Multi-source heterogeneous terrain data fusion method and electronic equipment

Abstract

This invention provides a method and electronic device for fusing multi-source heterogeneous terrain data, relating to the field of heterogeneous terrain data fusion technology. The method includes: collecting multi-source terrain data for each data point in a target area; determining the terrain spatial attribute data of void areas, inputting the terrain spatial attribute data into a pre-trained calibration model to obtain elevation difference data for the void areas; adding the global digital elevation data (GDMA) of the void areas to their corresponding elevation difference data to determine calibrated GDMA data; using the calibrated GDMA data to complete the terrain data of the void areas; and fusing the calibrated GDMA data, the completed terrain data, and the surface elevation data to obtain the fused terrain data for the target area. This invention can reduce errors and missing data in anomalous areas, thereby significantly improving the calibration accuracy and reliability of terrain data in anomalous areas.

Description

Technical Field

[0001] This invention relates to the field of heterogeneous terrain data fusion technology, and in particular to a method and electronic device for fusion of multi-source heterogeneous terrain data. Background Technology

[0002] Traditional topographic data acquisition primarily relies on aerial remote sensing imagery and ground surveying, which suffers from problems such as limited data sources, limited accuracy, and lagging update frequency, making it difficult to meet the high-quality demands of modern surveying and mapping. Multi-source topographic data encompasses various types, including satellite imagery, lidar, and UAV imagery, providing diverse pathways for topographic data acquisition. Through multi-source data fusion technology, data accuracy can be effectively improved, data integrity enhanced, and large-scale topographic data processing automated, increasing processing efficiency. However, current multi-source topographic data fusion technology still faces many technical challenges.

[0003] Before fusing multi-source terrain data, outliers need to be corrected. Existing outlier correction techniques typically rely on sparse surface elevation data combined with interpolation algorithms. Due to the sparse distribution of sampling points, areas far from the sampling points are prone to large errors, leading to inaccurate outlier correction.

[0004] The multi-source terrain data fusion results obtained under such circumstances will not only reduce the overall data accuracy due to the residue of outliers, but will also further affect the effectiveness of subsequent processing steps such as noise removal and hole filling, forming a chain of errors. Summary of the Invention

[0005] This invention provides a method and electronic device for fusing multi-source heterogeneous terrain data to solve the problem of inaccurate calibration of abnormal data in the current stage of terrain data fusion.

[0006] In a first aspect, embodiments of the present invention provide a method for fusing multi-source heterogeneous terrain data, including: Collect multi-source terrain data for each data point in the target area; the multi-source terrain data includes SRTM1 data, global digital elevation data, and surface elevation data; The topographic spatial attribute data of the cavity region in the SRTM1 data is determined, and the topographic spatial attribute data is input into the pre-trained calibration model to obtain the elevation difference data of the cavity region; the elevation difference data is used to characterize the elevation difference between the surface elevation data of the cavity region and the global digital elevation data. The global digital elevation data under the cavity area is added to its corresponding elevation difference data to determine the calibrated global digital elevation data; and the calibrated global digital elevation data is used to complete the SRTM1 data of the cavity area. The calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data are fused to obtain the fused terrain data of the target area.

[0007] In one possible implementation, the pre-trained calibration model is obtained in the following way: Historical multi-source topographic data of target data points in the sample area were collected; among which, historical multi-source topographic data included historical global digital elevation data and historical surface elevation data; Calculate the elevation difference between historical surface elevation data and historical global digital elevation data for each data point of the target; The historical terrain spatial attribute data of each target data point is used as input, and the corresponding elevation difference is used as output to train the calibration model to obtain a pre-trained calibration model. Among them, historical topographic spatial attribute data are determined based on historical global digital elevation data.

[0008] In one possible implementation, calibrated global digital elevation data, completed SRTM1 data, and surface elevation data are fused to obtain fused terrain data for the target area, including: The calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data are fused to obtain initial fused terrain data. Anomalies are then identified in the initial fused terrain data. Determine the type of abnormal region, and based on the type, determine the corresponding filtering algorithm for each abnormal region; The initial fused terrain data under each abnormal region is filtered using the filtering algorithm corresponding to each abnormal region in order to obtain the fused terrain data of the target region.

[0009] In one possible implementation, the anomalous regions include complex terrain regions, elevation anomaly regions, and gentle terrain regions; the type of the anomalous region is determined, and based on the type, a corresponding filtering algorithm is determined for each anomalous region, including: If the abnormal region is a complex terrain region, then the low-order surface fitting algorithm will be used as the filtering algorithm corresponding to the abnormal region. If the abnormal area is an elevation abnormal area, then the median filtering algorithm will be used as the filtering algorithm corresponding to the abnormal area. If the abnormal area is a flat terrain area, then the mean filtering algorithm will be used as the filtering algorithm for the abnormal area. Among them, complex terrain areas are used to characterize areas where the spatial distribution of elevation is not uniform and there are changes in terrain structure; Elevation anomaly regions are used to characterize areas where erroneous elevation data exists; Gentle terrain regions are used to characterize areas with uniform spatial distribution of elevation and no local abrupt changes.

[0010] In one possible implementation, after fusing the calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data to obtain the fused terrain data for the target area, the method further includes: Based on the fused terrain data of the target area, a terrain simulation model of the target area is constructed.

[0011] In one possible implementation, a terrain simulation model of the target area is constructed based on the fused terrain data of the target area, including: Vulnerability detection is performed on the fused terrain data of the target area to identify vulnerable areas; The vulnerability area is filled in to obtain complete terrain data of the target area; Based on complete terrain data, construct a terrain simulation model of the target area.

[0012] In one possible implementation, the vulnerable area is filled in to obtain complete terrain data of the target area, including: Extract point cloud data within a pre-defined area, including the vulnerability region, from the stereo mapping results; Based on the point cloud data, determine the initial elevation point cloud data of the vulnerability area; The initial elevation point cloud data is processed, and the processed initial elevation point cloud data is used to fill in the gaps in the target area to obtain complete terrain data of the target area.

[0013] In one possible implementation, the initial elevation point cloud data of the vulnerability area is determined based on the point cloud data, including: Based on the point cloud data, generate ground point cloud data for the vulnerability area; The ground point cloud data is filtered to obtain the initial elevation point cloud data of the vulnerability area.

[0014] In one possible implementation, the initial elevation point cloud data is processed, and the processed initial elevation point cloud data is used to complete the vulnerability area, resulting in complete terrain data of the target area, including: Geometric processing is performed on the initial elevation point cloud data to obtain the target elevation point cloud data; The target elevation point cloud data is repaired, and the repaired target elevation point cloud data is used to fill in the vulnerability area to obtain the complete terrain data of the target area.

[0015] In a second aspect, embodiments of the present invention provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect or any possible implementation thereof.

[0016] In this embodiment of the invention, a calibration model trained based on historical multi-source terrain data is used to predict errors in void areas of SRTM1 data. This achieves accurate calibration of global digital elevation data (GEM) while simultaneously filling in void areas in SRTM1 data. This method eliminates systematic elevation deviations between different data sources, reduces step effects and data discontinuities during the fusion process, and improves the consistency and matching degree of multi-source terrain data. This method differs from traditional direct stitching and simple interpolation methods that rely solely on spatial proximity fitting. It fully utilizes the error distribution patterns and terrain correlation characteristics inherent in historical multi-source terrain data to specifically compensate for systematic deviations between void areas in SRTM1 and GEM. Therefore, it can more accurately restore the true terrain elevation, reduce error amplification and distortion in anomalous areas, and significantly improve the calibration accuracy and reliability of terrain data in anomalous areas. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the implementation of the multi-source heterogeneous terrain data fusion method provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of the multi-source heterogeneous terrain data fusion device provided in the embodiment of the present invention. Detailed Implementation

[0018] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0019] Figure 1 This is a flowchart illustrating the implementation of the multi-source heterogeneous terrain data fusion method provided in this embodiment of the invention. Figure 1 As shown, the method includes: Step 110: Collect multi-source terrain data for each data point in the target area; the multi-source terrain data includes SRTM1 data, global digital elevation data, and surface elevation data.

[0020] First, determine the geographical extent of the target area and obtain its latitude and longitude coordinates, projected coordinate system, and coverage boundaries. Then, transform the target area using the projected coordinate system and rasterize it into a regular grid. The geospatial feature point corresponding to the smallest pixel or raster unit in the resulting regular grid is used as the data point.

[0021] For each data point, based on its latitude and longitude coordinates, global digital elevation data (Shuttle Radar Topography Mission 1 Arc-Second Global, SRTM1) is obtained through satellite radar; global digital elevation data is obtained through the Global Digital Elevation Model (GDEM); and surface elevation data is obtained through the Surface Elevation Data (SED).

[0022] Among them, SRTM1 data covers most of the global land area, but contains voids; Global Digital Elevation (GDE) data is complete across the entire region, but has moderate accuracy and contains systematic errors; Surface Elevation data has the highest accuracy and most closely approximates the actual terrain, but its coverage is limited and cannot cover the entire region. Compared with SRTM1 and GDE data, it has limitations such as high data density, excessively high resolution, and scale mismatch. Therefore, not every data point corresponds to all three types of data, but all data points include GDE data.

[0023] The three types of data acquired are preprocessed separately. Preprocessing methods may include format conversion, coordinate system transformation, and data cleaning to ensure consistency in spatial location and coverage across all data. Specifically: Convert these three types of data to the same format, such as GeoTIFF, ERDAS IMG, or ASCIIGrid.

[0024] Then, through projection transformation, these three types of data are transformed into a coordinate system, that is, the original projection system of these data is transformed into a unified projection system to ensure that all data are aligned in geographic space.

[0025] The data cleaning process includes outlier detection and outlier filtering. Outliers are caused by equipment malfunctions, cloud cover, or terrain complexity. For these three types of data, statistical methods such as Z-scores and terrain rules such as slope thresholds can be used for outlier detection. For the detected outliers, filtering conditions can be set based on statistical results or terrain rules to filter them out.

[0026] Step 120: Determine the topographic spatial attribute data of the cavity area in the SRTM1 data, and input the topographic spatial attribute data into the pre-trained calibration model to obtain the elevation difference data of the cavity area; wherein, the elevation difference data is used to characterize the elevation difference between the surface elevation data of the cavity area and the global digital elevation data.

[0027] Step 130: Add the global digital elevation data under the cavity area to its corresponding elevation difference data to determine the calibrated global digital elevation data; and use the calibrated global digital elevation data to complete the SRTM1 data of the cavity area.

[0028] In this embodiment, the terrain spatial attribute data includes the latitude and longitude coordinates of each data point, and the terrain slope information of each data point; wherein, the terrain slope information of each data point is determined in the following way: For any given data point, determine the adjacent data points within a preset range based on the latitude and longitude coordinates of that data point; the preset range is calculated based on the actual accuracy requirements.

[0029] Calculate the terrain slope value for each adjacent data point based on the global digital elevation data of the adjacent data points.

[0030] The average of the terrain slope values ​​of all adjacent data points is taken as the terrain slope information corresponding to that data point.

[0031] Considering the discrepancy between surface elevation data and global digital elevation data (DEM), and that surface elevation data is more accurate and precise, but its coverage is limited and cannot cover the entire region, this embodiment of the invention uses terrain slope information as a basis to determine the ideal elevation difference data for each data point. The DEM data for the void area is then added to its corresponding elevation difference data to determine the calibrated DEM data. Simultaneously, the calibrated DEM data is used to complete the SRTM1 data for the void area.

[0032] In other words, this embodiment only calibrates the global digital elevation data (GEM) for cavity areas. This is because non-cavity areas already have SRTM1 data as a valid elevation reference, requiring no additional calibration. However, cavity areas lack valid SRTM1 data and must rely on GEM data for completion, thus requiring high-precision calibration to ensure accurate completion results. For non-cavity areas, which already have SRTM1 and surface elevation data, the accuracy and reliability of which are superior to uncalibrated GEM data, can be directly used as reliable terrain data; therefore, calibration of the GEM data for these areas is unnecessary.

[0033] In an optional embodiment, the void region is determined in the following way: Read SRTM1 data, iterate through the elevation values ​​of each raster cell, and identify the data identifier corresponding to each elevation value. Extract the data identifiers without data and mark all raster locations with invalid values ​​to form hole areas.

[0034] Step 140: Fuse the calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data to obtain the fused terrain data of the target area.

[0035] For these three types of data after correction and completion, spatial weighted fusion method, regional replacement fusion method, or kriging / inverse distance weighted interpolation fusion method can be used to achieve fusion to obtain the fused terrain data of the target area.

[0036] Among them, spatial weighted fusion allocates weights according to data accuracy. Specifically, surface elevation data has the highest weight, followed by calibrated global digital elevation data, and SRTM1 data has the lowest weight. The fused elevation is obtained by weighting by location.

[0037] In the regional replacement fusion, areas with available surface elevation data directly use the surface elevation data; areas without surface elevation data but with SRTM1 data use the SRTM1 data; and areas with cavities use calibrated global digital elevation data.

[0038] Kriging / inverse distance weighted interpolation fusion uses high-precision surface elevation data as control points to perform spatial interpolation across the entire region, smoothing boundaries and generating continuous fused terrain.

[0039] In summary, the method provided by this invention, based on a corresponding elevation difference under terrain slope data, determines the elevation difference of each data point in the cavity area using Global Digital Elevation (GDE) data. The sum of the GDE data and the elevation difference for each data point is then used as the calibrated data. This process calibrates the GDE data and fills in the cavity area, improving the consistency and matching degree of multi-source terrain data. This method differs from traditional methods that rely solely on spatial proximity fitting through direct stitching and simple interpolation. It fully utilizes the error distribution patterns and terrain correlation characteristics inherent in historical multi-source terrain data to specifically compensate for the systematic deviation between the cavity area in SRTM1 and the GDE data. Therefore, it can more accurately restore the true terrain elevation, reduce error amplification and distortion in anomalous areas, and significantly improve the calibration accuracy and reliability of terrain data in anomalous areas.

[0040] The method is illustrated below through some optional embodiments: In an optional embodiment, the pre-trained calibration model is obtained as follows: Historical multi-source topographic data of target data points in the sample area were collected; among which, historical multi-source topographic data included historical global digital elevation data and historical surface elevation data.

[0041] Calculate the elevation difference between historical surface elevation data and historical global digital elevation data for each data point of the target.

[0042] The historical terrain spatial attribute data of each target data point is used as input, and the corresponding elevation difference is used as output to train the calibration model to obtain a pre-trained calibration model.

[0043] Among them, historical topographic spatial attribute data are determined based on historical global digital elevation data.

[0044] In this embodiment, firstly, historical multi-source terrain data of target data points in the sample area is collected. The sample area may include hollow areas and non-hollow areas. Correspondingly, the historical multi-source terrain data may also include historical SRTM1 data to identify hollow and non-hollow areas.

[0045] Due to the characteristics of these three types of data, not all data points correspond to all three types. This embodiment uses points that simultaneously include historical global digital elevation data and historical surface elevation data as target data points.

[0046] The collected historical multi-source terrain data is preprocessed. The collection and processing process can refer to the specific implementation method in step 110.

[0047] For each target data point in the sample area, the elevation difference between historical surface elevation data and historical global digital elevation data is calculated. Simultaneously, for each target data point, its corresponding historical topographic spatial attribute data is calculated based on the historical global digital elevation data. For hollow areas, the average value can be used to determine the corresponding historical topographic spatial attribute data; for non-hollow areas, the topographic slope value calculated from the historical global digital elevation data can be directly used as the historical topographic spatial attribute data.

[0048] Using a neural network model as the architecture, the historical terrain spatial attribute data of each target data point is taken as input and the corresponding elevation difference is taken as output for model training. During the training process, the prediction error is minimized by adjusting the model's weights and biases until the model meets the requirements, thus obtaining a pre-trained calibration model.

[0049] In an optional embodiment, step 140 involves fusing the calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data to obtain fused terrain data for the target area. This may include: Step 141: Fuse the calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data to obtain initial fused terrain data, and identify abnormal areas in the initial fused terrain data.

[0050] Step 142: Determine the type of abnormal region, and based on the type, determine the corresponding filtering algorithm for each abnormal region.

[0051] Step 143: Use the filtering algorithm corresponding to each abnormal region to filter the initial fused terrain data under each abnormal region to obtain the fused terrain data of the target region.

[0052] In the process of fusing multi-source heterogeneous terrain data, due to differences in data sources and terrain complexity, systematic errors and outliers can occur. Direct data stitching can result in obvious step anomalies or contour distortions. Therefore, selecting an appropriate fusion method is crucial. This involves both mosaicking multi-source data and eliminating systematic errors and elevation anomalies to smooth the terrain surface. Specifically: For the initially fused terrain data, neighborhood elevation difference statistics are performed to calculate the elevation difference between each data point and its surrounding points; areas where the elevation difference significantly exceeds the normal range of terrain variation are marked as abnormal areas. The normal range of terrain variation is obtained based on historical data statistics.

[0053] This method can quickly identify elevation abrupt changes, spikes, and sudden jumps caused by multi-source data splicing and accuracy differences, and locate areas that need optimization.

[0054] For different types of anomaly areas, targeted filtering methods are selected to avoid terrain distortion caused by a one-size-fits-all approach and to ensure filtering accuracy. Based on the anomaly type, the selected filtering algorithm is applied to the corresponding area. During the filtering process, elevation values ​​can be smoothed using 3×3 or 5×5 neighborhood windows. After filtering, the normal terrain trend is preserved, resulting in the final fused terrain data. This eliminates splicing noise and outliers, making the terrain transition natural, continuous, and smooth, thus improving the final data quality.

[0055] In an optional embodiment, the anomalous regions include complex terrain regions, elevation anomaly regions, and gentle terrain regions; step 142, determining the type of the anomalous region and, based on the type, determining the corresponding filtering algorithm for each anomalous region, may include: If the abnormal region is a complex terrain region, then the low-order surface fitting algorithm will be used as the filtering algorithm corresponding to the abnormal region.

[0056] If the abnormal region is an elevation abnormal region, then the median filtering algorithm will be used as the filtering algorithm corresponding to the abnormal region.

[0057] If the abnormal area is a flat terrain area, then the mean filtering algorithm will be used as the filtering algorithm for that abnormal area.

[0058] Among them, complex terrain regions are used to characterize areas where the spatial distribution of elevation is not uniform and there are changes in terrain structure.

[0059] Elevation anomaly regions are used to characterize areas where erroneous elevation data exists.

[0060] Gentle terrain regions are used to characterize areas with uniform spatial distribution of elevation and no local abrupt changes.

[0061] Based on the terrain undulation characteristics and elevation data quality, this embodiment divides the abnormal areas into three types: complex terrain areas, elevation anomaly areas, and gentle terrain areas.

[0062] Complex terrain areas feature significant topographic relief, rapid slope changes, ridges, and valleys, resulting in uneven elevation distribution. Areas with elevation anomalies contain erroneous data, exhibiting abrupt highs, lows, jumps, and spikes. Flat terrain areas, on the other hand, have minimal elevation changes, a uniform elevation distribution, and no significant abrupt changes.

[0063] When determining complex terrain, the standard deviation of local slope and the variance of elevation are calculated. If they exceed the corresponding thresholds, the terrain is classified as complex. These thresholds are determined based on historical data.

[0064] Anomaly identification is performed by calculating the elevation difference between the current point and the average elevation of its neighbors. If the difference exceeds a preset elevation range, the area is identified as an anomaly.

[0065] The identification of flat terrain areas also calculates the local slope standard deviation and elevation variance. If the calculated local slope standard deviation and elevation variance are both less than the corresponding threshold, it is determined to be flat terrain.

[0066] Low-order surface fitting can fit the overall terrain trend, while preserving the real terrain structure such as ridges and valleys, smoothing local noise, and avoiding the loss of terrain features due to over-filtering. Therefore, for complex terrain areas, low-order surface fitting algorithms are selected for filtering.

[0067] Median filtering is effective in removing impulse noise and outliers, and can effectively eliminate erroneous elevation data without excessively blurring normal terrain boundaries. Therefore, median filtering is chosen for areas with elevation anomalies.

[0068] Mean filtering can smoothly eliminate small fluctuations, making the terrain smoother and more continuous. It is simple to calculate and does not destroy the overall consistency of flat areas. Therefore, the mean filtering algorithm is chosen for flat terrain areas.

[0069] Alternatively, the principle of low-order surface fitting is to smooth data points by fitting a low-order surface within a local region of the data, i.e., a sliding window, thereby preserving the overall trend of the data while reducing the impact of noise.

[0070] When using this algorithm, a sliding window is first selected, which can be a rectangular area of ​​size m×n. Within the window, a surface model is fitted based on the coordinates and values ​​of the data points. The data point at the center of the window is then replaced with the value of the center point of the fitted surface. The window is then slid to the next position, and the above process is repeated until all data has been processed.

[0071] The expression for the surface model can be:

[0072] In the formula, to These are the fitting coefficients. , Represents the coordinates of the center point.

[0073] In this process, the objective function can be to minimize the sum of squared errors, i.e.:

[0074] In the formula, Let i be the coordinates of data point i.

[0075] The fitting coefficients can be obtained using the least squares method, which can be determined by constructing a system of linear equations, i.e.:

[0076]

[0077] Where A is the design matrix, which contains The polynomial terms. 'b' represents the fitting coefficients to be determined. 'b' represents the observed value. The vector.

[0078] After solving, the smoothing value at the center of the window is obtained from Confirmed, among which These are the coordinates of the window center.

[0079] Besides complex terrain areas, this algorithm is suitable for the following scenarios: Suitable for smoothing two-dimensional or three-dimensional data, such as terrain modeling, image denoising, and sensor data processing. Ideal for data with a certain degree of continuity and local trends. Also suitable for continuous data and spatial data, such as image pixels and elevation data.

[0080] This algorithm can preserve the local trends and geometric properties of the data, making it suitable for smoothing complex surface data. It also offers high flexibility, allowing the smoothing degree to be controlled by adjusting the window size and the order of the fitted surface.

[0081] The basic principle of the median filtering algorithm is to replace the original value at the center of the window with the median of the data within the sliding window, thereby effectively removing impulse noise, such as salt-and-pepper noise, while preserving edge information.

[0082] The specific calculation steps of the algorithm are as follows: First, select a sliding window, such as 3×3 or 5×5. Collect the values ​​of all data points within the sliding window. Sort these values ​​and select the median value as the smoothed value of the window center point. Slide the window and repeat the above process.

[0083] For a one-dimensional signal, assume the window size is... The data in the window is The median filter output is:

[0084] In the formula, Indicates signal value The median.

[0085] For two-dimensional images, the window size can be set to... The output is the median of the pixel values ​​within the window, i.e.:

[0086] In the formula, These The median pixel value.

[0087] In addition to areas with elevation anomalies, this algorithm is also applicable to the following scenarios: removing salt-and-pepper noise in image processing; eliminating spike noise or outliers in signal processing; and scenarios where edge information needs to be preserved.

[0088] This algorithm is suitable for processing discrete signals and image data.

[0089] This algorithm is highly robust to impulse noise and outliers. It effectively preserves edge information of images or signals, avoiding blurring. Furthermore, the algorithm is simple and easy to implement.

[0090] The basic principle of the mean filtering algorithm is to replace the original value at the center of the window with the average value of the data within the sliding window to reduce random noise. Mean filtering assumes that the noise is zero-mean random noise, and the original signal can be approximately restored after smoothing.

[0091] The specific steps of this algorithm are as follows: Select a sliding window, for example, 3×3. Calculate the average of all data within the window. Replace the data point in the center of the window with the average. Slide the window and repeat the above process.

[0092] For a one-dimensional signal, the window size is... The mean filter output is:

[0093] In the formula, i and j represent the position of the signal, and x j Let y represent the j-th signal value. i This represents the average value of all signal values ​​within a 2k+1 window.

[0094] For a two-dimensional image, with a window size of m×n or n×m, the output is:

[0095] In the formula, m represents the number of rows in the 2D window, n represents the number of columns in the 2D window, i and j represent the pixel positions, and x represents the pixel position. (i,j) Let y represent the pixel value at point (i,j). （i，j） This represents the average value of all pixels within an m×n window.

[0096] In addition to flat terrain areas, this algorithm is also suitable for the following scenarios: removing Gaussian noise in image processing; smoothing random noise in signal processing; and scenarios requiring fast smoothing.

[0097] This algorithm is simple, fast, and suitable for real-time processing. It also exhibits good suppression of Gaussian noise.

[0098] In an optional embodiment, after fusing the calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data in step 140 to obtain the fused terrain data of the target area, the method further includes: Step 150: Construct a terrain simulation model of the target area based on the fused terrain data of the target area.

[0099] The fused terrain data can be used to construct a digital elevation model (DEM) and serve as a terrain simulation model for the target area.

[0100] Alternatively, an irregular triangular network (TIN) model can be constructed and used as a terrain simulation model for the target area.

[0101] Other types of terrain simulation models can be constructed according to actual needs, and will not be elaborated here.

[0102] In an optional embodiment, step 150, which involves constructing a terrain simulation model of the target area based on the fused terrain data of the target area, may include: Step 151: Perform vulnerability detection on the merged terrain data of the target area to identify vulnerable areas.

[0103] Step 152: Complete the vulnerability area to obtain the complete terrain data of the target area.

[0104] Step 153: Construct a terrain simulation model of the target area based on complete terrain data.

[0105] Considering that there may be vulnerable areas in the merged data, vulnerability detection is required for the merged data in order to ensure the accuracy of the terrain simulation model.

[0106] Vulnerable regions refer to areas that still exhibit defects after fusion and filtering, such as missing local elevations, abnormal depressions / protrusions, data discontinuities, and small areas of null values. The formation of vulnerable regions is mainly due to the following reasons: When stitching together heterogeneous data from multiple sources, the boundary coordinates are not perfectly aligned, resulting in tiny null values ​​or breaks; isolated noise and bad pixels exist in the original data being fused, and local anomalies still remain after filtering; during raster resampling and format conversion, individual pixel values ​​are lost or distorted.

[0107] Vulnerable regions differ from void regions. Vulnerable regions are small-scale defects remaining after multi-source data fusion and filtering. They are local flaws introduced after data processing, with a small range and scattered distribution. Vulnerable regions, on the other hand, specifically refer to large areas of no values ​​in the original SRTM1 data. They are missing from the data source itself, with a larger range and fixed location.

[0108] During vulnerability area detection, the merged terrain grid is traversed, and small areas containing invalid values, as well as isolated anomalies and small blocks where the difference between the elevation value and the average value of the surrounding neighborhood exceeds a set error threshold, are marked as vulnerability areas.

[0109] For vulnerable areas, neighborhood interpolation can be used to complete the vulnerable areas and obtain complete terrain data of the target area, ensuring the accuracy of the data used to construct the terrain simulation model of the target area.

[0110] In an optional embodiment, step 152, which involves completing the vulnerable area to obtain complete terrain data of the target area, may include: Point cloud data, including the vulnerability area, is extracted from the stereo mapping results.

[0111] Based on the point cloud data, determine the initial elevation point cloud data of the vulnerability area.

[0112] The initial elevation point cloud data is processed, and the processed initial elevation point cloud data is used to fill in the gaps in the target area to obtain complete terrain data of the target area.

[0113] The initial elevation point cloud data used to determine the vulnerability area, based on point cloud data, may include: Based on the point cloud data, generate ground point cloud data for the vulnerable area.

[0114] The ground point cloud data is filtered to obtain the initial elevation point cloud data of the vulnerability area.

[0115] The process involves processing the initial elevation point cloud data and using the processed data to fill in the gaps in the target area, resulting in complete terrain data for the target region, including: Geometric processing is performed on the initial elevation point cloud data to obtain the target elevation point cloud data.

[0116] The target elevation point cloud data is repaired, and the repaired target elevation point cloud data is used to fill in the vulnerability area to obtain the complete terrain data of the target area.

[0117] In this embodiment, for a vulnerable area with data vulnerabilities, a preset area corresponding to the vulnerable area is determined. This preset area includes the vulnerable area and a buffer zone surrounding it. The buffer zone is designed to ensure that the point cloud data covers the vulnerable area and connects to the surrounding terrain. The buffer zone can be set to extend 30-50 meters outward from the edge of the vulnerable area.

[0118] Based on the geographic coordinates and elevation information from the stereo mapping results, three-dimensional point cloud data within a preset area are extracted in batches using point cloud extraction tools. The extracted point cloud data undergoes preliminary screening to remove obvious extreme anomalies, yielding ground point cloud data for the vulnerability area. The stereo mapping results can be aerial triangulation results.

[0119] The initial screening process can be as follows: Noise removal and missing value filling improve the accuracy and efficiency of subsequent processing. Missing value filling requires using stereo mapping results within a preset area as control data and stereo images within that area as a reference. Connection points are obtained through multi-view registration of the missing image and the stereo image. The positioning parameters of the missing satellite image are optimized using joint adjustment calculations of the stereo mapping results and the missing satellite image. Rational Polynomial Coefficients (RPC) parameters are fitted to eliminate systematic errors in the RPC parameters. Then, using aerial triangulation results and the missing stereo satellite image as basic data, operations such as generating epipolar alignment images, estimating disparity range, and performing disparity map transformation and rasterization are performed to match and generate ground point cloud data for the missing area, i.e., Digital Surface Model (DSM) point cloud data.

[0120] The obtained point cloud data is filtered to remove non-ground points, resulting in the initial elevation point cloud data of the vulnerability area, also known as the initial Digital Elevation Model (DEM) point cloud data. This process can be summarized as follows: 1) Initialization of filter parameters Window size setting: The window size determines the effective range of the filter. The initial window size is usually set based on the resolution of the point cloud data and the terrain features.

[0121] Elevation difference threshold setting: The elevation difference threshold is used to distinguish between ground points and non-ground points, and is determined based on the terrain slope and the height variation of the study area.

[0122] 2) Filtering Erosion operation: Simplifies the shape of an object by removing unnecessary parts, which helps to remove noise and small non-ground objects in point clouds.

[0123] Inflation operation: Increases the volume of an object based on its original size, which helps to fill in some voids and make the details fuller.

[0124] Opening and closing operations: Opening is an operation that erodes first and then dilates, which helps to eliminate small objects and smooth the boundaries of larger objects; closing is an operation that dilates first and then erodes, which helps to fill small holes inside objects and smooth the boundaries.

[0125] 3) Separation of ground points and non-ground points After filtering, the point cloud data is divided into ground points and non-ground points by comparing the elevation value of each point with the filtered elevation surface. This step is crucial for generating the initial elevation point cloud data, because only ground points can be used to build the elevation model.

[0126] 4) Filter parameter update and iteration Based on the results of the previous filtering, the filtering parameters, such as window size and elevation difference threshold, are gradually adjusted to adapt to terrain changes and the size of non-ground objects.

[0127] Iterative morphological filtering is performed until the filter window size exceeds the preset maximum threshold or a satisfactory filtering effect is achieved.

[0128] After filtering, all non-ground points in the DSM point cloud data are removed, resulting in all ground point cloud data within the DSM point cloud, which constitutes the initial elevation point cloud data for the vulnerability area. However, the filtering process also introduces numerous holes into the initial elevation point cloud data due to the removal of non-ground point data, requiring patching as well. Since point clouds are discretized data, direct patching is difficult; geometric processing of the initial elevation point cloud data is necessary to obtain the target elevation point cloud data. This geometric processing can include point cloud mesh construction and ray intersection calculation.

[0129] Point cloud networking can include triangular mesh networking and regular grid networking.

[0130] Triangular mesh construction forms a triangular grid by connecting adjacent ground points, and is suitable for areas with complex and varied terrain.

[0131] Regular grid construction divides ground points into regular grids, with each grid cell containing one or more ground points. It is suitable for areas with flat terrain and little variation.

[0132] The mesh size during the mesh construction process is determined based on the resolution of the point cloud data and the accuracy requirements of the DEM. A smaller mesh size can improve the accuracy of the DEM, but it will increase the computational and data storage requirements; a larger mesh size may reduce the accuracy of the DEM.

[0133] Ray intersection determination determines the elevation value through ray emission, ray intersection, and elevation value assignment; specifically: Ray emission: A ray is emitted vertically from the center point of each cell in the DEM.

[0134] Ray intersection: Perform an intersection operation between the ray and the ground point cloud data after network construction, find the intersection point and obtain its elevation value, where the ground point cloud data is also the initial elevation point cloud data of the vulnerability area.

[0135] Elevation value assignment: The obtained intersection elevation value is assigned to the corresponding cell as the elevation value of that cell.

[0136] The target elevation point cloud data is obtained through the above method.

[0137] The target elevation point cloud data is repaired, and the repaired data is used to fill in the missing areas, obtaining complete terrain data for the target region. The repair process includes smoothing, hole filling, and accuracy assessment. Smoothing eliminates noise and outliers; algorithms used include low-order surface fitting, mean filtering, and median filtering. Hole filling can be performed using interpolation algorithms, including linear interpolation, bilinear interpolation, and cubic convolutional interpolation.

[0138] The accuracy of the generated DEM data is evaluated, including comparison with known elevation data and the preservation of terrain features. The evaluation results determine the accuracy and usability of the DEM data.

[0139] By using the above methods, the integrity of the fused terrain data can be guaranteed to the greatest extent, thereby improving the accuracy of the terrain simulation model.

[0140] In summary, this invention, based on a calibration model trained from historical multi-source terrain data, predicts errors in void areas within SRTM1 data. This achieves accurate calibration of global digital elevation data (GEM) while simultaneously filling in void areas in SRTM1 data. This method eliminates systematic elevation deviations between different data sources, reduces step effects and data discontinuities during fusion, and improves the consistency and matching degree of multi-source terrain data. Unlike traditional methods that rely solely on spatial proximity for fitting, this method fully utilizes the error distribution patterns and terrain correlation characteristics inherent in historical multi-source terrain data to specifically compensate for systematic deviations between void areas in SRTM1 and GEM. Therefore, it can more accurately restore true terrain elevations, reduce error amplification and distortion in anomalous areas, and significantly improve the calibration accuracy and reliability of terrain data in anomalous areas.

[0141] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0142] The following are device embodiments of the present invention. For details not described in detail, please refer to the corresponding method embodiments described above.

[0143] Figure 2 A schematic diagram of the multi-source heterogeneous terrain data fusion device provided in an embodiment of the present invention is shown. For ease of explanation, only the parts related to the embodiment of the present invention are shown, and are described in detail below: like Figure 2 As shown, the multi-source heterogeneous terrain data fusion device 2 includes: The acquisition module 21 is used to acquire multi-source terrain data for each data point in the target area; the multi-source terrain data includes SRTM1 data, global digital elevation data and surface elevation data. The determination module 22 is used to determine the topographic spatial attribute data of the cavity area in the SRTM1 data, and input the topographic spatial attribute data into the pre-trained calibration model to obtain the elevation difference data of the cavity area; wherein, the elevation difference data is used to characterize the elevation difference between the surface elevation data of the cavity area and the global digital elevation data. The calibration module 23 is used to add the global digital elevation data under the cavity area to its corresponding elevation difference data to determine the calibrated global digital elevation data; and to use the calibrated global digital elevation data to complete the SRTM1 data of the cavity area. The fusion module 24 is used to fuse the calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data to obtain the fused terrain data of the target area.

[0144] This invention also provides an electronic device, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the method described in the above method embodiments.

[0145] In the above embodiments, the descriptions of each embodiment have their own emphasis. Parts not detailed or described in a particular embodiment can be referred to in the relevant descriptions of other embodiments. Unless otherwise specified or in conflict with logic, the terminology and / or descriptions between different embodiments are consistent and can be referenced interchangeably. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0146] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A multi-source heterogeneous terrain data fusion method, characterized in that, include: Collect multi-source terrain data for each data point in the target area; wherein, the multi-source terrain data includes SRTM1 data, global digital elevation data, and surface elevation data; The topographic spatial attribute data of the cavity region in the SRTM1 data is determined, and the topographic spatial attribute data is input into a pre-trained calibration model to obtain the elevation difference data of the cavity region; wherein, the elevation difference data is used to characterize the elevation difference between the surface elevation data of the cavity region and the global digital elevation data; The global digital elevation data under the cavity area is added to its corresponding elevation difference data to determine the calibrated global digital elevation data; and the calibrated global digital elevation data is used to complete the SRTM1 data of the cavity area. The calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data are fused to obtain the fused terrain data of the target area.

2. The multi-source heterogeneous terrain data fusion method according to claim 1, characterized in that, The pre-trained calibration model is obtained in the following way: Historical multi-source terrain data of target data points in the sample area are collected; wherein, the historical multi-source terrain data includes historical global digital elevation data and historical surface elevation data; Calculate the elevation difference between the historical surface elevation data and the historical global digital elevation data for each target data point; The historical terrain spatial attribute data of each target data point is used as input, and the corresponding elevation difference is used as output to train the calibration model to obtain a pre-trained calibration model. The historical terrain spatial attribute data is determined based on the historical global digital elevation data.

3. The multi-source heterogeneous terrain data fusion method according to claim 1, characterized in that, The process of fusing the calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data to obtain the fused terrain data for the target area includes: The calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data are fused to obtain initial fused terrain data, and the initial fused terrain data is then identified to determine abnormal areas. Determine the type of the abnormal region, and based on the type, determine the filtering algorithm corresponding to each abnormal region; The initial fused terrain data under each abnormal region is filtered using the filtering algorithm corresponding to each abnormal region in order to obtain the fused terrain data of the target region.

4. The multi-source heterogeneous terrain data fusion method according to claim 3, characterized in that, The anomalous regions include complex terrain regions, elevation anomaly regions, and gentle terrain regions; determining the type of the anomalous region and, based on the type, determining the corresponding filtering algorithm for each anomalous region includes: If the abnormal region is a complex terrain region, then the low-order surface fitting algorithm will be used as the filtering algorithm corresponding to the abnormal region. If the abnormal region is an elevation abnormal region, then the median filtering algorithm will be used as the filtering algorithm corresponding to the abnormal region. If the abnormal region is a flat terrain region, then the mean filtering algorithm will be used as the filtering algorithm corresponding to the abnormal region. The complex terrain region is used to characterize areas where the spatial distribution of elevation is not uniform and there are changes in terrain structure; The elevation anomaly region is used to characterize areas where erroneous elevation data exists. The gently sloping terrain region is used to characterize areas with uniform spatial elevation distribution and no local abrupt changes.

5. The multi-source heterogeneous terrain data fusion method according to claim 1, characterized in that, After fusing the calibrated global digital elevation data, the completed SRTM1 data, and the surface elevation data to obtain the fused terrain data for the target area, the method further includes: Based on the fused terrain data of the target area, a terrain simulation model of the target area is constructed.

6. The multi-source heterogeneous terrain data fusion method according to claim 5, characterized in that, The step of constructing a terrain simulation model of the target area based on the fused terrain data of the target area includes: Vulnerability detection is performed on the fused terrain data of the target area to identify vulnerable areas; The defective area is filled in to obtain complete terrain data of the target area; Based on the complete terrain data, a terrain simulation model of the target area is constructed.

7. The multi-source heterogeneous terrain data fusion method according to claim 6, characterized in that, The process of completing the defective area to obtain the complete terrain data of the target area includes: Extract point cloud data within a preset area, including the aforementioned vulnerability region, from the stereoscopic mapping results; Based on the point cloud data, determine the initial elevation point cloud data of the vulnerability area; The initial elevation point cloud data is processed, and the processed initial elevation point cloud data is used to fill in the gaps in the target area to obtain complete terrain data of the target area.

8. The multi-source heterogeneous terrain data fusion method according to claim 7, characterized in that, The step of determining the initial elevation point cloud data of the vulnerability area based on the point cloud data includes: Based on the point cloud data, generate ground point cloud data for the vulnerability area; The ground point cloud data is filtered to obtain the initial elevation point cloud data of the vulnerability area.

9. The multi-source heterogeneous terrain data fusion method according to claim 7, characterized in that, The process of processing the initial elevation point cloud data and using the processed initial elevation point cloud data to fill in the missing areas to obtain complete terrain data of the target area includes: Geometric processing is performed on the initial elevation point cloud data to obtain the target elevation point cloud data; The target elevation point cloud data is repaired, and the repaired target elevation point cloud data is used to fill in the defective area to obtain the complete terrain data of the target area.

10. An electronic device, characterized in that, It includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the method as described in any one of claims 1 to 9.

Images