainty index; S33, for the identified uncertain nodes, fusing information from their temporal context, spatial neighbor nodes, and semantic fingerprint similar nodes, performing label re-estimation and temporal position fine-tuning based on the semantic confidence propagation mechanism, and finally generating a continuous fusion graph with enhanced confidence consistency.

[0011] Optionally, S31 includes: S311, dividing the multi-source data and the multimodal interpolation candidate set according to spatial units. and time slices Mapped to fusion nodes in the dynamic semantic evolution graph S312, in each spatial unit Within, indexed by time Connect adjacent nodes in sequence to form time evolution path edges. S313, in the same time slice Next, determine any spatial unit. Its adjacent unit Does the fusion node satisfy the spatial adjacency relationship, and generate a spatiotemporal multi-source fusion path? .

[0012] Optionally, S32 includes: S321, facilitating the spatiotemporal multi-source fusion path. Perform a traversal, extracting the fusion nodes in the path one by one. and at the same time The following constructs its spatial neighborhood node set. S322, for fusion nodes Calculate its spatial units degree of boundary overlap with neighboring spatial units ,like This indicates that the fusion node is located in a spatial boundary intersection or ambiguity region, where, S323, based on the degree of dispersion between the semantic fingerprint vector of the fused node and the semantic fingerprint of the neighboring nodes, calculate the semantic uncertainty index. Furthermore, by integrating spatial and semantic conditions, a rule for determining uncertain nodes is defined.

[0013] Optionally, S33 includes: S331, for an uncertain set of nodes. Construct a propagation neighborhood graph of the adjacency structure. Including time context nodes Spatial Neighborhood Nodes and semantically similar nodes S332, For uncertain nodes, their semantic labels are represented as the original label confidence vectors. Based on the propagation neighborhood graph Each uncertain node Tag information and dissemination weight Calculate the re-estimated label confidence vector S333, if the re-estimated label confidence vector Compared with the original label confidence vector The distance between them exceeds the offset threshold If the uncertain node label has drifted effectively, a correction operation is performed, and its temporal position is fine-tuned based on the re-estimated temporal consistency probability density of the label. All corrected uncertain nodes are then reconnected to the spatiotemporal multi-source fusion path to generate a continuous fusion map with enhanced confidence consistency.

[0014] A spatiotemporal multi-source data intelligent fusion system is provided to implement the aforementioned spatiotemporal multi-source data intelligent fusion method. The system comprises the following modules: a data acquisition module: acquiring multi-source data from multiple heterogeneous data sources and extracting their corresponding timestamps, spatial coordinates, and native semantic labels; a semantic evolution map construction module: generating a dynamic semantic evolution map describing the semantic change path of the same spatial region at different time periods based on semantic fingerprint matching and historical state trajectory reasoning; a modal fault identification and interpolation module: identifying modal fault segments in the multi-source data based on the time span and modal missingness between nodes in the dynamic semantic evolution map, constructing prediction windows at multiple time scales, and generating a multimodal interpolation candidate set for the fault area by combining the spatial-semantic correlation of adjacent nodes; and a confidence fusion and correction module: fusing the multimodal interpolation candidate set with the multi-source data into a spatiotemporal multi-source fusion path, identifying uncertain nodes located in the spatial boundary region of the fusion path based on spatial boundary overlap measurement and semantic confidence propagation mechanism, and re-estimating the confidence of their semantic labels and temporal positions, and correcting the boundary reinforcement, outputting a continuous fusion map with enhanced confidence consistency.

[0015] The beneficial effects of this invention are as follows: By constructing a dynamic semantic evolution map based on semantic fingerprint matching and historical trajectory reasoning, this invention can accurately identify the semantic evolution path of multi-source heterogeneous data in the spatiotemporal dimension without relying on a complete dataset. It effectively captures regional semantic drift and semantic instability, solves the problem that traditional multi-source fusion methods cannot perceive regional semantic dynamic changes, and improves the adaptability of the fusion model to spatiotemporal heterogeneous environments.

[0016] This invention introduces a modal tomography identification and structural compensation reasoning mechanism. In complex environments where multimodal data is missing or acquired asynchronously, it can construct an interpolation prediction window based on multiple time scales and integrate time series interpolation, spatial neighborhood regression, and semantic similarity propagation algorithms to generate a multimodal interpolation candidate set. This enables the structural recovery of multi-source data in regions with incomplete modalities, solving the problems of coarse interpolation granularity and poor structural consistency in existing methods.

[0017] This invention, by introducing a spatial boundary overlap metric and a semantic confidence propagation mechanism, can identify and correct uncertain nodes in the fusion path that are in regions with ambiguous boundaries or discrete semantics. Through semantic label confidence re-evaluation and temporal position fine-tuning, a continuous fusion map with enhanced confidence consistency is generated, thereby significantly improving the stability of the fusion result in the spatial boundary region and the consistency of temporal evolution, making up for the shortcomings of existing fusion methods in boundary semantic determination and dynamic correction. Attached Figure Description

[0018] 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 only for this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 is a schematic flowchart of the fusion method according to an embodiment of the present invention; Figure 2 is a schematic diagram of the system functional modules according to an embodiment of the present invention. Detailed Implementation

[0020] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. Those skilled in the art may employ other alternative methods to implement some well-known technologies; moreover, the accompanying drawings are only for more specific description of the embodiments and are not intended to specifically limit the present invention.

[0021] As shown in Figure 1, a spatiotemporal multi-source data intelligent fusion method includes the following steps: S1, collecting multi-source data from multiple heterogeneous data sources, extracting their timestamps, spatial coordinates, and original semantic labels, and generating a dynamic semantic evolution map describing the semantic change path of the same spatial region at different time periods based on semantic fingerprint matching and historical state trajectory reasoning; S2, identifying modal fault segments in the current multi-source data based on the time span and modal missing status between nodes in the dynamic semantic evolution map, and constructing prediction windows at multiple time scales by combining the spatial-semantic correlation of adjacent nodes, and generating a multimodal interpolation candidate set for the fault area through structural compensation reasoning; S3, fusing the multimodal interpolation candidate set with the multi-source data into a spatiotemporal multi-source fusion path, identifying uncertain nodes located in the spatial boundary region in the spatiotemporal multi-source fusion path based on boundary overlap measurement and semantic confidence propagation mechanism, re-estimating the confidence of their semantic labels and time positions and correcting the boundary, and outputting a continuous fusion map with enhanced confidence consistency.

[0022] S1 includes: S11, which collects multi-source data from different modal data sources, including remote sensing images, environmental sensors, traffic flow, social media event streams, and geographic tag data, and structures all multi-source data into tuple form, represented as follows: ;in, For the first 1 data record For timestamps, For spatial coordinates, For data modalities, such as images, text, sensors, etc. For native semantic labels, such as "water body", "traffic congestion", "high temperature", etc.; S12, discretize all multi-source data according to spatial range and time window, generating standardized spatial units and time slice grid indexes, denoted as ,in, For the first The spatial unit of the first An index unit consisting of time slices Encoding spatial units, such as grids, administrative divisions, and hotspot clusters. For time segments, such as hours, days, weeks; S13, for each semantic tag set within Generate semantic fingerprint vector This is used to capture the comprehensive semantic state of this region-time period, and is represented as: ;in, semantic tags The corresponding embedding vector, For aggregation functions; S14, grouping the same spatial units different time slices semantic fingerprint vector Construct a sequence of nodes and calculate the degree of semantic drift between adjacent nodes. , is represented as: S15, based on the degree of semantic drift Compared with the preset semantic drift threshold By comparing and identifying nodes with significant semantic changes, if And mark it as a semantic evolution trajectory node. Ultimately, a set of semantic drift indicators is generated to identify unstable regions. and dynamic semantic evolution graph ,in, , Evolutionary edges in temporal order; semantic drift threshold The setting uses historical sliding window statistics, and is represented as follows: ;in, In a certain spatial region Historical time window Mean semantic drift within, This represents the corresponding semantic drift standard deviation. It is a regulating factor.

[0023] S2 includes: S21, in the dynamic semantic evolution graph, traversing the semantic evolution trajectory nodes under each spatial unit, analyzing the time span and modal data integrity between adjacent nodes, if any modality is missing in the corresponding time slice, and the time interval between its adjacent nodes exceeds the set sampling frequency tolerance threshold, then the time period where the node is located is marked as a modal fault segment; S22, for the identified modal fault segments, constructing multiple sets of interpolation prediction windows at multiple time scales, using time series interpolation, spatial neighborhood regression and semantic similarity propagation algorithms to perform structural compensation reasoning on the missing modal information, generating a set of candidate modal values, each candidate modal value is also associated with its prediction source, and finally outputting a multimodal interpolation candidate set.

[0024] S21 includes: S211, for each spatial unit in the dynamic semantic evolution graph. Extract its time slices semantic evolution trajectory nodes And count each mode In the corresponding time slice Does the data contain valid data? Construct a modality coverage matrix. , is represented as: ;in, , S212 refers to a set of modalities in multi-source data, such as images, temperature, and text; S212 refers to the same spatial unit. Statistically analyze the time span of adjacent time slice nodes in its semantic evolution trajectory. , is represented as: ;in, S213 sets the sampling frequency tolerance threshold for each mode, using the timestamp of the previous node. If a mode exists satisfy and Then the corresponding time slice Or the preceding and following window segments are marked as modal fault segments. Sampling frequency tolerance threshold Represented as: ;in, For modality The average sampling interval within the historical time window, This represents the standard deviation of the corresponding time interval. This is the tolerance coefficient.

[0025] S22 includes: S221, for the identified modal fault segments, using the semantic evolution trajectory nodes corresponding to the fault nodes. Centered on time, a set of timescales is constructed along the time dimension. Including short timescale windows and long-term timescale windows , is represented as: ; ; ;in, S222 represents the time scale radius; within each short-time scale window and long-time scale window, for the missing mode... Candidate time series interpolation elements are generated from the temporal, spatial, and semantic dimensions, respectively. Candidates for Spatial Neighborhood Regression and semantic similarity propagation candidates , is represented as: ;in, For interpolation operators, such as linear interpolation, , These are the observations of this mode in adjacent time slices; ;in, spatial unit The neighborhood set, The modal values ​​of neighboring cells in the same time slice. Spatial weights, Geographical distance between spatial units; ;in, A set of semantically similar nodes , These are the corresponding semantic fingerprint vectors. S223 uses the cosine similarity function to organize candidate modal values ​​from different interpolation paths into a unified candidate set, and adds a prediction source label to each candidate value to form a multimodal interpolation candidate set. , is represented as: ;in, , , These represent candidate values ​​originating from temporal interpolation, spatial regression, or semantic propagation paths, respectively.

[0026] S3 includes: S31, mapping the generated multimodal imputation candidate set and multi-source data together to the dynamic semantic evolution graph, reconstructing the connection relationship between nodes within each spatial unit based on temporal order and spatial adjacency, forming a spatiotemporal multi-source fusion path; S32, traversing all nodes in the spatiotemporal multi-source fusion path, identifying uncertain nodes located in the boundary ambiguity zone or sparse data distribution area based on the spatial boundary overlap metric function and semantic uncertainty index; S33, for the identified uncertain nodes, fusing information from their temporal context, spatial neighbor nodes, and semantic fingerprint similar nodes, performing label re-estimation and temporal position fine-tuning based on the semantic confidence propagation mechanism, and finally generating a continuous fusion graph with enhanced confidence consistency.

[0027] S31 includes: S311, which combines multi-source data with multimodal interpolation candidate sets according to spatial units. and time slices Mapped to fusion nodes in the dynamic semantic evolution graph , is represented as: ;in, This is the set of modal features at this spatiotemporal node; ;in, These are the original modal values. For interpolation candidate values; S312, in each spatial unit Within, indexed by time Connect adjacent nodes in sequence to form time evolution path edges. , is represented as: S313, in the same time slice Next, determine any spatial unit. Its adjacent unit Does the fusion node satisfy the spatial adjacency relationship, and generate a spatiotemporal multi-source fusion path? , is represented as: ;in, Spatial edges represent connections between different spatial units within the same time slice. This is the adjacency judgment function, where 1 indicates adjacent. .

[0028] S32 includes: S321, the spatiotemporal multi-source fusion path. Perform a traversal, extracting the fusion nodes in the path one by one. and at the same time The following constructs its spatial neighborhood node set. Used to characterize the boundary properties of nodes in a spatial structure, represented as: S322, for fusion nodes Calculate its spatial units degree of boundary overlap with neighboring spatial units ,like This indicates that the fusion node is located in a spatial boundary intersection or ambiguity region, where, The spatial boundary overlap threshold is expressed as: ;in, spatial unit With neighboring spatial units The overlapping area, spatial unit area, S323, based on the degree of dispersion between the semantic fingerprint vector of the fused node and the semantic fingerprint of the neighboring nodes, calculate the semantic uncertainty index. Furthermore, considering both spatial and semantic conditions, a rule for determining uncertain nodes is defined, expressed as: ; ;in, , These are the semantic fingerprint vectors of the fused node and its neighboring nodes, respectively. For an uncertain set of nodes, This is the threshold for semantic uncertainty. , , respectively, are the mean semantic Euclidean distances in the full image or the sliding window. Standard deviation This is the adjustment coefficient.

[0029] S33 includes: S331, which is for an uncertain set of nodes. Construct a propagation neighborhood graph of the adjacency structure. Including time context nodes Spatial Neighborhood Nodes and semantically similar nodes , is represented as: ; ; ;in, This is the semantic fingerprint vector of the candidate node. S332, for uncertain nodes, their semantic label is represented as the original label confidence vector. Based on the propagation neighborhood graph Each uncertain node Tag information and dissemination weight Calculate the re-estimated label confidence vector , is represented as: ; ;in, This is the semantic label vector of the neighboring nodes. For time distance, For spatial distance, The cosine similarity function is used. , These are the propagation attenuation factors that regulate time and space, respectively. S333 is the normalization factor; if the re-estimated label confidence vector Compared with the original label confidence vector The distance between them exceeds the offset threshold If the value is set to 0.2, then the uncertain node label is considered to have experienced a valid drift. A correction operation is performed, and its temporal position is fine-tuned based on the re-estimated temporal consistency probability density of the label. All corrected uncertain nodes are then reconnected to the spatiotemporal multi-source fusion path, generating a continuous fusion map with enhanced confidence consistency, represented as: .

[0030] As shown in Figure 2, a spatiotemporal multi-source data intelligent fusion system is used to implement the aforementioned spatiotemporal multi-source data intelligent fusion method. It includes the following modules: a data acquisition module: acquiring multi-source data from multiple heterogeneous data sources and extracting their corresponding timestamps, spatial coordinates, and native semantic labels; a semantic evolution map construction module: generating a dynamic semantic evolution map describing the semantic change path of the same spatial region at different time periods based on semantic fingerprint matching and historical state trajectory reasoning; a modal fault identification and interpolation module: identifying modal fault segments in the multi-source data based on the time span and modal missingness between nodes in the dynamic semantic evolution map, constructing prediction windows at multiple time scales, and generating a multimodal interpolation candidate set for the fault area by combining the spatial-semantic correlation of adjacent nodes; and a confidence fusion and correction module: fusing the multimodal interpolation candidate set with the multi-source data into a spatiotemporal multi-source fusion path, identifying uncertain nodes located in the spatial boundary region of the fusion path based on spatial boundary overlap measurement and semantic confidence propagation mechanism, and re-estimating the confidence of their semantic labels and temporal positions, and correcting the boundary reinforcement, outputting a continuous fusion map with enhanced confidence consistency.

[0031] This invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of this invention. To provide the public with a thorough understanding of this invention, specific details are described in detail in the following preferred embodiments; however, those skilled in the art will fully understand the invention even without these details. Furthermore, to avoid unnecessary misunderstanding of the essence of this invention, well-known methods, processes, procedures, components, and circuits are not described in detail.

[0032] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for intelligent fusion of spatiotemporal multi-source data, characterized in that, Includes the following steps: S1. Collect multi-source data from multiple heterogeneous data sources, extract their timestamps, spatial coordinates, and original semantic labels, and generate a dynamic semantic evolution map describing the semantic change path of the same spatial region at different time periods based on semantic fingerprint matching and historical state trajectory reasoning; S2. Based on the time span and modal missing status between nodes in the dynamic semantic evolution map, identify the modal fault segments existing in the current multi-source data, and construct prediction windows at multiple time scales by combining the spatial-semantic correlation of adjacent nodes, and generate a multimodal imputation candidate set for the fault area through structural compensation reasoning; S3. Fuse the multimodal imputation candidate set with the multi-source data into a spatiotemporal multi-source fusion path, and identify uncertain nodes located in the spatial boundary region in the spatiotemporal multi-source fusion path based on boundary overlap measurement and semantic confidence propagation mechanism, re-evaluate the confidence of their semantic labels and time positions and perform boundary reinforcement correction, and output a continuous fusion map with enhanced confidence consistency.

2. The spatiotemporal multi-source data intelligent fusion method according to claim 1, characterized in that, S1 includes: S11, collecting multi-source data from different modal data sources, including remote sensing images, environmental sensors, traffic flow, social media event streams, and geographic tag data, and uniformly structuring all multi-source data into tuple form, represented as: ;in, For the first 1 data record For timestamps, For spatial coordinates, For data modalities, For native semantic labels; S12, discretize all multi-source data according to spatial range and time window to generate standardized spatial units and time slice grid indexes, denoted as ,in, For the first The spatial unit of the first An index unit consisting of time slices Encoding of spatial units, For time segments; S13, for each semantic tag set within Generate semantic fingerprint vector S14 is used to capture the comprehensive semantic state of the region-time period; S15 combines the same spatial unit. Different time slices semantic fingerprint vector Construct a sequence of nodes and calculate the degree of semantic drift between adjacent nodes. S15, based on the degree of semantic drift Compared with the preset semantic drift threshold By comparing and identifying nodes with significant semantic changes, if And mark it as a semantic evolution trajectory node. Ultimately, a set of semantic drift indicators is generated to identify unstable regions. and dynamic semantic evolution graph ,in, ， These are evolutionary edges in chronological order.

3. The spatiotemporal multi-source data intelligent fusion method according to claim 2, characterized in that, S2 includes: S21, in the dynamic semantic evolution graph, traversing the semantic evolution trajectory nodes under each spatial unit, analyzing the time span and modal data integrity between adjacent nodes, if any modality is missing in the corresponding time slice, and the time interval between its adjacent nodes exceeds the set sampling frequency tolerance threshold, then the time period where the node is located is marked as a modal fault segment; S22, for the identified modal fault segments, constructing multiple sets of interpolation prediction windows at multiple time scales, using time series interpolation, spatial neighborhood regression and semantic similarity propagation algorithms to perform structural compensation reasoning on the missing modal information, generating a set of candidate modal values, each candidate modal value is also associated with its prediction source, and finally outputting a multimodal interpolation candidate set.

4. The spatiotemporal multi-source data intelligent fusion method according to claim 3, characterized in that, S21 includes: S211, for each spatial unit in the dynamic semantic evolution graph Extract its time slices semantic evolution trajectory nodes And count each mode In the corresponding time slice Does the data contain valid data? Construct a modality coverage matrix. S212, for the same spatial unit Statistically analyze the time span of adjacent time slice nodes in its semantic evolution trajectory. S213, Set the sampling frequency tolerance threshold for each mode. If a mode exists satisfy and Then the corresponding time slice Or the preceding and following window segments are marked as modal fault segments. 。 5. The spatiotemporal multi-source data intelligent fusion method according to claim 4, characterized in that, S22 includes: S221, for the identified modal fault segments, using the semantic evolution trajectory nodes corresponding to the fault nodes. Centered on time, a set of timescales is constructed along the time dimension. Including short timescale windows and long-term timescale windows S222, within each short-term and long-term time-scale window, for the missing modality Candidate time series interpolation elements are generated from the temporal, spatial, and semantic dimensions, respectively. Candidates for Spatial Neighborhood Regression and semantic similarity propagation candidates S223: Candidate modal values ​​from different interpolation paths are organized into a candidate set, and a prediction source label is attached to each candidate value to form a multimodal interpolation candidate set. 。 6. The spatiotemporal multi-source data intelligent fusion method according to claim 5, characterized in that, S3 includes: S31, mapping the generated multimodal imputation candidate set and multi-source data together to the dynamic semantic evolution graph, and reconstructing the connection relationship between nodes within each spatial unit based on temporal order and spatial adjacency to form a spatiotemporal multi-source fusion path; S32, traversing all nodes in the spatiotemporal multi-source fusion path, and identifying uncertain nodes located in the boundary ambiguity zone or the sparse data distribution area based on the spatial boundary overlap metric function and semantic uncertainty index; S33, for the identified uncertain nodes, fusing the information of their temporal context, spatial neighbor nodes and semantic fingerprint similar nodes, performing label re-estimation and temporal position fine-tuning based on the semantic confidence propagation mechanism, and finally generating a continuous fusion graph with enhanced confidence consistency.

7. The spatiotemporal multi-source data intelligent fusion method according to claim 6, characterized in that, S31 includes: S311, dividing the multi-source data and the multimodal interpolation candidate set according to spatial units. and time slices Mapped to fusion nodes in the dynamic semantic evolution graph S312, in each spatial unit Within, indexed by time Connect adjacent nodes in sequence to form time evolution path edges. S313, in the same time slice Next, determine any spatial unit. Its adjacent unit Does the fusion node satisfy the spatial adjacency relationship, and generate a spatiotemporal multi-source fusion path? 。 8. The spatiotemporal multi-source data intelligent fusion method according to claim 7, characterized in that, S32 includes: S321, facilitating the spatiotemporal multi-source fusion path. Perform a traversal, extracting the fusion nodes in the path one by one. and at the same time The following constructs its spatial neighborhood node set. S322, for fusion nodes Calculate its spatial units degree of boundary overlap with neighboring spatial units ,like This indicates that the fusion node is located in a spatial boundary intersection or ambiguity region, where, S323, based on the degree of dispersion between the semantic fingerprint vector of the fused node and the semantic fingerprint of the neighboring nodes, calculate the semantic uncertainty index. Furthermore, by integrating spatial and semantic conditions, a rule for determining uncertain nodes is defined.

9. The spatiotemporal multi-source data intelligent fusion method according to claim 8, characterized in that, S33 includes: S331, for the uncertain node set Construct a propagation neighborhood graph of the adjacency structure. Including time context nodes Spatial Neighborhood Nodes and semantically similar nodes S332, For uncertain nodes, their semantic labels are represented as the original label confidence vectors. Based on the propagation neighborhood graph Each uncertain node Tag information and dissemination weight Calculate the re-estimated label confidence vector S333, if the re-estimated label confidence vector Compared with the original label confidence vector The distance between them exceeds the offset threshold If the uncertain node label has drifted effectively, a correction operation is performed, and its temporal position is fine-tuned based on the re-estimated temporal consistency probability density of the label. All corrected uncertain nodes are then reconnected to the spatiotemporal multi-source fusion path to generate a continuous fusion map with enhanced confidence consistency.

10. A spatiotemporal multi-source data intelligent fusion system, used to implement the spatiotemporal multi-source data intelligent fusion method as described in any one of claims 1-9, characterized in that, It includes the following modules: Data acquisition module: collects multi-source data from multiple heterogeneous data sources and extracts their corresponding timestamps, spatial coordinates and native semantic tags; Semantic Evolution Graph Construction Module: Based on semantic fingerprint matching and historical state trajectory reasoning, a dynamic semantic evolution graph describing the semantic change path of the same spatial region at different time periods is generated; Modal Fault Recognition and Imputation Module: Based on the time span and modal missing status between nodes in the dynamic semantic evolution graph, modal fault segments in multi-source data are identified, and prediction windows are constructed at multiple time scales. Combined with the spatial-semantic correlation of adjacent nodes, a multimodal imputation candidate set for the fault area is generated; Confidence fusion and correction module: The multimodal interpolation candidate set and multi-source data are fused into a spatiotemporal multi-source fusion path. Based on the spatial boundary overlap measurement and semantic confidence propagation mechanism, the uncertain nodes located in the spatial boundary region in the fusion path are identified, and their semantic labels and temporal positions are re-estimated with confidence and boundary strengthening correction, and a continuous fusion map with enhanced confidence consistency is output.

Images