Digital elevation model correction method and device based on random forest regression model

Abstract

The present invention relates to a method and device for correcting a digital elevation model based on a random forest regression model. The method comprises: loading a digital elevation model into an elevation correction regression model constructed based on a random forest algorithm to modify the elevation; a model training process comprising: selecting a geographic location for reference from satellite footprint terrain elevation data, selecting elevation data of an area within a first distance range from the geographic location from the digital elevation model, and comparing the data with the satellite footprint terrain elevation data to obtain an elevation difference as an elevation correction; and training the elevation correction regression model using the elevation correction and evaluation attributes constructed based on the digital elevation model. The evaluation attributes include surface cover factors, terrain feature factors, spatial distribution factors, and data source quality factors. Compared with the prior art, the method proposed in the present invention can effectively improve the elevation accuracy of the digital elevation model, with the elevation accuracy improvement ratio before and after correction being approximately 29-42%.

Description

Technical Field

[0001] The present invention relates to the technical field of satellite-borne laser altimetry correction, and in particular to a digital elevation model correction method and device based on a random forest regression model. Background Art

[0002] The global digital elevation model (DEM) is an essential component of global science research and applications, and improving its elevation accuracy has attracted widespread attention from researchers both domestically and internationally. Improving the elevation accuracy of the global DEM using satellite-borne laser altimetry, with its high ranging accuracy and wide coverage, is a key strategy for addressing this issue. Previous studies have primarily employed two approaches: 1) using the footprint elevations of satellite-borne laser altimetry as elevation control points, supplementing the primary data source to improve the accuracy of the DEM during its production; and 2) using the footprint elevations of satellite-borne laser altimetry as elevation references to directly modify the elevation of the generated DEM to improve its accuracy.

[0003] The first method involves complex data processing steps and a large amount of data, especially for global regions. Furthermore, the diverse data sources of digital elevation models necessitate a mixed approach to correction, further increasing the processing load. Unlike the first method, the second method offers flexible data processing and has been widely favored by researchers in recent years. Previous studies on the second method primarily used laser altimetry data from the ICESat satellite. Due to limitations in the satellite's detection method and resolution, these studies have focused on correcting the elevation of digital elevation models in typical regions, such as vegetation and the polar regions.

[0004] With the advancement of spaceborne laser altimetry technology, the ICESat-2 satellite utilizes single-photon detection, a new generation of spaceborne laser altimetry technology that differs from the waveform energy detection technology used by ICESat. This technology is expected to further unlock the potential of spaceborne laser altimetry in this field. However, due to the satellite's relatively recent launch, existing methods for correcting elevation in digital elevation models focus solely on the ground near-track area (30 meters). Summary of the Invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a digital elevation model correction method and device based on a random forest regression model, in order to break through the limitations of previous methods in correcting the elevation of digital elevation models in the adjacent area along the track on the ground, thereby further enhancing the application potential of satellite-borne laser altimetry technology in this field.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A digital elevation model correction method based on a random forest regression model comprises: loading a digital elevation model into a pre-built and trained elevation correction regression model, modifying the elevation in the digital elevation model, and obtaining a corrected digital elevation model;

[0008] The elevation correction regression model is constructed based on the random forest algorithm.

[0009] The training process of the elevation correction regression model includes: selecting a geographic location for reference from the satellite footprint terrain elevation data, obtaining a digital elevation model, selecting elevation data of an area within a first distance range from the geographic location from the digital elevation model, and comparing the elevation with the satellite footprint terrain elevation data to obtain an elevation difference as an elevation correction for the digital elevation model;

[0010] The elevation correction regression model constructed based on the random forest algorithm is trained using the elevation correction amount and the evaluation attributes constructed according to the digital elevation model. The evaluation attributes include surface cover factors, terrain feature factors, spatial distribution factors and data source quality factors within the distribution area of ​​the digital elevation model.

[0011] Furthermore, the surface coverage factor is to use global surface coverage data to represent the surface within the distribution area of ​​the digital elevation model.

[0012] Furthermore, the terrain characteristic factors are slope, roughness, aspect, curvature and elevation statistics of the digital elevation model within a preset first window.

[0013] Furthermore, the spatial distribution factor is the three-dimensional coordinates of each pixel point of the digital elevation model relative to the zero point coordinate calculated using the minimum value of the plane geographic coordinates of the digital elevation model in the preset first area in the X and Y directions and the minimum elevation value as the zero point coordinate reference.

[0014] Furthermore, the data source quality factor of the model is the data source quality obtained according to the quality product file attached to the digital elevation model.

[0015] Furthermore, the value of the first distance is within the range of 10-20m.

[0016] Furthermore, the modification of the elevation in the digital elevation model is specifically as follows:

[0017] Modify the elevation of the portion of the digital elevation model where the slope is less than 25 degrees.

[0018] The present invention also provides a digital elevation model correction device based on a random forest regression model, comprising:

[0019] The model building and training module is configured to build an elevation correction regression model based on the random forest algorithm;

[0020] The training process of the elevation correction regression model includes: selecting a geographic location for reference from the satellite footprint terrain elevation data, obtaining a digital elevation model, selecting elevation data of an area within a first distance range from the geographic location from the digital elevation model, and comparing the elevation with the satellite footprint terrain elevation data to obtain an elevation difference as an elevation correction for the digital elevation model;

[0021] Training an elevation correction regression model constructed based on a random forest algorithm using elevation corrections and evaluation attributes constructed based on the digital elevation model, wherein the evaluation attributes include surface cover factors, terrain characteristics factors, spatial distribution factors, and data source quality factors within the distribution area of ​​the digital elevation model;

[0022] The digital elevation model correction module loads the digital elevation model into the trained elevation correction regression model, modifies the elevation in the digital elevation model, and obtains the corrected digital elevation model.

[0023] Furthermore, the surface coverage factor is to use global surface coverage data to characterize the surface within the distribution area of ​​the digital elevation model;

[0024] The terrain characteristic factors are the slope, roughness, aspect, curvature and elevation statistics of the digital elevation model within a preset first window;

[0025] The spatial distribution factor is the three-dimensional coordinates of each pixel point of the digital elevation model relative to the zero point coordinates, calculated using the minimum values ​​of the plane geographic coordinates of the digital elevation model in the X and Y directions and the minimum elevation value of the digital elevation model in the preset first area as the zero point coordinate reference;

[0026] The data source quality factor of the model is the data source quality obtained according to the quality product file attached to the digital elevation model.

[0027] Furthermore, the value of the first distance is within the range of 10-20m;

[0028] The modification of the elevation in the digital elevation model is specifically as follows:

[0029] Modify the elevation of the portion of the digital elevation model where the slope is less than 25 degrees.

[0030] Compared with the prior art, the present invention has the following advantages:

[0031] The present invention proposes a digital elevation model correction method based on a random forest regression model, in order to overcome the limitations of previous methods in correcting the digital elevation model in the adjacent area along the ground track. The proposed method was verified by using SRTM products and the ATL08 product of the ICESat-2 satellite as digital elevation model experimental data and space-borne laser altimetry experimental data, respectively, and using FROM-GLC10 products and GFCC30TC products as auxiliary data, as well as airborne LiDAR data as elevation accuracy verification data. The experimental results show that after the method proposed by the present invention, the accuracy of the SRTM product after elevation correction is significantly improved in flat land, hilly and mountainous areas, with the accuracy improvement ratio being 29-42%. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic flow chart of a digital elevation model correction method based on a random forest regression model provided in an embodiment of the present invention;

[0033] Figure 2 This is a schematic diagram of an extraction result analysis provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0035] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention as claimed, but rather merely represents selected embodiments of the present invention. All other embodiments derived by persons of ordinary skill in the art based on the embodiments of the present invention without creative effort are intended to fall within the scope of protection of the present invention.

[0036] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, it does not require further definition or explanation in subsequent drawings.

[0037] In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, or are the orientation or position relationship in which the product of the invention is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they should not be understood as limiting the present invention.

[0038] It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, "plurality" means two or more, unless otherwise specifically defined.

[0039] In addition, the terms "horizontal" and "vertical" do not mean that the components must be absolutely horizontal or overhanging, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", not that the structure must be completely horizontal, but can be slightly tilted.

[0040] Example 1

[0041] This embodiment provides a digital elevation model correction method based on a random forest regression model, comprising: loading a digital elevation model into a pre-built and trained elevation correction regression model, modifying the elevation in the digital elevation model, and obtaining a corrected digital elevation model;

[0042] The elevation correction regression model is built based on the random forest algorithm.

[0043] The training process of the elevation correction regression model includes: selecting a geographic location for reference from the satellite footprint terrain elevation data to obtain a digital elevation model; selecting elevation data of an area within a first distance range from the geographic location from the digital elevation model; comparing the elevation data with the satellite footprint terrain elevation data; and obtaining the elevation difference as the elevation correction value of the digital elevation model. The value of the first distance is within the range of 10-20 meters.

[0044] The elevation correction regression model constructed based on the random forest algorithm is trained using the elevation correction value and the evaluation attributes constructed according to the digital elevation model. The evaluation attributes include the surface coverage factor, terrain feature factor, spatial distribution factor and data source quality factor of the digital elevation model distribution area.

[0045] As a preferred implementation, the surface cover factor is to use global surface cover data to represent the surface within the distribution area of ​​the digital elevation model.

[0046] In other words, land cover can be characterized using open-source global land cover data, such as GlobeLand30 and FROM-GLC10. In particular, the impact of vegetation cover, a key factor affecting land cover, on the elevation accuracy of digital elevation models requires further evaluation.

[0047] As a preferred embodiment, the terrain characteristic factors are the slope, roughness, aspect, curvature and elevation statistics of the digital elevation model within a preset first window.

[0048] In other words, the terrain characteristic factors can be mainly used to evaluate the degree of influence of the elevation accuracy of the digital elevation model within a specific window (such as 3×3 pixels), including slope, roughness (elevation variation coefficient), aspect, curvature, and elevation statistics (minimum elevation, maximum elevation, median elevation, mean elevation, etc.).

[0049] As a preferred embodiment, the spatial distribution factor is the three-dimensional coordinates of each pixel point of the digital elevation model relative to the zero point coordinate calculated using the minimum value of the plane geographic coordinates of the digital elevation model in the preset first area in the X and Y directions and the minimum elevation as the zero point coordinate reference.

[0050] Equivalently, the spatial distribution can use the minimum values ​​of the plane geographic coordinates in the X and Y directions and the minimum elevation value of the digital elevation model in a specific area (such as 1°×1° longitude and latitude) as the zero-point coordinate reference, and calculate the three-dimensional coordinates of each pixel point of the model relative to the zero-point coordinate as an evaluation attribute of the degree of influence of the spatial distribution on the elevation accuracy of the digital elevation model.

[0051] As a preferred implementation, the data source quality factor of the model is the data source quality obtained from the quality product file attached to the digital elevation model.

[0052] Equivalently, the quantitative source quality factor of the digital elevation model can be evaluated using the quality product file attached to the digital elevation model as an attribute.

[0053] In this embodiment, after constructing the above-mentioned evaluation attributes, the geographical location of the high-quality ICESat-2 (Ice, Cloud and land Elevation Satellite) satellite footprint terrain elevation is used as a reference, and the part closest to these geographical elevations (<15m) is selected from the digital elevation model, and the elevation of this part is compared with the ICESat-2 satellite footprint terrain elevation, and the elevation difference between the two is used as the elevation correction of the digital elevation model. In this process, high-quality ICESat-2 satellite footprint terrain elevation can be obtained by processing a method for extracting elevation control points from satellite-borne single-photon laser altimetry based on evaluation tags, namely, the document A Method of Extracting High-accuracy Elevation ControlPoints from ICESat-2 Altimetry Data (Li B., Xie H., Tong X., et al. Photogrammetric Engineering & Remote Sensing, 2021b, 84 (9): 579–589).

[0054] Finally, the aforementioned evaluation attributes and elevation corrections were trained using a random forest algorithm to obtain a DEM elevation correction regression model. This correction regression model was then used to correct the remaining DEM elevations. Considering that high-quality ICESat-2 satellite footprint terrain elevations are distributed over areas with slopes less than 25 degrees, to ensure the accuracy of the DEM elevation corrections, elevation corrections were only performed on the remaining DEM portions with slopes less than 25 degrees.

[0055] In summary, based on the above evaluation labels, the flowchart of the digital elevation model elevation correction method based on the random forest regression model proposed in this embodiment is as follows: Figure 1 shown.

[0056] This embodiment also provides a digital elevation model correction device based on a random forest regression model, comprising:

[0057] The model building and training module is configured to build an elevation correction regression model based on the random forest algorithm;

[0058] The training process of the elevation correction regression model includes: selecting a geographic location for reference from the satellite footprint terrain elevation data to obtain a digital elevation model; selecting elevation data of an area within a first distance range from the geographic location from the digital elevation model; comparing the elevation data with the satellite footprint terrain elevation data; and obtaining an elevation difference as an elevation correction value for the digital elevation model;

[0059] The elevation correction regression model built using the random forest algorithm is trained using elevation correction values ​​and evaluation attributes constructed based on the digital elevation model. The evaluation attributes include surface cover factors, terrain characteristics factors, spatial distribution factors within the distribution area of ​​the digital elevation model, and data source quality factors of the model.

[0060] The digital elevation model correction module loads the digital elevation model into the trained elevation correction regression model, modifies the elevation in the digital elevation model, and obtains the corrected digital elevation model.

[0061] Each module may include a memory and a processor, the memory stores a computer program, and the processor calls the computer program to execute the steps of the configuration method shown above.

[0062] The land cover factor is to use global land cover data to represent the land surface in the distribution area of ​​the digital elevation model;

[0063] The terrain characteristic factors are the slope, roughness, aspect, curvature and elevation statistics of the digital elevation model within the preset first window;

[0064] The spatial distribution factor is the three-dimensional coordinates of each pixel point of the digital elevation model relative to the zero point coordinates, calculated using the minimum values ​​of the plane geographic coordinates of the digital elevation model in the X and Y directions and the minimum elevation value in the preset first area as the zero point coordinate reference;

[0065] The data source quality factor of the model is the data source quality obtained from the quality product file attached to the digital elevation model.

[0066] The value of the first distance is within the range of 10-20m;

[0067] The specific modification of the elevation in the digital elevation model is as follows:

[0068] Modify the elevation of the portion of the digital elevation model where the slope is less than 25 degrees.

[0069] Experimental results and discussion:

[0070] In order to verify the method proposed in this embodiment, this embodiment selected the land area of ​​North Auckland, New Zealand and its vicinity (longitude: 174°-175° east longitude; latitude: 36°-37° south latitude) as the study area. In the study area, SRTM products and ATL08 products of ICESat-2 satellite are used as digital elevation model experimental data and spaceborne laser altimetry experimental data. The spatial resolution of SRTM products is 30m, which can provide coverage of all global land areas between 60 degrees north latitude and 56 degrees south latitude. In the ATL08 product, the length of the along-track distance interval is 100m, and the best-fit terrain elevation (referred to as "terrain elevation") at the center position of each along-track distance interval in non-water areas is used as the elevation correction reference for the digital elevation model. In the process of constructing the evaluation attributes, FROM-GLC10 products and GFCC30TC products are used as auxiliary data to generate surface cover evaluation attributes. Airborne LiDAR data, covering approximately half of the land area of ​​the study area, was used as elevation accuracy verification data. Its vertical accuracy is better than 10 cm and the point density is approximately 12 pts / m2.

[0071] According to the method proposed in this embodiment, the elevation of the SRTM product in the study area is corrected, and the correction results are verified and analyzed using the airborne LiDAR data in the study area. The results are as follows: Figure 2 As shown.

[0072] from Figure 2 It can be seen that after the method proposed in this embodiment, the accuracy of the SRTM product after elevation correction is significantly improved in flat land, hilly and mountainous areas, with the accuracy improvement ratio being 29-42%.

[0073] in conclusion:

[0074] This embodiment proposes a method for correcting the elevation of a digital elevation model based on a random forest regression model, in order to overcome the limitations of previous methods in correcting the elevation of digital elevation models in adjacent areas along the ground track. The proposed method is verified by using SRTM products and the ATL08 product of the ICESat-2 satellite as digital elevation model experimental data and satellite-borne laser altimetry experimental data, respectively, and using FROM-GLC10 products and GFCC30TC products as auxiliary data, as well as airborne LiDAR data as elevation accuracy verification data. The experimental results show that after the method proposed in this embodiment, the accuracy of the SRTM product after elevation correction is significantly improved in flat land, hilly areas, and mountainous areas, with the accuracy improvement ratio being 29-42%.

[0075] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A digital elevation model correction method based on a random forest regression model, characterized in that: include: The digital elevation model is loaded into a pre-built and trained elevation correction regression model, and the elevation in the digital elevation model is modified to obtain a corrected digital elevation model; The elevation correction regression model is constructed based on the random forest algorithm. The training process of the elevation correction regression model includes: selecting a geographic location for reference from the satellite footprint terrain elevation data, obtaining a digital elevation model, selecting elevation data of an area within a first distance range from the geographic location from the digital elevation model, and comparing the elevation with the satellite footprint terrain elevation data to obtain an elevation difference as an elevation correction for the digital elevation model; Training an elevation correction regression model constructed based on a random forest algorithm using elevation corrections and evaluation attributes constructed based on the digital elevation model, wherein the evaluation attributes include surface cover factors, terrain characteristics factors, spatial distribution factors, and data source quality factors within the distribution area of ​​the digital elevation model; The surface cover factor is to use global surface cover data to represent the surface within the distribution area of ​​the digital elevation model; The terrain characteristic factors are the slope, roughness, aspect, curvature and elevation statistics of the digital elevation model within a preset first window; The spatial distribution factor is the three-dimensional coordinates of each pixel point of the digital elevation model relative to the zero point coordinates, calculated using the minimum values ​​of the plane geographic coordinates of the digital elevation model in the X and Y directions and the minimum elevation value of the digital elevation model in the preset first area as the zero point coordinate reference; The data source quality factor of the model is the data source quality obtained according to the quality product file attached to the digital elevation model.

2. The digital elevation model correction method based on the random forest regression model according to claim 1, characterized in that: The value of the first distance is within the range of 10-20m.

3. The digital elevation model correction method based on the random forest regression model according to claim 1, characterized in that: The modification of the elevation in the digital elevation model is specifically as follows: Modify the elevation of the portion of the digital elevation model where the slope is less than 25 degrees.

4. A digital elevation model correction device based on a random forest regression model, characterized in that: include: The model building and training module is configured to build an elevation correction regression model based on the random forest algorithm; The training process of the elevation correction regression model includes: selecting a geographic location for reference from the satellite footprint terrain elevation data, obtaining a digital elevation model, selecting elevation data of an area within a first distance range from the geographic location from the digital elevation model, and comparing the elevation with the satellite footprint terrain elevation data to obtain an elevation difference as an elevation correction for the digital elevation model; Training an elevation correction regression model constructed based on a random forest algorithm using elevation corrections and evaluation attributes constructed based on the digital elevation model, wherein the evaluation attributes include surface cover factors, terrain characteristics factors, spatial distribution factors, and data source quality factors within the distribution area of ​​the digital elevation model; The digital elevation model correction module loads the digital elevation model into the trained elevation correction regression model, modifies the elevation in the digital elevation model, and obtains the corrected digital elevation model; The surface cover factor is to use global surface cover data to represent the surface within the distribution area of ​​the digital elevation model; The terrain characteristic factors are the slope, roughness, aspect, curvature and elevation statistics of the digital elevation model within a preset first window; The spatial distribution factor is the three-dimensional coordinates of each pixel point of the digital elevation model relative to the zero point coordinates, calculated using the minimum values ​​of the plane geographic coordinates of the digital elevation model in the X and Y directions and the minimum elevation value of the digital elevation model in the preset first area as the zero point coordinate reference; The data source quality factor of the model is the data source quality obtained according to the quality product file attached to the digital elevation model.

5. The digital elevation model correction device based on the random forest regression model according to claim 4 is characterized in that: The value of the first distance is within the range of 10-20m; The modification of the elevation in the digital elevation model is specifically as follows: Modify the elevation of the portion of the digital elevation model where the slope is less than 25 degrees.

Images