Intelligent surveying and mapping system and method based on remote sensing technology

Abstract

The invention relates to the technical field of remote sensing surveying and mapping, and particularly discloses an intelligent surveying and mapping system and method based on a remote sensing technology, and the method comprises the steps: obtaining a multi-time-sequence remote sensing image data set of a target region through multiple platforms carrying intelligent sensing elements; performing three-dimensional reconstruction and cross-time-phase accurate alignment on the time-phase data to generate a three-dimensional space coordinate sequence under a unified coordinate system; carrying out conjoint analysis on the multi-time-sequence attribute of each spatial point in the sequence, and generating a fused spatio-temporal feature by distinguishing a persistent structure and instantaneous interference; constructing a three-dimensional graph structure according to the features, performing hierarchical aggregation, and identifying a point set with similar features and continuous space as an entity unit with a single semantic identifier; based on the semantic and geometric attributes of the entity units, automatically establishing a spatial topology and time evolution relationship, and generating a structured information set containing complete space-time dimension information; according to the invention, high-precision intelligent surveying and mapping from static modeling to dynamic spatio-temporal knowledge generation is realized.

Description

Technical Field

[0001] This invention relates to the field of remote sensing mapping technology, and more specifically to an intelligent mapping system and method based on remote sensing technology. Background Technology

[0002] With the rapid development of remote sensing technologies such as satellites, drones, and near-ground mobile platforms, the sensing capabilities of the intelligent sensing elements mounted on them (such as high-resolution multispectral imagers and lidar) are constantly improving. This enables us to acquire multi-source, multi-temporal remote sensing image datasets of the Earth's surface with unprecedented frequency and precision. These intelligent sensing elements can not only automatically record high-precision spatiotemporal information but also perform on-orbit or real-time preprocessing, providing a rich data foundation for dynamic and refined geographic information monitoring.

[0003] The existing technology has the following shortcomings:

[0004] When using oblique photogrammetry remote sensing technology to perform 3D mapping of targets with complex surface features (such as exquisite carvings on historical buildings), the point cloud data at a single moment is essentially a surface visual sample. In detailed areas, there are blurred boundaries, adhesion, and voids. This causes traditional geometric feature-based analysis methods to be unable to effectively distinguish between "decorative details" and "main structure," resulting in systematic errors in automated 3D semantic segmentation, loss of key details and true topological relationships, and inability to meet the requirements of high-precision mapping. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent mapping system and method based on remote sensing technology to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A smart mapping method based on remote sensing technology includes the following steps:

[0008] S1: Acquire a set of multi-time series remote sensing image data of the target area at multiple different time points;

[0009] S2: Based on a multi-time series remote sensing image data set, a three-dimensional spatial coordinate sequence under a unified spatial coordinate system is generated. The three-dimensional spatial coordinate sequence contains the geometric and optical properties of each spatial point at multiple time points.

[0010] S3: Jointly analyze the multiple temporal attributes of each spatial point in the three-dimensional spatial coordinate sequence. Process the data by distinguishing between persistent structures and transient disturbances in the temporal analysis method to generate the fused spatiotemporal features of each spatial point. The persistent structures are the entity features that exist continuously at different time points, and the transient disturbances are the non-entity features caused by lighting, occlusion or temporary objects.

[0011] S4: Based on the spatiotemporal characteristics of all spatial points, perform three-dimensional semantic analysis on the target area of ​​the survey, and aggregate points with similar spatiotemporal characteristics and spatial continuity into entity units with a single semantic identifier.

[0012] S5: Based on the semantic identifiers and geometric attributes of all entity units, generate a structured information set containing time dimension information. The structured information set marks each entity unit with at least one of the following time information: appearance time, duration of existence, and disappearance time, and establishes the spatial topology and temporal evolution relationship between each entity unit.

[0013] As a further aspect of the present invention: S2 specifically includes:

[0014] Three-dimensional reconstruction is performed on the image data at each time point in the multi-time series remote sensing image data set to generate an initial three-dimensional point cloud for each independent time point. Each initial three-dimensional point cloud contains the spatial coordinates and optical attributes of the corresponding time point.

[0015] The initial 3D point clouds at each independent time point are associated and precisely aligned point by point. By analyzing the local surface curvature and texture distribution stability of spatial points with the same name at adjacent time points, a multi-temporal dense matching point set with a unified spatial reference is output.

[0016] Based on a multi-temporal dense matching point set, the spatial coordinates and optical properties of each successfully matched spatial point are integrated in chronological order at different time points to generate a three-dimensional spatial coordinate sequence.

[0017] As a further aspect of the present invention: the output multi-temporal dense matching point set with a unified spatial reference specifically includes:

[0018] Multidimensional stable features are extracted from each spatial point of each initial 3D point cloud. These multidimensional stable features include high-order surface curvature features and local texture spectrum features that maintain statistical consistency between adjacent time points.

[0019] A global joint optimization framework is constructed with feature consistency as a constraint and spatial transformation parameters as the optimization objective. The global joint optimization framework simultaneously minimizes the feature differences and geometric position residuals of matching points between all adjacent time point pairs.

[0020] Solve the global joint optimization framework and simultaneously calculate a set of spatial transformation parameters that enable all time point clouds to achieve the best alignment state under a unified spatial reference.

[0021] Spatial transformation parameters are applied to perform coordinate transformation and resampling on all initial 3D point clouds to generate multi-temporal dense matching point sets.

[0022] As a further aspect of the present invention: S3 specifically includes:

[0023] For each spatial point in a three-dimensional spatial coordinate sequence, the variation pattern of the optical properties of each spatial point at multiple different time points is analyzed, and the stability and periodicity of the property values ​​of each spatial point at different time points are calculated.

[0024] Based on the stability and periodicity of attribute values, the optical attributes of spatial points at different time points are subjected to temporal adaptive weighted fusion to generate principal component features representing the persistent optical attributes of the corresponding spatial points.

[0025] By comparing the deviations between the measured optical properties of spatial points at different time points and the principal component features, the anomalous attribute components characterizing transient disturbances are extracted and quantified.

[0026] The principal component features, anomalous attribute components, and temporal statistics of geometric attributes of combined spatial points are used to form the fused spatiotemporal features of the corresponding spatial points.

[0027] As a further aspect of the present invention: the extraction and quantification of the abnormal attribute components characterizing transient disturbances specifically includes:

[0028] Calculate the attribute deviation between the measured optical attribute values ​​of a spatial point at each time point and the principal component eigenvalues ​​at the corresponding time point;

[0029] Sequence analysis was performed on the attribute deviation values ​​at all time points to identify and separate isolated deviation patterns from periodic deviation patterns with regular fluctuation characteristics.

[0030] The intensity and duration of isolated bias patterns and periodic bias patterns are quantified respectively to generate transient anomalous components that characterize sudden light events and cyclic anomalous components that characterize seasonal vegetation shading.

[0031] The transient anomaly component and the cyclic anomaly component are merged and encoded to form the anomaly attribute component of the corresponding spatial point.

[0032] As a further aspect of the present invention: S4 specifically includes:

[0033] Based on the spatiotemporal characteristics of all spatial points, a three-dimensional graph structure that integrates feature similarity and spatial adjacency is constructed.

[0034] Perform a hierarchical node aggregation operation based on a 3D graph structure. In each level, the hierarchical node aggregation operation first merges nodes with feature similarity higher than a first set threshold and spatially directly adjacent nodes to generate a primary aggregation unit.

[0035] In subsequent levels, the generated primary aggregation units are iteratively merged based on the spatial proximity of the boundary points of the primary aggregation units and the statistical consistency of the spatiotemporal features of the overall fusion of the units, to form higher-level semantic aggregation units.

[0036] When a semantic aggregation unit meets the preset conditions of geometric integrity and feature homogeneity, the aggregation process stops, and the semantic aggregation unit is output as an entity unit with a single semantic identifier.

[0037] As a further aspect of the present invention: the construction of a three-dimensional graph structure that integrates feature similarity and spatial adjacency specifically includes:

[0038] Based on the three-dimensional coordinates of spatial points, multi-scale spatial unit division is carried out to generate three-dimensional voxel units with hierarchical relationships, and each spatial point is assigned to the corresponding smallest scale voxel unit.

[0039] Calculate the spatiotemporal feature similarity of all spatial point pairs within any two directly adjacent voxel units in three-dimensional space, and select the statistical median of the similarity as the initial feature connection weight between the two voxel units.

[0040] By verifying and correcting the spatial proximity relationships of voxel units, false spatial adjacency connections caused by uneven point cloud density or noise are removed.

[0041] By using voxel units that satisfy spatial continuity constraints as graph nodes and modifying the initial feature connection weights as edge weights between corresponding nodes, the 3D graph structure is constructed.

[0042] As a further aspect of the present invention: S5 specifically includes:

[0043] By tracing back the timestamps of all spatial points corresponding to each entity unit in the three-dimensional spatial coordinate sequence, the initial appearance time and final disappearance time of the entity unit can be determined.

[0044] Based on the geometric boundaries of all entity units in a unified spatial coordinate system, parallel determination of three-dimensional spatial relationships is performed to establish spatial topological relationships of inclusion, adjacency, and separation between entity units.

[0045] Based on the appearance and disappearance time of entity units and the spatial topological relationship of entity units, the temporal evolution relationship between entity units is derived and recorded by analyzing the state changes of the topological relationship in different time windows.

[0046] The semantic identifiers, geometric attributes, temporal information, spatial topological relationships, and temporal evolution relationships of each entity unit are integrated to generate a structured information set.

[0047] As a further aspect of the present invention: the derivation and recording of the temporal evolution relationship between entity units specifically includes:

[0048] Based on the appearance and disappearance time of entity units, the entire observation period is divided into continuous time windows, and within each time window, a relational state code is generated for each pair of entity units with topological relationships according to spatial topological relationships.

[0049] Analyze the changes in the entity unit relationship state encoding in the continuous time window sequence, identify the critical time window when the relationship state changes, and record the specific type of state change;

[0050] Based on the preset entity evolution rule base, the state transition type is mapped to a specific time evolution relationship type, which includes derivation, fusion, splitting and replacement;

[0051] For each pair of entity units that undergo a state transition, record the type of time evolution relationship corresponding to the entity unit, the critical time window, and the relationship state encoding before and after the transition, thus completing the derivation and recording of the time evolution relationship.

[0052] An intelligent mapping system based on remote sensing technology includes:

[0053] The multi-temporal remote sensing data acquisition module is used to acquire a set of multi-temporal remote sensing image data of the target area at multiple different time points;

[0054] The spatiotemporal data alignment and serialization module generates a three-dimensional spatial coordinate sequence in a unified spatial coordinate system based on a multi-time series of remote sensing image data. The three-dimensional spatial coordinate sequence contains the geometric and optical properties of each spatial point at multiple time points.

[0055] The fusion spatiotemporal feature generation module is used to jointly analyze the multiple temporal attributes of each spatial point in the three-dimensional spatial coordinate sequence. It is processed by a temporal analysis method that distinguishes between persistent structures and transient disturbances to generate fusion spatiotemporal features for each spatial point. Persistent structures are entity features that exist continuously at different time points, while transient disturbances are non-entity features caused by lighting, occlusion, or temporary objects.

[0056] The 3D semantic entity aggregation module performs 3D semantic analysis on the survey target area based on the spatiotemporal characteristics of all spatial points, and aggregates points with similar spatiotemporal characteristics and spatial continuity into entity units with a single semantic identifier.

[0057] The spatiotemporal knowledge structuring generation module generates a structured information set containing time dimension information based on the semantic identifiers and geometric attributes of all entity units. The structured information set marks each entity unit with at least one of the following time information: appearance time, duration of existence, and disappearance time, and establishes the spatial topology and temporal evolution relationship between each entity unit.

[0058] The beneficial effects of this invention are:

[0059] (1) This invention introduces joint analysis and spatiotemporal feature extraction of multi-temporal remote sensing data, which can automatically identify and filter out transient interference caused by changes in illumination, temporary occlusion, etc., while strengthening the expression of persistent entity features through temporal weighted fusion. This enables the subsequent semantic parsing process to achieve accurate component-level identification and aggregation in complex structural areas (such as facades with exquisite carvings) based on purer and more discriminative features. The generated entity units have clear boundaries and clear semantics, which improves the quality of surveying and mapping results in fields such as cultural heritage digitization and urban fine modeling.

[0060] (2) By constructing a time dimension that runs through the entire data processing process, this invention not only records the appearance and disappearance times of entity units, but also automatically derives and establishes evolutionary relationships (such as derivation, fusion, splitting, and replacement) between entities by analyzing the topological relationship changes in multi-temporal data. The final generated structured information set is a spatiotemporal knowledge base that integrates semantics, geometry, time, and correlation. This enables the method to not only describe "what the target is and where it is", but also to answer "how the target changes and how it interacts with its surroundings in time and space", providing an unprecedented dynamic data foundation and analytical capabilities for advanced applications such as smart city asset management, dynamic monitoring of illegal buildings, disaster damage assessment, and process backtracking. Attached Figure Description

[0061] The invention will now be further described with reference to the accompanying drawings.

[0062] Figure 1 This is a flowchart of the method of the present invention;

[0063] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0064] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0065] Please see Figure 1 As shown, this invention is an intelligent mapping method based on remote sensing technology, comprising the following steps:

[0066] S1: Acquire a set of multi-time series remote sensing image data of the target area at multiple different time points;

[0067] S2: Based on a multi-time series remote sensing image data set, a three-dimensional spatial coordinate sequence under a unified spatial coordinate system is generated. The three-dimensional spatial coordinate sequence contains the geometric and optical properties of each spatial point at multiple time points.

[0068] S3: Jointly analyze the multiple temporal attributes of each spatial point in the three-dimensional spatial coordinate sequence. Process the data by distinguishing between persistent structures and transient disturbances in the temporal analysis method to generate the fused spatiotemporal features of each spatial point. The persistent structures are the entity features that exist continuously at different time points, and the transient disturbances are the non-entity features caused by lighting, occlusion or temporary objects.

[0069] S4: Based on the spatiotemporal characteristics of all spatial points, perform three-dimensional semantic analysis on the target area of ​​the survey, and aggregate points with similar spatiotemporal characteristics and spatial continuity into entity units with a single semantic identifier.

[0070] S5: Based on the semantic identifiers and geometric attributes of all entity units, generate a structured information set containing time dimension information. The structured information set marks each entity unit with at least one of the following time information: appearance time, duration of existence, and disappearance time, and establishes the spatial topology and temporal evolution relationship between each entity unit.

[0071] In S1, data acquisition relies on a multi-platform collaborative observation network equipped with intelligent sensing elements. This network includes at least satellite platforms, aircraft platforms, and near-ground mobile observation platforms. Satellite platforms provide large-scale, periodic coverage data, aircraft platforms (such as UAVs) provide high-resolution, highly mobile regional data, while near-ground platforms are used for supplementary, detailed observations of specific areas. The intelligent sensing elements on each platform possess high-precision spatiotemporal information recording and data preprocessing capabilities; their core components are multispectral imaging sensors and lidar sensors.

[0072] The acquired multi-time-series remote sensing image data set specifically includes multi-source data directly generated by the intelligent sensing element or after preliminary processing. This data mainly includes: high spatial resolution multispectral digital images, lidar point cloud data, and corresponding sensor interior / exterior orientation elements and imaging timestamps. For each time point, data acquisition must ensure complete coverage of the target area, and the time interval between adjacent time points is set according to the mapping objective; for example, for dynamic monitoring, the interval may be several days or weeks, thus forming a complete time-series data set.

[0073] During data acquisition, the raw observation data undergoes real-time quality checks and standardized preprocessing via the embedded processing unit of the intelligent sensing element or its connected edge computing nodes. This includes radiometric correction of image data to eliminate atmospheric and illumination effects, and noise filtering and coarse registration of point cloud data. This ensures that data acquired at different time points possess the basic geometric and radiometric consistency required for subsequent processes, thereby forming a standardized multi-temporal remote sensing image dataset that can be directly used for subsequent processing.

[0074] In S2, 3D reconstruction is performed on the image data at each independent time point in the multi-temporal remote sensing image dataset. For each time point, the structure-of-motion reconstructing method is used to calculate the 3D coordinates of pixels based on the multi-view digital image and its corresponding sensor parameters. Specifically, high-precision exterior orientation elements of each image are first calculated using feature point matching and bundle adjustment. Then, a dense disparity map is generated using a multi-view stereo matching algorithm. Finally, the 3D spatial coordinates of each image point are calculated using forward intersection, and optical attribute values ​​derived from the original image are assigned to them, thus forming the initial 3D point cloud for that time point. Each initial 3D point cloud is a set consisting of a large number of spatial points and their corresponding optical attributes, representing the instantaneous 3D state of the Earth's surface at that moment.

[0075] The initial 3D point clouds generated at each time point are precisely aligned across time phases to obtain a multi-temporal dense matching point set under a unified spatial reference. The first step in this process is feature extraction: for each spatial point in each initial 3D point cloud, its multi-dimensional stable features are calculated. These features include higher-order surface curvature features and local texture spectrum features. The higher-order surface curvature features are calculated as follows: with the point as the center, 15 neighboring points are selected, a local quadratic surface is fitted, the two principal curvatures of the surface are calculated, and its Gaussian curvature and root mean square curvature are further calculated as descriptors. The local texture spectrum features are calculated as follows: the point is projected back onto all visible original images, a 16-pixel multiplied image patch centered on the projection point is extracted, the gray-level co-occurrence matrix of the image patch in different color channels is calculated, and three statistics—contrast, correlation, and entropy—are extracted from the matrix to form a feature vector. These features need to exhibit statistical consistency among potential matching point pairs at adjacent time points, that is, the Euclidean distance of their feature vectors is less than a preset consistency threshold, which is determined by statistically analyzing the distribution of feature distances of all candidate point pairs and taking the top 15% quantile.

[0076] The second step is to construct and solve the global joint optimization framework. This framework uses the spatial transformation parameters between all time-series point clouds as the variables to be optimized. The optimization objective function is defined as the sum of two terms: the first term is the sum of the squares of the feature vector differences between all adjacent time-series point pairs that have been successfully matched by the aforementioned multi-dimensional stable features; the second term is the sum of the squares of the three-dimensional coordinate residuals of these matched point pairs after the spatial transformation in the current iteration. These two terms are assigned different weights in the objective function. The weight of the first term is set to 1, and the weight of the second term is determined experimentally to be 0.5, in order to balance the contributions of feature consistency and geometric alignment. Through iterative optimization algorithms, such as the Levenberg-Marquardt method, the spatial transformation parameters of each time-series point cloud are continuously adjusted until the objective function value converges to the minimum value, at which point a set of optimal global spatial transformation parameters is obtained.

[0077] Finally, the optimal spatial transformation parameters obtained from the solution are applied to transform the coordinates of all initial 3D point clouds, converting them all to the same absolute spatial coordinate system. Since the point clouds may not be strictly aligned on the same spatial grid after transformation, resampling is necessary. Specifically, a uniform 3D regular grid with a side length of 0.1 meters is defined in the target coordinate system. For each voxel in the grid, the spatial points that fall into or are closest to that voxel in all time-series point clouds are found. The arithmetic mean of the spatial coordinates of these points is taken as the spatial coordinates of the voxel's center at that time point, and the average of their optical properties is taken as the optical properties of the voxel's center at that time point. If a voxel has no points falling into it at a certain time point, the attribute at that time point is marked as missing. This process generates the aforementioned multi-temporal densely matched point set.

[0078] Based on this multi-temporal densely matched point set, the spatial coordinates and optical properties of each successfully matched spatial point are organized in chronological order, forming a data structure indexed by point location and ordered by time, namely the aforementioned three-dimensional spatial coordinate sequence. This sequence fully records the continuous observations of the evolution of the surface geometry and surface optical properties of each spatial location over time.

[0079] In S3, firstly, for each spatial point in the three-dimensional spatial coordinate sequence, the variation pattern of its optical properties at multiple different time points is analyzed. Here, optical properties refer to those extracted from remote sensing images, such as near-infrared reflectance values. Assume a spatial point... In total The optical property values ​​observed at various points in time constitute a time series. Calculate the stability of the attribute values ​​at this spatial point. With periodicity of change Stability of attribute values The overall fluctuation of the optical properties at a given point is quantified by the standard deviation of all values ​​in the time series. Periodicity of change. To identify whether an attribute exhibits regular fluctuations with seasonal factors, the calculation is performed using periodogram analysis: A discrete Fourier transform is performed on the time series, and peaks with significantly higher energy than background noise are identified in the frequency domain. The period corresponding to these peaks (in units of time points) is the candidate period. The period is verified by calculating the correlation coefficient between the original sequence and its lagged sequence of the candidate period. If the correlation coefficient is greater than 0.7, the periodicity is confirmed, and the length of this period is recorded. (Unit: number of time points) is used as a periodicity parameter. If no significant periodicity is found, it is marked as aperiodic.

[0080] Secondly, based on the calculated stability of the attribute values With periodicity of change The optical attribute values ​​of the spatial point at different time points are then subjected to temporal adaptive weighted fusion to generate principal component features representing its persistent optical properties. The core of this process is to construct a weighting function that assigns a weight to each observation in the time series. Weight It consists of two parts: one part is based on the average difference of attribute values ​​between the observation time and all other times in the sequence (i.e., local stability), with smaller differences carrying higher weights; the other part is based on the periodicity analysis results, assigning higher weights to times with the same phase if periodicity exists, to strengthen the periodic stability signal. Specifically, the principal component eigenvalues... Calculated using the following weighted average formula: ;

[0081] in, Representing a spatial point In the Measured optical property values ​​at each time point. For stability-based weight components, the calculation formula is as follows: , and These are sequences The mean and standard deviation. For weighted components based on periodic phase, if a point is determined to be aperiodic, then all... If there exists a period of length of ,but: , Expressing the request Divide by The remainder after calculation makes the weights at the same phase point close to 1, and the weights at different phase points close to 0. This formula is used to calculate... It is a characteristic value that filters out random fluctuations and some periodic interferences, and better reflects the optical properties of the substrate at that point.

[0082] Subsequently, by comparing the measured optical property values ​​of spatial points at different time points... The principal component eigenvalues ​​calculated above The deviation between these deviations is used to extract and quantify the anomalous attribute components that characterize transient disturbances. The first step is to calculate the attribute deviation value at each time point. The calculation expression is: The second step is to process the attribute deviation value sequence for all time points. Sequence analysis is performed to identify and separate different bias patterns. This step is achieved by setting two thresholds: an amplitude threshold. and duration threshold Amplitude threshold Set as sequence 1.5 times the standard deviation of the absolute value. Scan the sequence to identify points within consecutive time points. The segment. If the segment duration is less than (For example, If the deviation is set to 5% of the total number of time points, then the segment is marked as an "isolated deviation pattern," corresponding to transient events such as cloud shadows and vehicles. Simultaneously, periodicity analysis is performed on the deviation sequence (using the same method as before). If significant periodicity is identified (correlation coefficient > 0.6), it is marked as a "periodic deviation pattern," corresponding to seasonal vegetation growth and withering. The third step involves quantifying these two patterns separately. For an isolated deviation pattern segment, its "intensity" is quantified as the sum of all... The average absolute value; "duration" is the number of time points contained in the segment. For a periodic bias pattern, its "intensity" is quantified within a complete period. The average amplitude (obtained by fitting a sine curve), and the "duration" record the length of its period. The fourth step is to merge and encode the quantization results. The encoding is a quadruple (Type, I, Dur, Phase). Here, Type represents the type (0 for isolated type, 1 for periodic type); I represents the intensity scalarization value (e.g., normalized to the 0-1 interval); Dur represents the duration or period length; and Phase is only valid for periodic types, representing the initial phase. The combination of all the anomalous pattern codes for a spatial point constitutes its "anomaly attribute component".

[0083] Finally, combine the principal component features of this spatial point. The final fused spatiotemporal features of a spatial point are formed by the temporal statistics of its anomalous attribute components and geometric attributes (such as elevation). The temporal statistics of the geometric attributes must include at least the maximum, minimum, and average elevation values ​​and the slope of the linear trend at all time points. The fused spatiotemporal features can be represented as a structured feature vector, for example: [ , , [Geometric maximum, geometric minimum, geometric mean, geometric trend slope, outlier component code 1, outlier component code 2]. This feature vector comprehensively expresses the stable essential attributes, changing patterns and rules, and sudden disturbances of the spatial point during the observation period, providing highly discriminative spatiotemporal dimensional information for subsequent semantic parsing.

[0084] In S4, firstly, based on the 3D coordinates of all spatial points and their fused spatiotemporal features, a 3D graph structure integrating feature similarity and spatial adjacency is constructed. The first step in this process is to perform multi-scale spatial unit partitioning. Based on the 3D coordinate range of all spatial points, a minimum voxel grid with a side length of 0.05 meters is defined, and each spatial point is assigned to its corresponding minimum voxel unit. Subsequently, every 8 adjacent minimum voxel units are merged into a larger voxel unit of the next higher level, and this process is recursively repeated to generate a voxel unit hierarchy with a 3-layer pyramid structure. Each non-empty minimum voxel unit will serve as a candidate node in the subsequent graph structure, and its represented feature vector is the arithmetic mean of the fused spatiotemporal features of all spatial points within that unit.

[0085] The second step is to calculate the initial feature connection weights between any two directly adjacent voxel units in 3D space. Two voxel units are considered directly adjacent if they share a face, an edge, or a corner point in 3D space. For such a pair of adjacent units, the fused spatiotemporal feature vectors of all spatial points contained in each unit are extracted, and the similarity between each pair of feature vectors from the first unit and the second unit is calculated. The specific calculation method for similarity is to first calculate the cosine value of the angle between the two feature vectors, and then normalize the result to between 0 and 1. After calculating the similarity of all point pairs, the median of these similarity values ​​is taken as the initial feature connection weight between the two voxel units.

[0086] The third step is to verify and correct the aforementioned spatial proximity relationships to remove spurious adjacency connections caused by uneven point cloud density or noise. This verification is based on spatial continuity constraints. Specifically, for each pair of voxel units initially determined to be adjacent, it is checked whether there is continuous point cloud coverage in their common boundary region. By calculating the point projection density of the two units on their adjacent faces, if the difference in projection density between the two units exceeds 5 times, or if there is a continuous hole in the common boundary region greater than twice the voxel side length, the adjacency relationship is determined to be a spurious connection and removed. Only voxel unit pairs that pass this verification are considered to have true spatial continuity.

[0087] The fourth step is to complete the construction of the 3D graph structure. The validated voxel units are used as nodes in the graph structure, and the corrected initial feature connection weights are used as the weights of the edges connecting corresponding nodes. At this point, a 3D graph structure with voxel units as nodes, spatial continuity as the connection basis, and feature similarity as the connection weights is completed.

[0088] Next, a hierarchical node aggregation operation based on the aforementioned 3D graph structure is performed. In the first aggregation level, a first similarity threshold of 0.85 is set. All edges in the graph are traversed. If the feature similarity (i.e., edge weight) of two nodes connected by an edge is higher than 0.85, then these two nodes are merged into a primary aggregation unit. The feature vector of this primary aggregation unit is updated to the weighted average of the feature vectors of all nodes it contains, with the weight being the number of original spatial points contained in each node.

[0089] In subsequent second and higher aggregation levels, the generated primary aggregation units are iteratively merged. At this point, the criteria for merging two units are based on two points: the spatial proximity of the unit boundary points and the statistical consistency of the overall features of the units. Specifically, two primary aggregation units are considered spatially proximity if there are at least three pairs of boundary points (from different units) within a distance of less than 0.1 meters in three-dimensional space. Statistical consistency is measured by calculating the Mahalanobis distance between the feature vectors of the two units; if this distance is less than the 95% confidence interval threshold calculated based on the global feature distribution, they are considered consistent. Units that simultaneously satisfy the conditions of spatial proximity and statistical consistency will be merged to form higher-level semantic aggregation units.

[0090] Finally, the conditions for stopping aggregation are set. The aggregation process stops when the newly generated semantic aggregation unit meets the preset geometric integrity and feature homogeneity conditions. Geometric integrity means that the ratio of the volume of the circumscribed cube of the point cloud cluster formed by the unit in 3D space to the total volume of the voxels actually occupied by the cluster is less than 1.5, indicating that the shape is relatively compact. Feature homogeneity means that the magnitude of the standard deviation vector of all node feature vectors in the unit is less than 0.5 times the average magnitude of the standard deviation vector of all units in the entire scene. Semantic aggregation units that meet these conditions will be output as entity units with a single semantic identifier (such as "building exterior wall", "tree crown", "hardened ground"), thus completing the 3D semantic parsing.

[0091] In S5, firstly, for each identified entity unit, its temporal lifecycle information is determined. Specifically, this involves tracing back the timestamps of all spatial points corresponding to that entity unit in the three-dimensional spatial coordinate sequence. These timestamps record the specific time each spatial point was observed. By statistically analyzing the distribution of these timestamps, the time when the entity unit first appeared at all its corresponding spatial points is determined, and this time is recorded as its initial appearance time. Similarly, the last time all its corresponding spatial points were observed is determined, and this is recorded as its final disappearance time. If an entity unit still has corresponding spatial points at the end of the observation period, its disappearance time is marked as "persistent." In this way, the time nodes of each entity unit's appearance, persistence, or disappearance are precisely marked.

[0092] Secondly, based on the geometric boundaries of all entity units in a unified spatial coordinate system, spatial topological relationships are established between them. Specifically, the geometric boundary of each entity unit is calculated in parallel and represented by its minimum circumscribed cuboid or convex hull. The determination of spatial topological relationships is based on the positional relationships between these geometric boundaries. An inclusion relationship means that the geometric boundary of one entity unit is completely inside the geometric boundary of another entity unit, and the distance between the two boundaries is less than 0.01 meters. An adjacency relationship means that the shortest distance between the geometric boundaries of two entity units in three-dimensional space is less than 0.1 meters, and there is no third unit completely separating them. A disjoint relationship means that the shortest distance between the two boundaries is greater than or equal to 0.1 meters. By calculating these spatial relationships between all entity unit pairs in parallel, a complete set of spatial topological relationships is established.

[0093] Then, based on the appearance and disappearance times of entity units and their spatial topological relationships, the temporal evolution relationships between entity units are derived and recorded. The specific implementation includes four sub-steps. First, the entire observation period is divided into a series of continuous and non-overlapping time windows of fixed duration (e.g., 30 days). Within each time window, based on the current state of each entity unit and its spatial topological relationships, a unique relationship state code is generated for each pair of simultaneously existing entity units that have an inclusion or adjacency relationship. This code is an integer, where its binary bits represent the existence of a specific spatial relationship (e.g., inclusion, being included, adjacency).

[0094] Second, analyze the changes in the relational state encoding of each pair of entity units within a continuous time window sequence. By sequentially comparing the encoding values ​​of adjacent time windows, identify the critical time windows where the encoding values ​​change. Record the start and end times of these windows, as well as the specific encoding states before and after the change; this is the type of state transition.

[0095] Third, based on a pre-defined entity evolution rule base, the aforementioned state transition types are mapped to specific temporal evolution relationship types. This rule base is predefined based on the logic of geographic entity evolution. For example, if, within a critical time window, the relationship between entity units A and B changes from "separate" to "adjacent," and subsequently A disappears while B persists, and B inherits some spatial features from A after A's disappearance, this is mapped to a "replacement" relationship. If the relationship between two entity units changes from "separate" to "merging into a new unit," this is mapped to a "fusion" relationship. The main relationship types include derivation, fusion, splitting, and replacement.

[0096] Fourth, for each pair of entity units that have undergone a state transition, create a temporal evolution relationship record. This record contains the following fields: the identifiers of the two entity units whose relationship has changed, the identified temporal evolution relationship type (such as fusion), the start and end times of the critical time window, the relationship state code before the transition, and the relationship state code after the transition.

[0097] Finally, all information is integrated to generate a final structured information set. This set records the semantic identifier, geometric attributes (such as circumscribed cuboid parameters, volume, and surface area), and lifecycle time information (appearance time and disappearance time) for each entity unit. Simultaneously, this set contains two relationship tables: a static spatial topology relationship table, recording the inclusion, adjacency, or separation relationships between all entity unit pairs; and a dynamic temporal evolution relationship table, recording all derived temporal relationships such as derivation, fusion, splitting, or replacement. This structured information set comprehensively represents the state, association, and evolutionary history of all geographic entities within the surveyed target area in the spatiotemporal dimensions.

[0098] Please see Figure 2As shown, an intelligent mapping system based on remote sensing technology includes:

[0099] The multi-temporal remote sensing data acquisition module is used to acquire a set of multi-temporal remote sensing image data of the target area at multiple different time points;

[0100] The spatiotemporal data alignment and serialization module generates a three-dimensional spatial coordinate sequence in a unified spatial coordinate system based on a multi-time series of remote sensing image data. The three-dimensional spatial coordinate sequence contains the geometric and optical properties of each spatial point at multiple time points.

[0101] The fusion spatiotemporal feature generation module is used to jointly analyze the multiple temporal attributes of each spatial point in the three-dimensional spatial coordinate sequence. It is processed by a temporal analysis method that distinguishes between persistent structures and transient disturbances to generate fusion spatiotemporal features for each spatial point. Persistent structures are entity features that exist continuously at different time points, while transient disturbances are non-entity features caused by lighting, occlusion, or temporary objects.

[0102] The 3D semantic entity aggregation module performs 3D semantic analysis on the survey target area based on the spatiotemporal characteristics of all spatial points, and aggregates points with similar spatiotemporal characteristics and spatial continuity into entity units with a single semantic identifier.

[0103] The spatiotemporal knowledge structuring generation module generates a structured information set containing time dimension information based on the semantic identifiers and geometric attributes of all entity units. The structured information set marks each entity unit with at least one of the following time information: appearance time, duration of existence, and disappearance time, and establishes the spatial topology and temporal evolution relationship between each entity unit.

[0104] The working principle of this invention is as follows: First, a multi-temporal remote sensing image data set of the target area at multiple different time points is acquired. Second, a three-dimensional spatial coordinate sequence under a unified spatial coordinate system is generated based on this data set. This sequence contains the geometric and optical attributes of each spatial point at multiple time points. Next, by jointly analyzing the multi-temporal attributes of each spatial point, a temporal analysis method capable of distinguishing between persistent structures and transient disturbances is used to generate the fused spatiotemporal features of each spatial point. Then, based on the fused spatiotemporal features of all spatial points, a three-dimensional graph structure of fused feature similarity and spatial adjacency is constructed. Through hierarchical aggregation operations, points with similar features and spatial continuity are aggregated into entity units with a single semantic identifier. Finally, based on the semantic identifier and geometric attributes of all entity units, lifecycle time information is labeled for them, and spatial topological relationships and temporal evolution relationships between entities are automatically established, thereby generating a structured information set containing complete spatiotemporal dimensional information, realizing intelligent and automated analysis and recording of the status, association, and evolution of geographic entities.

[0105] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent mapping method based on remote sensing technology, characterized in that, Includes the following steps: S1: Acquire a set of multi-time series remote sensing image data of the target area at multiple different time points; S2: Based on a multi-time series remote sensing image data set, a three-dimensional spatial coordinate sequence under a unified spatial coordinate system is generated. The three-dimensional spatial coordinate sequence contains the geometric and optical properties of each spatial point at multiple time points. S3: Jointly analyze the multiple temporal attributes of each spatial point in the three-dimensional spatial coordinate sequence. Process the data by distinguishing between persistent structures and transient disturbances in the temporal analysis method to generate the fused spatiotemporal features of each spatial point. The persistent structures are the entity features that exist continuously at different time points, and the transient disturbances are the non-entity features caused by lighting, occlusion or temporary objects. S4: Based on the spatiotemporal characteristics of all spatial points, perform three-dimensional semantic analysis on the target area of ​​the survey, and aggregate points with similar spatiotemporal characteristics and spatial continuity into entity units with a single semantic identifier. S5: Based on the semantic identifiers and geometric attributes of all entity units, generate a structured information set containing time dimension information. The structured information set marks each entity unit with at least one of the following time information: appearance time, duration of existence, and disappearance time, and establishes the spatial topology and temporal evolution relationship between each entity unit.

2. The intelligent mapping method based on remote sensing technology according to claim 1, characterized in that, S2 specifically includes: Three-dimensional reconstruction is performed on the image data at each time point in the multi-time series remote sensing image data set to generate an initial three-dimensional point cloud for each independent time point. Each initial three-dimensional point cloud contains the spatial coordinates and optical attributes of the corresponding time point. The initial 3D point clouds at each independent time point are associated and precisely aligned point by point. By analyzing the local surface curvature and texture distribution stability of spatial points with the same name at adjacent time points, a multi-temporal dense matching point set with a unified spatial reference is output. Based on a multi-temporal dense matching point set, the spatial coordinates and optical properties of each successfully matched spatial point are integrated in chronological order at different time points to generate a three-dimensional spatial coordinate sequence.

3. The intelligent mapping method based on remote sensing technology according to claim 2, characterized in that, The output, which has a unified spatial reference, is a multi-temporal densely matched point set, specifically including: Multidimensional stable features are extracted from each spatial point of each initial 3D point cloud. These multidimensional stable features include high-order surface curvature features and local texture spectrum features that maintain statistical consistency between adjacent time points. A global joint optimization framework is constructed with feature consistency as a constraint and spatial transformation parameters as the optimization objective. The global joint optimization framework simultaneously minimizes the feature differences and geometric position residuals of matching points between all adjacent time point pairs. Solve the global joint optimization framework and simultaneously calculate a set of spatial transformation parameters that enable all time point clouds to achieve the best alignment state under a unified spatial reference. Spatial transformation parameters are applied to perform coordinate transformation and resampling on all initial 3D point clouds to generate multi-temporal dense matching point sets.

4. The intelligent mapping method based on remote sensing technology according to claim 1, characterized in that, S3 specifically includes: For each spatial point in a three-dimensional spatial coordinate sequence, the variation pattern of the optical properties of each spatial point at multiple different time points is analyzed, and the stability and periodicity of the property values ​​of each spatial point at different time points are calculated. Based on the stability and periodicity of attribute values, the optical attributes of spatial points at different time points are subjected to temporal adaptive weighted fusion to generate principal component features representing the persistent optical attributes of the corresponding spatial points. By comparing the deviations between the measured optical properties of spatial points at different time points and the principal component features, the anomalous attribute components characterizing transient disturbances are extracted and quantified. The principal component features, anomalous attribute components, and temporal statistics of geometric attributes of combined spatial points are used to form the fused spatiotemporal features of the corresponding spatial points.

5. The intelligent mapping method based on remote sensing technology according to claim 4, characterized in that, The extraction and quantification of anomalous attribute components characterizing transient disturbances specifically includes: Calculate the attribute deviation between the measured optical attribute values ​​of a spatial point at each time point and the principal component eigenvalues ​​at the corresponding time point; Sequence analysis was performed on the attribute deviation values ​​at all time points to identify and separate isolated deviation patterns from periodic deviation patterns with regular fluctuation characteristics. The intensity and duration of isolated bias patterns and periodic bias patterns are quantified respectively to generate transient anomalous components that characterize sudden light events and cyclic anomalous components that characterize seasonal vegetation shading. The transient anomaly component and the cyclic anomaly component are merged and encoded to form the anomaly attribute component of the corresponding spatial point.

6. The intelligent mapping method based on remote sensing technology according to claim 1, characterized in that, S4 specifically includes: Based on the spatiotemporal characteristics of all spatial points, a three-dimensional graph structure that integrates feature similarity and spatial adjacency is constructed. Perform a hierarchical node aggregation operation based on a 3D graph structure. In each level, the hierarchical node aggregation operation first merges nodes with feature similarity higher than a first set threshold and spatially directly adjacent nodes to generate a primary aggregation unit. In subsequent levels, the generated primary aggregation units are iteratively merged based on the spatial proximity of the boundary points of the primary aggregation units and the statistical consistency of the spatiotemporal features of the overall fusion of the units, to form higher-level semantic aggregation units. When a semantic aggregation unit meets the preset conditions of geometric integrity and feature homogeneity, the aggregation process stops, and the semantic aggregation unit is output as an entity unit with a single semantic identifier.

7. The intelligent mapping method based on remote sensing technology according to claim 6, characterized in that, The construction of the three-dimensional graph structure, which integrates feature similarity and spatial adjacency, specifically includes: Based on the three-dimensional coordinates of spatial points, multi-scale spatial unit division is carried out to generate three-dimensional voxel units with hierarchical relationships, and each spatial point is assigned to the corresponding smallest scale voxel unit. Calculate the spatiotemporal feature similarity of all spatial point pairs within any two directly adjacent voxel units in three-dimensional space, and select the statistical median of the similarity as the initial feature connection weight between the two voxel units. By verifying and correcting the spatial proximity relationships of voxel units, false spatial adjacency connections caused by uneven point cloud density or noise are removed. By using voxel units that satisfy spatial continuity constraints as graph nodes and modifying the initial feature connection weights as edge weights between corresponding nodes, the 3D graph structure is constructed.

8. The intelligent mapping method based on remote sensing technology according to claim 1, characterized in that, S5 specifically includes: By tracing back the timestamps of all spatial points corresponding to each entity unit in the three-dimensional spatial coordinate sequence, the initial appearance time and final disappearance time of the entity unit can be determined. Based on the geometric boundaries of all entity units in a unified spatial coordinate system, parallel determination of three-dimensional spatial relationships is performed to establish spatial topological relationships of inclusion, adjacency, and separation between entity units; Based on the appearance and disappearance time of entity units and the spatial topological relationship of entity units, the temporal evolution relationship between entity units is derived and recorded by analyzing the state changes of the topological relationship in different time windows. The semantic identifiers, geometric attributes, temporal information, spatial topological relationships, and temporal evolution relationships of each entity unit are integrated to generate a structured information set.

9. The intelligent mapping method based on remote sensing technology according to claim 8, characterized in that, The derivation and recording of the temporal evolution relationship between entity units specifically includes: Based on the appearance and disappearance time of entity units, the entire observation period is divided into continuous time windows, and within each time window, a relational state code is generated for each pair of entity units with topological relationships according to spatial topological relationships. Analyze the changes in the entity unit relationship state encoding in the continuous time window sequence, identify the critical time window when the relationship state changes, and record the specific type of state change; Based on the preset entity evolution rule base, the state transition type is mapped to a specific time evolution relationship type, which includes derivation, fusion, splitting and replacement; For each pair of entity units that undergo a state transition, record the type of time evolution relationship corresponding to the entity unit, the critical time window, and the relationship state encoding before and after the transition, thus completing the derivation and recording of the time evolution relationship.

10. An intelligent mapping system based on remote sensing technology, characterized in that, An intelligent mapping method based on remote sensing technology as described in any one of claims 1-9 includes: The multi-temporal remote sensing data acquisition module is used to acquire a set of multi-temporal remote sensing image data of the target area at multiple different time points; The spatiotemporal data alignment and serialization module generates a three-dimensional spatial coordinate sequence in a unified spatial coordinate system based on a multi-time series of remote sensing image data. The three-dimensional spatial coordinate sequence contains the geometric and optical properties of each spatial point at multiple time points. The fusion spatiotemporal feature generation module is used to jointly analyze the multiple temporal attributes of each spatial point in the three-dimensional spatial coordinate sequence. It is processed by a temporal analysis method that distinguishes between persistent structures and transient disturbances to generate fusion spatiotemporal features for each spatial point. Persistent structures are entity features that exist continuously at different time points, while transient disturbances are non-entity features caused by lighting, occlusion, or temporary objects. The 3D semantic entity aggregation module performs 3D semantic analysis on the survey target area based on the spatiotemporal characteristics of all spatial points, and aggregates points with similar spatiotemporal characteristics and spatial continuity into entity units with a single semantic identifier. The spatiotemporal knowledge structuring generation module generates a structured information set containing time dimension information based on the semantic identifiers and geometric attributes of all entity units. The structured information set marks each entity unit with at least one of the following time information: appearance time, duration of existence, and disappearance time, and establishes the spatial topology and temporal evolution relationship between each entity unit.

Images