Remote sensing information processing method and system for calibrating terrain

Abstract

The invention discloses a terrain calibration remote sensing information processing method and system, belongs to the technical field of remote sensing surveying and mapping, and is used for solving the problems that a traditional remote sensing data processing method limits the utilization depth of dynamic characteristics and extensive resource allocation aggravates the processing efficiency bottleneck. And large-range, high-time-efficiency and multi-element comprehensive surveying and mapping requirements are difficult to meet. The method comprises the following steps: respectively extracting geometrical characteristics of a fixed marker and time sequence characteristics of a dynamic marker; fusing the geometric features and the time sequence features to obtain fused features; performing terrain partitioning on the target region based on the fusion features to obtain different types of calibrated terrain regions; according to the processing priority of the calibrated terrain area, remote sensing information processing is carried out through an adaptive processing algorithm, and terrain change information is obtained; constructing a dynamic terrain inversion model to obtain terrain inversion change information; and carrying out cross validation on the topographic change information and the topographic inversion change information to obtain accurate topographic change information.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing mapping, and in particular to a remote sensing information processing method and system for calibrating terrain. Background Art

[0002] In the field of remote sensing and mapping, although existing technologies have made certain progress, traditional remote sensing data processing technology has long been limited by static processing frameworks and extensive resource allocation models, and still faces certain technical bottlenecks, which are mainly reflected in the following three dimensions:

[0003] 1. Spatial adaptability defects caused by the static processing paradigm: Existing technical systems generally adopt a fixed feature labeling strategy based on prior knowledge, which simply defines artificial facilities such as towers and buildings as geometric reference points. This processing method has two limitations: first, the manual labeling process is highly dependent on the subjective experience of the operator, which is prone to position drift and attribute misjudgment under complex terrain conditions; second, fixed features are only used as terrain spatial positioning benchmarks and are not included in the dynamic partitioning optimization process. Traditional methods usually use regular grids (such as UTM partitions) or administrative boundaries for data segmentation, resulting in urban buildings and surrounding farmland being forced into the same processing unit, which not only destroys the spatial continuity of the features, but also causes redundant calculations in urban areas due to over-segmentation, while linear features such as ridges in farmland areas are blurred due to insufficient resolution.

[0004] 2. Lack of value extraction of temporal dynamic features: For landforms with significant temporal variations, such as vegetation and water bodies, traditional technology systems only extract single-phase spectral features, completely ignoring the following key dimensions: NDVI fluctuations caused by changes in vegetation phenology, seasonal variations in chlorophyll concentration in water bodies, and the interaction between dynamic landforms and topographical elements (such as the spatial expansion of flood inundation areas). This processing model leads to two major technical flaws: first, the lack of a data verification mechanism under temporal constraints; second, the inability to invert the terrain evolution process based on the changes in dynamic landform characteristics.

[0005] 3. Resource mismatch in a globally unified processing architecture: Traditional systems employ a one-size-fits-all processing strategy, applying a unified process to all terrain data. This includes setting resolution using the same spatial resolution for both simple terrain (e.g., plains) and complex terrain (e.g., mountains). Algorithm parameters employ fixed threshold combinations, disregarding the impact of terrain relief on spectral response. Furthermore, computational resource allocation prioritizes both simple and complex terrain equally. This model results in a double waste of resources: redundant computation in simple terrain areas and feature loss in complex terrain areas.

[0006] The above technical defects form a chain reaction: the static processing framework limits the depth of utilization of dynamic features, and the extensive resource allocation aggravates the bottleneck of processing efficiency. These three factors together make it difficult for traditional remote sensing data processing systems to cope with the large-scale, high-efficiency, and multi-factor comprehensive terrain mapping needs. Summary of the Invention

[0007] Embodiments of the present invention provide a remote sensing information processing method and system for terrain calibration, which are used to solve the following technical problems: the static processing framework of traditional remote sensing data processing methods limits the depth of utilization of dynamic features, while extensive resource allocation exacerbates the processing efficiency bottleneck, making it difficult to meet the needs of large-scale, high-efficiency, and multi-factor comprehensive terrain mapping.

[0008] The embodiment of the present invention adopts the following technical solutions:

[0009] In one aspect, an embodiment of the present invention provides a remote sensing information processing method for calibrating terrain, the method comprising: extracting geometric features of fixed markers and temporal features of dynamic markers based on historical remote sensing information of a target area;

[0010] Based on a two-stream neural network, a multimodal feature fusion model is constructed, and the geometric features and the temporal features are fused to obtain fused features;

[0011] Constructing a dynamic terrain partition model, and performing terrain partitioning on the target area based on the fusion features to obtain different types of calibrated terrain areas;

[0012] Acquire real-time remote sensing information of each calibrated terrain area and call the corresponding adaptive processing algorithm; process the remote sensing information through the adaptive processing algorithm according to the processing priority of the calibrated terrain area to obtain terrain change information;

[0013] Building a dynamic terrain inversion model based on the LSTM network, and inputting the real-time remote sensing information and historical remote sensing information into the dynamic terrain inversion model to obtain terrain inversion change information;

[0014] The terrain change information is cross-validated with the terrain inversion change information to obtain accurate terrain change information.

[0015] In a feasible implementation, based on the historical remote sensing information of the target area, the geometric features of the fixed markers and the temporal features of the dynamic markers are extracted respectively, specifically including:

[0016] Acquire historical remote sensing data collected over multiple historical periods of the target area; wherein the historical remote sensing data includes at least: SAR satellite remote sensing data, airborne lidar point cloud data, and multispectral image data;

[0017] Inputting the multispectral image data into the DeepLabV3+ network for target recognition and classification, obtaining fixed markers and dynamic markers in the target area; wherein the dynamic markers include at least vegetation and water bodies;

[0018] Based on the closest point search algorithm, SAR satellite remote sensing data collected over multiple historical periods are aligned with airborne lidar point cloud data, and the RANSAC algorithm is used to extract features from the aligned data to obtain the geometric features of the fixed marker; wherein the geometric features include at least the outer contour features and the inclination angle features relative to the ground;

[0019] Extracting the Normalized Vegetation Index (NDVI) and the Normalized Water Index (NDWI) from each frame of multispectral image data of the historical remote sensing information, and constructing NDVI time series data and NDWI time series data;

[0020] The time series characteristics of the dynamic marker are extracted from the NDVI time series data and the NDWI time series data; wherein the time series characteristics include at least the growth state characteristics and area change characteristics of vegetation, and the texture change characteristics and area change characteristics of water bodies.

[0021] In a feasible implementation, a multimodal feature fusion model is constructed based on a two-stream neural network, and the geometric features and the temporal features are fused to obtain fused features, specifically including:

[0022] Based on the standard residual network, a coordinate attention module is inserted after each residual block, and each channel feature is adaptively weighted through the SE module to obtain a geometric flow neural network;

[0023] Construct a Transformer encoder and alternate between window attention and global attention in the Transformer encoder to obtain a temporal stream neural network;

[0024] Connecting the output end of the geometric stream neural network and the output end of the temporal stream neural network to the cross attention module respectively to form a two-stream neural network;

[0025] Construct a multi-level feature pyramid and fuse features of different levels using bilinear interpolation upsampling and element-by-element addition;

[0026] Connecting the multi-level feature pyramid to the output end of the cross attention module to form the multimodal feature fusion model together with the two-stream neural network;

[0027] The geometric features and the temporal features are input into the multimodal feature fusion model, and fusion features are output.

[0028] In a feasible implementation, constructing a dynamic terrain partitioning model specifically includes:

[0029] Constructing a multi-level partitioning architecture including a coarse-grained administrative partitioning module, a medium-grained clustering partitioning module, and a fine-grained supervoxel partitioning module to obtain the dynamic terrain partitioning model;

[0030] The specific construction methods include:

[0031] Obtaining administrative division vector surface data and constructing the coarse-grained administrative division module for dividing administrative divisions and extracting regional features of each administrative division;

[0032] Based on the DBSCAN clustering algorithm, the medium-granularity cluster partitioning module is constructed to perform cluster analysis on the fusion features, determine each cluster partition and extract regional features;

[0033] Based on the supervoxel segmentation algorithm, the fine-grained supervoxel partitioning module is constructed to voxelize the airborne lidar point cloud data in the historical remote sensing information to obtain voxel cloud data, and perform supervoxel segmentation on the voxel cloud data to obtain various supervoxel partitions and extract regional features.

[0034] In a feasible implementation, the target area is topographically partitioned based on the fusion features to obtain different types of calibrated terrain areas, specifically including:

[0035] Inputting the fused features into the dynamic terrain partition model to obtain a plurality of administrative partitions, cluster partitions, and supervoxel partitions, as well as regional features corresponding to each partition; wherein the regional features include at least: building density, texture entropy, night light value, NDVI seasonal amplitude, land regularity, elevation difference, NDWI, surface temperature, texture contrast, NIR reflectivity, red edge band slope, and elevation;

[0036] Determine the terrain type of each partition based on the terrain discrimination conditions satisfied by the regional characteristics corresponding to each partition; wherein the terrain types include at least: urban building area, farmland area, water area, forest area and mountain area;

[0037] Each partition is calibrated according to its terrain type to obtain different types of calibrated terrain areas.

[0038] In a feasible implementation, before obtaining the real-time remote sensing information of each calibrated terrain area and calling the corresponding adaptive processing algorithm, the method further includes:

[0039] Different adaptive processing algorithms are constructed for different calibrated terrain areas, and a remote sensing information processing algorithm library is built; among them, a high-precision geometric processing algorithm is used in urban built-up areas, a time-series spectral analysis algorithm is used in farmland areas, a sub-pixel decomposition algorithm is used in water areas, and a three-dimensional point cloud processing algorithm is used in forest areas and mountainous areas;

[0040] The processing priority of each calibrated terrain area is determined according to the historical regional change rate of the calibrated terrain area; among them, the calibrated terrain area with the largest regional change law has the highest processing priority.

[0041] In a feasible implementation, real-time remote sensing information of each calibrated terrain area is obtained, and a corresponding adaptive processing algorithm is called; remote sensing information is processed by the adaptive processing algorithm according to the processing priority of the calibrated terrain area to obtain terrain change information, specifically including:

[0042] Obtaining the terrain type of the current calibrated terrain area, and extracting the corresponding adaptive processing algorithm from the remote sensing information processing algorithm library according to the terrain type;

[0043] Calculate the historical area change rate of the current calibrated terrain area, compare and sort it with other calibrated terrain areas in the target area, and determine the processing priority of the current calibrated terrain area;

[0044] After the processing of the calibrated terrain area of the previous processing priority is completed, the real-time remote sensing information of the current calibrated terrain area is processed by the adaptive processing algorithm to obtain the terrain change information.

[0045] In a feasible implementation, a dynamic terrain inversion model is constructed based on an LSTM network, specifically including:

[0046] A three-layer LSTM network structure is used, and residual connections are added between the second and third LSTM network structures to construct a stacked LSTM network. The three-layer LSTM network structure is used to capture short-term temporal dependencies, medium-term temporal dependencies, and long-term temporal dependencies respectively.

[0047] Construct an attention enhancement module and connect it to the output of the stacked LSTM network to dynamically adjust the importance weight of each time step and the contribution weight of the input features;

[0048] The output end of the attention enhancement module is connected to the output layer to complete the construction of the dynamic terrain inversion model, and the model is trained using the pre-built NASA GEDI laser elevation dataset.

[0049] In a feasible implementation, cross-validating the terrain change information with the terrain inversion change information to obtain accurate terrain change information specifically includes:

[0050] Extracting a first change feature set from the terrain change information, and extracting a second change feature set from the terrain inversion change information;

[0051] Calculating feature consistency between each change feature in the first change feature set and the corresponding change feature in the second change feature set;

[0052] Determine the changed features whose feature consistency is lower than a preset threshold as abnormal features, send an early warning and carry the location information of the abnormal features to the maintenance terminal;

[0053] After receiving the confirmation features returned by the maintenance terminal, they are combined with the non-abnormal features to form the precise terrain change information.

[0054] On the other hand, an embodiment of the present invention further provides a remote sensing information processing system for calibrating terrain, the system comprising:

[0055] The terrain calibration module is used to extract the geometric features of fixed markers and the temporal features of dynamic markers based on the historical remote sensing information of the target area; construct a multimodal feature fusion model based on a two-stream neural network, and fuse the geometric features with the temporal features to obtain fused features; construct a dynamic terrain partitioning model, and partition the target area into different types of calibrated terrain areas based on the fused features;

[0056] A remote sensing information processing module is used to obtain real-time remote sensing information of each calibrated terrain area and call the corresponding adaptive processing algorithm; based on the processing priority of the calibrated terrain area, the remote sensing information is processed by the adaptive processing algorithm to obtain terrain change information;

[0057] A terrain inversion module is used to build a dynamic terrain inversion model based on an LSTM network, and input the real-time remote sensing information and historical remote sensing information into the dynamic terrain inversion model to obtain terrain inversion change information;

[0058] The cross-validation module is used to cross-validate the terrain change information with the terrain inversion change information to obtain accurate terrain change information.

[0059] Compared with the prior art, the remote sensing information processing method and system for terrain calibration provided by the embodiments of the present invention have the following beneficial effects:

[0060] 1. This invention uses a dynamic terrain zoning model to achieve intelligent segmentation of terrain units, enabling differentiated processing granularity for different characteristic areas such as urban buildings and farmland. This effectively solves the spatial mismatch problem caused by traditional grid division.

[0061] 2. This invention establishes a terrain complexity-computing resource mapping model through an adaptive processing algorithm, improving spectral feature retention in complex terrain areas such as mountains while reducing computational redundancy in plains. A dynamic priority scheduling mechanism achieves a precise match between computing resources and terrain complexity.

[0062] 3. The geometric-temporal fusion feature space constructed by the dual-stream neural network significantly enhances the ability to analyze spatiotemporal correlations between terrain features, providing a richer feature dimension for terrain evolution inversion. Furthermore, the constructed LSTM dynamic terrain inversion model maximizes the value of time series data.

[0063] 4. Finally, the bidirectional verification mechanism for terrain change information and inversion information in this invention significantly enhances the credibility of the final results, forming a complete chain of evidence from observation to mechanism inversion. This mechanism effectively eliminates the uncertainty of single-source data and establishes a closed-loop system for multi-dimensional verification.

[0064] This invention forms a disruptive improvement to the traditional static processing framework by constructing an innovative processing paradigm of "feature fusion-dynamic partitioning-intelligent processing-inversion verification". It shows significant technical advantages in key areas such as complex terrain processing, dynamic land feature monitoring, and resource optimization and allocation, laying a core methodological foundation for building a new generation of intelligent remote sensing processing system. BRIEF DESCRIPTION OF THE DRAWINGS

[0065] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments described in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0066] Figure 1 A flow chart of a remote sensing information processing method for calibrating terrain provided by an embodiment of the present invention;

[0067] Figure 2 A schematic structural diagram of a remote sensing information processing system for terrain calibration provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0068] In order to enable those skilled in the art to better understand the technical solutions of the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0069] The embodiment of the present invention provides a remote sensing information processing method for calibrating terrain, such as Figure 1 As shown, the remote sensing information processing method for calibrating terrain specifically includes steps S101-S105:

[0070] S101. Based on the historical remote sensing information of the target area, the geometric features of the fixed markers and the temporal features of the dynamic markers are extracted respectively.

[0071] Specifically, historical remote sensing data collected in multiple historical periods of the target area are obtained; wherein the historical remote sensing data at least includes: SAR satellite remote sensing data, airborne lidar point cloud data and multispectral image data.

[0072] Furthermore, the multispectral image data is input into the DeepLabV3+ network for target recognition and classification to obtain fixed markers and dynamic markers in the target area; among them, dynamic markers include at least vegetation and water bodies.

[0073] In one embodiment, fixed markers include fixed objects on the ground, such as buildings, towers, and dams, while dynamic markers include objects that change with the seasons, such as vegetation and water bodies.

[0074] Furthermore, based on the nearest point search algorithm, the SAR satellite remote sensing data collected in multiple historical periods are aligned with the airborne lidar point cloud data, and the features of the aligned data are extracted by the random sampling consistency RANSAC algorithm to obtain the geometric features of the fixed markers; wherein, the geometric features include at least the outer contour features and the inclination features with the ground.

[0075] Furthermore, the Normalized Difference Vegetation Index (NDVI) and the Normalized Water Index (NDWI) are extracted from each frame of multispectral image data from historical remote sensing information, and NDVI and NDWI time series data are constructed. Time series features of dynamic markers are extracted from the NDVI and NDWI time series data; these time series features include at least vegetation growth characteristics and area change characteristics, as well as water body texture and area change characteristics.

[0076] S102. Based on the dual-stream neural network, a multimodal feature fusion model is constructed, and the geometric features and temporal features are fused to obtain fused features.

[0077] Specifically, based on the standard residual network, a coordinate attention module is inserted after each residual block, and each channel feature is adaptively weighted through the SE module to obtain a geometric flow neural network.

[0078] Then, a Transformer encoder is constructed, and window attention and global attention are alternately used in the Transformer encoder to obtain a temporal stream neural network. The output of the geometric stream neural network and the output of the temporal stream neural network are connected to the cross attention module respectively to form a two-stream neural network.

[0079] As a feasible implementation, the Transformer encoder uses an 8-head self-attention mechanism with a hidden layer dimension of 512. Two attention modes are applied alternately: window attention, which calculates attention weights within a local 8×8 window, and global attention, which calculates attention over the full 128×128 spatial range.

[0080] Furthermore, a multi-level feature pyramid is constructed, and features at different levels are fused using bilinear interpolation upsampling and element-by-element addition. The multi-level feature pyramid is connected to the output of the cross-attention module, and together with the two-stream neural network, it forms a multimodal feature fusion model.

[0081] Finally, the geometric features and temporal features are input into the multimodal feature fusion model and the fused features are output.

[0082] As a feasible implementation, the cross-attention module extracts a 256-dimensional query vector from the geometric stream features and generates a 512-dimensional key-value pair based on the temporal stream features. The query vector is then divided into eight heads (each with 32 dimensions) for multi-head attention calculation, calculating the similarity matrix between each head and the key. Finally, attention weights are generated through softmax normalization, and the weighted aggregate value vector is added to obtain the fused feature.

[0083] S103: Construct a dynamic terrain partition model, and perform terrain partitioning on the target area based on the fusion features to obtain different types of calibrated terrain areas.

[0084] Specifically, a multi-level partitioning architecture including a coarse-grained administrative partitioning module, a medium-grained clustering partitioning module and a fine-grained supervoxel partitioning module is constructed to obtain a dynamic terrain partitioning model.

[0085] As a feasible implementation method, the specific construction method includes:

[0086] Administrative division vector surface data was obtained and a coarse-grained administrative division module was constructed to divide administrative divisions and extract regional features for each administrative division. Based on the DBSCAN clustering algorithm, a medium-grained clustering and partitioning module was constructed to perform cluster analysis on fused features, identify individual cluster partitions, and extract regional features. Based on the supervoxel segmentation algorithm, a fine-grained supervoxel partitioning module was constructed to voxelize airborne lidar point cloud data from historical remote sensing information to obtain voxel cloud data. This voxel cloud data was then subjected to supervoxel segmentation to obtain individual supervoxel partitions and extract regional features.

[0087] Furthermore, the fused features are input into the dynamic terrain zoning model to obtain several administrative divisions, cluster divisions and supervoxel divisions, as well as the regional features corresponding to each division; among which, the regional features include at least: building density, texture entropy, night light value, NDVI seasonal amplitude, plot regularity, elevation difference, NDWI, surface temperature, texture contrast, NIR reflectivity, red edge band slope and elevation.

[0088] Furthermore, the terrain type of each partition is determined based on the terrain discrimination conditions satisfied by the regional characteristics corresponding to each partition; wherein the terrain types include at least: urban building area, farmland area, water area, forest area and mountain area.

[0089] Finally, each partition is calibrated according to its terrain type to obtain different types of calibrated terrain areas.

[0090] As a feasible implementation method, a feature-type mapping table is pre-built, and the fields of the table include terrain type and discrimination conditions. For example, for an urban building area, if a certain area characteristic satisfies both "building density>15 / km 2 , texture entropy>6.5, night light value>50", then the area is an urban built-up area. For farmland areas, if the characteristics of a certain area simultaneously meet at least two of the following: "NDVI seasonal amplitude>0.3, plot regularity>0.7, elevation difference<2m", then the area is a farmland area. For water areas, if the characteristics of a certain area simultaneously meet the main item + any auxiliary item of "NDWI>0.2 (main item), surface temperature<surrounding 2℃ (auxiliary item), texture contrast<25 (auxiliary item)", then the area is a water area. For woodland areas, if the characteristics of a certain area meet at least two of the following: "NIR reflectance>0.4, red edge band slope>0.05, elevation>300m", then the area is a woodland area.

[0091] S104: Acquire real-time remote sensing information of each calibrated terrain area and call the corresponding adaptive processing algorithm; process the remote sensing information through the adaptive processing algorithm according to the processing priority of the calibrated terrain area to obtain terrain change information.

[0092] Specifically, different adaptive processing algorithms are constructed for different calibrated terrain areas, and a remote sensing information processing algorithm library is built; among them, a high-precision geometric processing algorithm is used in urban building areas, a time series spectral analysis algorithm is used in farmland areas, a sub-pixel decomposition algorithm is used in water areas, and a three-dimensional point cloud processing algorithm is used in forest areas and mountainous areas.

[0093] Furthermore, the processing priority of each calibrated terrain area is determined according to the historical regional change rate of the calibrated terrain area; wherein the calibrated terrain area with the largest regional change law has the highest processing priority.

[0094] Furthermore, the terrain type of the currently calibrated terrain area is obtained, and the corresponding adaptive processing algorithm is extracted from the remote sensing information processing algorithm library based on the terrain type. The historical regional change rate of the currently calibrated terrain area is calculated and compared and ranked with other calibrated terrain areas within the target area to determine the processing priority of the currently calibrated terrain area.

[0095] After the processing of the calibrated terrain area of the previous processing priority is completed, the real-time remote sensing information of the current calibrated terrain area is processed by the adaptive processing algorithm to obtain terrain change information.

[0096] S105. Construct a dynamic terrain inversion model based on the LSTM network, and input real-time remote sensing information and historical remote sensing information into the dynamic terrain inversion model to obtain terrain inversion change information; cross-validate the terrain change information with the terrain inversion change information to obtain accurate terrain change information.

[0097] Specifically, a three-layer LSTM network structure is adopted, and a residual connection is added between the second-layer LSTM network structure and the third-layer LSTM network structure to construct a stacked LSTM network; among them, the three-layer LSTM network structure is used to capture short-term timing dependencies, medium-term timing dependencies, and long-term timing dependencies, respectively.

[0098] Furthermore, an attention enhancement module is constructed and connected to the output of the stacked LSTM network to dynamically adjust the importance weights of each time step and the contribution weights of the input features. The output of the attention enhancement module is connected to the output layer to complete the dynamic terrain inversion model, which is then trained using the pre-built NASA GEDI laser elevation dataset.

[0099] Furthermore, a first change feature set is extracted from the terrain change information, and a second change feature set is extracted from the terrain inversion change information. The feature consistency of each change feature in the first change feature set and the corresponding change feature in the second change feature set is calculated.

[0100] Changes with feature consistency below a preset threshold are identified as abnormal. An alert is sent along with the abnormal feature's location information to the maintenance terminal. After receiving the confirmation signal from the maintenance terminal, it is combined with non-abnormal features to generate precise terrain change information. This precise terrain change information can then be displayed on a large screen.

[0101] As a feasible implementation method, if there are too many abnormal features in a survey, an algorithm iteration warning will be sent to check whether the current model and algorithm no longer meet the characteristics of the current target area, and remind maintenance personnel to adjust parameters and conduct additional training.

[0102] In addition, the embodiment of the present invention also provides a remote sensing information processing system for calibrating terrain, such as Figure 2 As shown, the remote sensing information processing system 200 for calibrating terrain specifically includes:

[0103] The terrain calibration module 210 is configured to extract the geometric features of fixed markers and the temporal features of dynamic markers based on historical remote sensing information of the target area; construct a multimodal feature fusion model based on a two-stream neural network, and fuse the geometric features with the temporal features to obtain fused features; construct a dynamic terrain partitioning model, and partition the target area into different types of calibrated terrain areas based on the fused features;

[0104] Remote sensing information processing module 220 is used to obtain real-time remote sensing information of each calibrated terrain area and call the corresponding adaptive processing algorithm; based on the processing priority of the calibrated terrain area, the remote sensing information is processed by the adaptive processing algorithm to obtain terrain change information;

[0105] A terrain inversion module 230 is configured to construct a dynamic terrain inversion model based on an LSTM network, and input the real-time remote sensing information and historical remote sensing information into the dynamic terrain inversion model to obtain terrain inversion change information;

[0106] The cross-validation module 240 is configured to cross-validate the terrain change information with the terrain inversion change information to obtain accurate terrain change information.

[0107] The various embodiments of the present invention are described in a progressive manner. Similar portions between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from the other embodiments. In particular, the device, apparatus, and non-volatile computer storage medium embodiments are generally similar to the method embodiments, so their descriptions are simplified. For relevant details, refer to the descriptions of the method embodiments.

[0108] The above description of specific embodiments of the present invention is provided. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0109] The foregoing is merely an embodiment of the present invention and is not intended to limit the present invention. It will be apparent to those skilled in the art that various modifications and variations may be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A remote sensing information processing method for calibrating terrain, characterized in that: The method comprises: Based on the historical remote sensing information of the target area, the geometric features of the fixed markers and the temporal features of the dynamic markers are extracted respectively; Based on a two-stream neural network, a multimodal feature fusion model is constructed, and the geometric features and the temporal features are fused to obtain fused features; Constructing a dynamic terrain partition model, and performing terrain partitioning on the target area based on the fusion features to obtain different types of calibrated terrain areas; Acquire real-time remote sensing information of each calibrated terrain area and call the corresponding adaptive processing algorithm; process the remote sensing information through the adaptive processing algorithm according to the processing priority of the calibrated terrain area to obtain terrain change information; Building a dynamic terrain inversion model based on the LSTM network, and inputting the real-time remote sensing information and historical remote sensing information into the dynamic terrain inversion model to obtain terrain inversion change information; The terrain change information is cross-validated with the terrain inversion change information to obtain accurate terrain change information.

2. A remote sensing information processing method for calibrating terrain according to claim 1, characterized in that: Based on the historical remote sensing information of the target area, the geometric features of fixed markers and the temporal features of dynamic markers are extracted respectively, including: Acquire historical remote sensing data collected over multiple historical periods of the target area; wherein the historical remote sensing data includes at least: SAR satellite remote sensing data, airborne lidar point cloud data, and multispectral image data; Inputting the multispectral image data into the DeepLabV3+ network for target recognition and classification, obtaining fixed markers and dynamic markers in the target area; wherein the dynamic markers include at least vegetation and water bodies; Based on the closest point search algorithm, SAR satellite remote sensing data collected over multiple historical periods are aligned with airborne lidar point cloud data, and the RANSAC algorithm is used to extract features from the aligned data to obtain the geometric features of the fixed marker; wherein the geometric features include at least the outer contour features and the inclination angle features relative to the ground; Extracting the Normalized Vegetation Index (NDVI) and the Normalized Water Index (NDWI) from each frame of multispectral image data of the historical remote sensing information, and constructing NDVI time series data and NDWI time series data; The time series characteristics of the dynamic marker are extracted from the NDVI time series data and the NDWI time series data; wherein the time series characteristics include at least the growth state characteristics and area change characteristics of vegetation, and the texture change characteristics and area change characteristics of water bodies.

3. The remote sensing information processing method for demarcating terrain according to claim 1, characterized in that: Based on the two-stream neural network, a multimodal feature fusion model is constructed, and the geometric features and the temporal features are fused to obtain fused features, which specifically include: Based on the standard residual network, a coordinate attention module is inserted after each residual block, and each channel feature is adaptively weighted through the SE module to obtain a geometric flow neural network; Construct a Transformer encoder and alternate between window attention and global attention in the Transformer encoder to obtain a temporal stream neural network; Connecting the output end of the geometric stream neural network and the output end of the temporal stream neural network to the cross attention module respectively to form a two-stream neural network; Construct a multi-level feature pyramid and fuse features of different levels using bilinear interpolation upsampling and element-by-element addition; Connecting the multi-level feature pyramid to the output end of the cross attention module to form the multimodal feature fusion model together with the two-stream neural network; The geometric features and the temporal features are input into the multimodal feature fusion model, and fusion features are output.

4. The remote sensing information processing method for demarcating terrain according to claim 1, characterized in that: Construct a dynamic terrain zoning model, including: Constructing a multi-level partitioning architecture including a coarse-grained administrative partitioning module, a medium-grained clustering partitioning module, and a fine-grained supervoxel partitioning module to obtain the dynamic terrain partitioning model; The specific construction methods include: Obtaining administrative division vector surface data and constructing the coarse-grained administrative division module for dividing administrative divisions and extracting regional features of each administrative division; Based on the DBSCAN clustering algorithm, the medium-granularity cluster partitioning module is constructed to perform cluster analysis on the fusion features, determine each cluster partition and extract regional features; Based on the supervoxel segmentation algorithm, the fine-grained supervoxel partitioning module is constructed to voxelize the airborne lidar point cloud data in the historical remote sensing information to obtain voxel cloud data, and perform supervoxel segmentation on the voxel cloud data to obtain various supervoxel partitions and extract regional features.

5. The remote sensing information processing method for demarcating terrain according to claim 4, characterized in that: The target area is divided into terrain regions based on the fusion features to obtain different types of calibrated terrain regions, specifically including: Inputting the fused features into the dynamic terrain partition model to obtain a plurality of administrative partitions, cluster partitions, and supervoxel partitions, as well as regional features corresponding to each partition; wherein the regional features include at least: building density, texture entropy, night light value, NDVI seasonal amplitude, land regularity, elevation difference, NDWI, surface temperature, texture contrast, NIR reflectivity, red edge band slope, and elevation; Determine the terrain type of each partition based on the terrain discrimination conditions satisfied by the regional characteristics corresponding to each partition; wherein the terrain types include at least: urban building area, farmland area, water area, forest area and mountain area; Each partition is calibrated according to its terrain type to obtain different types of calibrated terrain areas.

6. The remote sensing information processing method for demarcating terrain according to claim 1, characterized in that: Before obtaining the real-time remote sensing information of each calibrated terrain area and calling the corresponding adaptive processing algorithm, the method further includes: Different adaptive processing algorithms are constructed for different calibrated terrain areas, and a remote sensing information processing algorithm library is built; among them, a high-precision geometric processing algorithm is used in urban built-up areas, a time-series spectral analysis algorithm is used in farmland areas, a sub-pixel decomposition algorithm is used in water areas, and a three-dimensional point cloud processing algorithm is used in forest areas and mountainous areas; The processing priority of each calibrated terrain area is determined according to the historical regional change rate of the calibrated terrain area; among them, the calibrated terrain area with the largest regional change law has the highest processing priority.

7. The remote sensing information processing method for demarcating terrain according to claim 1, characterized in that: Acquire real-time remote sensing information for each calibrated terrain area and call the corresponding adaptive processing algorithm; process the remote sensing information using the adaptive processing algorithm based on the processing priority of the calibrated terrain area to obtain terrain change information, specifically including: Obtaining the terrain type of the current calibrated terrain area, and extracting the corresponding adaptive processing algorithm from the remote sensing information processing algorithm library according to the terrain type; Calculate the historical area change rate of the current calibrated terrain area, compare and sort it with other calibrated terrain areas in the target area, and determine the processing priority of the current calibrated terrain area; After the processing of the calibrated terrain area of the previous processing priority is completed, the real-time remote sensing information of the current calibrated terrain area is processed by the adaptive processing algorithm to obtain the terrain change information.

8. The remote sensing information processing method for terrain calibration according to claim 1, characterized in that: A dynamic terrain inversion model is constructed based on the LSTM network, specifically including: A three-layer LSTM network structure is used, and residual connections are added between the second and third LSTM network structures to construct a stacked LSTM network. The three-layer LSTM network structure is used to capture short-term temporal dependencies, medium-term temporal dependencies, and long-term temporal dependencies respectively. Construct an attention enhancement module and connect it to the output of the stacked LSTM network to dynamically adjust the importance weight of each time step and the contribution weight of the input features; The output end of the attention enhancement module is connected to the output layer to complete the construction of the dynamic terrain inversion model, and the model is trained using the pre-built NASA GEDI laser elevation dataset.

9. The remote sensing information processing method for demarcating terrain according to claim 1, characterized in that: Cross-validating the terrain change information with the terrain inversion change information to obtain accurate terrain change information, specifically including: Extracting a first change feature set from the terrain change information, and extracting a second change feature set from the terrain inversion change information; Calculating feature consistency between each change feature in the first change feature set and the corresponding change feature in the second change feature set; Determine the changed features whose feature consistency is lower than a preset threshold as abnormal features, send an early warning and carry the location information of the abnormal features to the maintenance terminal; After receiving the confirmation features returned by the maintenance terminal, they are combined with the non-abnormal features to form the precise terrain change information.

10. A remote sensing information processing system for calibrating terrain, characterized in that: The system comprises: The terrain calibration module is used to extract the geometric features of fixed markers and the temporal features of dynamic markers based on the historical remote sensing information of the target area; construct a multimodal feature fusion model based on a two-stream neural network, and fuse the geometric features with the temporal features to obtain fused features; construct a dynamic terrain partitioning model, and partition the target area into different types of calibrated terrain areas based on the fused features; A remote sensing information processing module is used to obtain real-time remote sensing information of each calibrated terrain area and call the corresponding adaptive processing algorithm; based on the processing priority of the calibrated terrain area, the remote sensing information is processed by the adaptive processing algorithm to obtain terrain change information; A terrain inversion module is used to build a dynamic terrain inversion model based on an LSTM network, and input the real-time remote sensing information and historical remote sensing information into the dynamic terrain inversion model to obtain terrain inversion change information; The cross-validation module is used to cross-validate the terrain change information with the terrain inversion change information to obtain accurate terrain change information.

Images