A DEM data fusion method based on multi-source data and machine learning

Abstract

The application relates to a DEM data fusion method based on multi-source data and machine learning, which comprises the following steps: collecting spaceplane radar terrain mapping task elevation model data, ICESat-2 satellite-borne laser radar data, land cover data and vegetation coverage data, and performing pretreatment; performing error factor analysis based on the pretreated data; sampling the data through a plurality of layered sampling methods for different factors, and respectively constructing a training set and a test set based on the sampled samples; constructing a DEM data fusion model based on extreme gradient boosting, inputting the training sets of different layered sampling methods into the DEM data fusion model, obtaining a DEM data fusion model, verifying the precision of the DEM data fusion model of the current layered sampling method through the test set corresponding to the layered sampling method, taking the highest precision as the final DEM data fusion model, and finally processing the abnormal values of the fused DEM data.

Description

Technical Field

[0001] This invention relates to a DEM data fusion method based on multi-source data and machine learning, belonging to the field of DEM data processing technology. Background Technology

[0002] Digital Elevation Models (DEMs) are among the most fundamental geospatial data types, widely used in the power industry for transmission line planning, geological and geographical risk assessment, power facility site selection, and the construction of 3D digital power grids. Due to limitations imposed by observation methods, processing techniques, and data acquisition time, single-source DEM data products inevitably suffer from low accuracy and inconsistent quality in areas with significant topographic relief. Therefore, leveraging the complementary characteristics of existing DEM data products to construct fusion strategies is one effective way to improve data usability.

[0003] For the fusion of DEM data of the same type, Karkee et al. used frequency domain transformation and fusion methods based on 90-meter resolution SRTM data and 30-meter resolution ASTER GDEM data to reduce the noise impact in high-resolution data. Fuss et al. extracted multiple DEMs of the same range from synthetic aperture radar images, used an improved k-means clustering method to obtain reliable elevation values ​​at the same pixel location, and averaged them to obtain the final fused product. Yamazaki et al. used ALOSAW3D, SRTM3, and VFP-DEM data, filtered out strip noise, speckle noise, tree height deviation, and absolute deviation in the data, and fused to generate the MERIT DEM dataset with a 3 arcsecond resolution covering a land area from 60°S to 82°N. For the fusion of different types of DEM data, spaceborne lidar is a widely used data format. Arefi and Reinartz used ICESat lidar altimetry data as control points and used interpolation algorithms to generate difference surfaces to correct outliers in ASTER GDEM.

[0004] However, existing DEM fusion methods only use a single data source for fusion, failing to fully utilize the advantages of multi-source data to assist in the fusion of DEM data. At the same time, they do not consider the prior errors of DEM datasets and do not analyze error-related factors, resulting in the fused DEM still having low accuracy in areas with complex terrain. Furthermore, existing DEM fusion methods have complex frameworks and high application thresholds, making them difficult to apply on a large scale. Summary of the Invention

[0005] To address the problems existing in the prior art, this invention proposes a DEM data fusion method based on multi-source data and machine learning.

[0006] The technical solution of the present invention is as follows:

[0007] On the one hand, this invention provides a DEM data fusion method based on multi-source data and machine learning, comprising the following steps:

[0008] Collect elevation model data from the Space Shuttle radar terrain mapping mission, ICESat-2 spaceborne lidar data, land cover data, and vegetation cover data, and preprocess the above data.

[0009] Based on the preprocessed data described above, an error factor analysis was performed on the elevation model data of the Space Shuttle radar terrain mapping mission.

[0010] By sampling the preprocessed data and the elevation model data of the Space Shuttle Radar Topographic Mapping Mission using various stratified sampling methods targeting different factors, the data of each laser point in the preprocessed ICESat-2 spaceborne lidar data and its corresponding elevation value, slope value, land cover type, and vegetation cover rate in the preprocessed land cover data and vegetation cover rate data were obtained. The data sampled by each of the above stratified sampling methods were divided into training samples and test samples, and training sets and test sets for different stratified sampling methods were constructed based on the training samples and test samples, respectively.

[0011] Based on the extreme gradient boosting and the error factor analysis results of the elevation model data of the above-mentioned space shuttle radar topographic mapping mission, a DEM data fusion model is constructed. The training sets of different hierarchical sampling methods are input into the DEM data fusion model, and the model parameters are updated by the grid search algorithm to obtain DEM data fusion models of different hierarchical sampling methods. The accuracy of the DEM data fusion model of the current hierarchical sampling method is verified by the test set of the corresponding hierarchical sampling method, and the one with the highest accuracy is taken as the final DEM data fusion model.

[0012] The final fused DEM data is obtained by fusing ICESat-2 spaceborne lidar data, land cover data, vegetation cover data, and elevation model data from the Space Shuttle radar topographic mapping mission, and by processing outliers in the fused DEM data.

[0013] As a preferred embodiment of the present invention, the preprocessing steps for the elevation model data of the space shuttle radar terrain mapping mission are as follows:

[0014] Slope data was calculated using elevation model data from the Space Shuttle radar terrain mapping mission.

[0015] The preprocessing steps for the ICESat-2 satellite-borne lidar data are as follows:

[0016] The laser point data corresponding to the strong laser beam is filtered by the flight direction of the aircraft, and the laser point data other than the laser point data corresponding to the strong laser beam is removed.

[0017] Remove laser point data whose cloud cover value is greater than a preset value from the laser point data corresponding to a strong laser beam;

[0018] Calculate the absolute value of the difference between the reference elevation value and the ICESat-2 interpolated land elevation value in each of the remaining laser point data, and retain the corresponding laser point data within three standard ranges of the absolute value of the difference from the mean;

[0019] Remove the laser point data whose absolute slope value is greater than the preset value from the above-mentioned retained laser point data, and then remove the laser point data that cannot correspond to the land cover data and vegetation cover data.

[0020] Finally, remove the laser point data in the remaining laser point data where the absolute value of the difference between the ICESat-2 interpolated land elevation value and the corresponding elevation value in the Space Shuttle Radar Topographic Mapping Mission elevation model data is greater than the preset value, and obtain the final laser point data.

[0021] The ICESat-2 interpolated land elevation values ​​of each laser point in the final laser point data are corrected to the same level as the elevation model data of the Space Shuttle Radar Topographic Mapping Mission, thus obtaining the preprocessed ICESat-2 spaceborne lidar data.

[0022] In a preferred embodiment of the present invention, the preprocessing steps for the land cover data and vegetation cover data are as follows:

[0023] The spatial coordinate system of the land cover data and vegetation cover data was transformed to be consistent with the elevation model data of the Space Shuttle Radar Topographic Mapping Mission. Then, the bilinear interpolation method was used to match all the pixels of the raster data with the elevation model data of the Space Shuttle Radar Topographic Mapping Mission to obtain the preprocessed land cover data and vegetation cover data.

[0024] As a preferred embodiment of the present invention, the error factor analysis step of the elevation model data for the space shuttle radar terrain mapping mission specifically includes:

[0025] Elevation, slope, vegetation coverage, and land cover type are classified. The elevation value in the ICESat-2 spaceborne lidar data is used as a reference value. The above-mentioned elevation, slope, vegetation coverage, and land cover type are used as error factors. The elevation error of the elevation model data of the space shuttle radar topographic mapping mission is calculated based on the reference value. The error factors of the elevation model data of the space shuttle radar topographic mapping mission are classified and analyzed by comparing the elevation error of the elevation model data of the space shuttle radar topographic mapping mission with the classification results of the above error factors.

[0026] The formula for calculating the elevation error of the space shuttle radar terrain mapping mission elevation model data based on reference values ​​is as follows:

[0027]

[0028] Where: error k The elevation error at point k; The elevation reference value for laser point k in the ICESat-2 spaceborne lidar data; z k This refers to the elevation value at point k in the elevation model data of the Space Shuttle radar terrain mapping mission.

[0029] The absolute value of the elevation error between the elevation model data of the space shuttle radar topographic mapping mission and the corresponding reference value is obtained. Based on this, the mean absolute error, root mean square error, and standard deviation of error are obtained respectively. These three indicators are used as indicators for error factor analysis.

[0030] In a preferred embodiment of the present invention, the mean absolute error (MAE) is used to reflect the overall error of the elevation model data from the space shuttle radar topographic mapping mission, as shown in the following formula:

[0031]

[0032] Where: n is the number of elevation reference values ​​in the ICESat-2 satellite-borne lidar data;

[0033] The root mean square error (RMSE) is used to reflect outliers in the elevation model data of the space shuttle radar topographic mapping mission, as shown in the following formula:

[0034]

[0035] The standard deviation of error (STD) is used to reflect the dispersion of the elevation model data error in the space shuttle radar topographic mapping mission, and to reflect the stability at different locations, as shown in the following formula:

[0036]

[0037] Where μ is the average value of the error.

[0038] In a preferred embodiment of the present invention, the training sample consists of the corresponding elevation values ​​of the laser points in the ICESat-2 spaceborne lidar data sampled by the corresponding hierarchical sampling method, the corresponding slope data of the laser points, the corresponding land cover type data of the laser points, and the corresponding vegetation coverage rate of the laser points in the elevation model data of the Space Shuttle radar terrain mapping mission.

[0039] The training set is a training data matrix constructed by taking the neighborhood window value. That is, in the above training samples, each valid ICESat-2 satellite lidar laser point data is traversed, and the data corresponding to the i×i window around the laser point data is obtained as the center, the training data matrix is ​​constructed, and the coordinates corresponding to the laser point data are added to the training data matrix.

[0040] In a preferred embodiment of the present invention, the outlier processing step is as follows:

[0041] For each pixel in the fused DEM data, take a j×j neighborhood window around the center pixel, calculate the difference between the elevation value of each pixel in the neighborhood window and the elevation value of the center pixel, and calculate the standard deviation of the elevation difference vector. Collect the number of pixels in the neighborhood whose absolute difference is greater than twice the standard deviation. If the number of pixels is greater than a preset value or the elevation value of the center pixel is obviously abnormal, the center pixel is judged as an abnormal point or noise point. Finally, replace the elevation value of the center pixel with the average elevation value of the neighborhood pixels.

[0042] On the other hand, the present invention also provides a DEM data fusion system based on multi-source data and machine learning, including a data preprocessing module, an error factor analysis module, a training sample selection module, a DEM data fusion module, and an anomaly handling module.

[0043] The data preprocessing module is used to collect elevation model data from the Space Shuttle radar terrain mapping mission, ICESat-2 spaceborne lidar data, land cover data, and vegetation cover data, and to preprocess the above data.

[0044] The error factor analysis module is used to perform error factor analysis on the elevation model data of the space shuttle radar terrain mapping mission based on the preprocessed data.

[0045] The training sample selection module is used to sample the preprocessed data and the elevation model data of the Space Shuttle Radar Topographic Mapping Mission using various stratified sampling methods targeting different factors. This results in the data of each laser point in the preprocessed ICESat-2 spaceborne lidar data and its corresponding elevation value, slope value, land cover type, and vegetation coverage rate in the preprocessed land cover data. The data sampled by each stratified sampling method is divided into training samples and test samples, and training sets and test sets for different stratified sampling methods are constructed based on the training samples and test samples, respectively.

[0046] The DEM data fusion module is used to construct a DEM data fusion model based on the error factor analysis results of the elevation model data of the Space Shuttle Radar Topographic Mapping Mission. The training sets of different stratified sampling methods are input into the DEM data fusion model, and the model parameters are updated through a grid search algorithm to obtain DEM data fusion models of different stratified sampling methods. The accuracy of the DEM data fusion model of the current stratified sampling method is verified by the test set of the corresponding stratified sampling method. The model with the highest accuracy is taken as the final DEM data fusion model. The final DEM data fusion model is used to fuse ICESat-2 spaceborne lidar data, land cover data, vegetation cover data and Space Shuttle Radar Topographic Mapping Mission elevation model data to obtain fused DEM data.

[0047] The anomaly handling module is used to process the abnormal values ​​of the DEM data after fusion to obtain the final fused DEM data.

[0048] In another aspect, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method as described in any embodiment of the present invention.

[0049] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in any embodiment of the present invention.

[0050] The present invention has the following beneficial effects:

[0051] 1. This invention applies the advantages of parallel and approximate computation of the extreme gradient boosting method to DEM fusion, thereby accelerating model training speed and improving fusion efficiency.

[0052] 2. This invention improves the DEM fusion accuracy in areas with complex land cover and large topographic variations by applying the analysis results of multi-source data and DEM error correlation factors to the construction of the fusion model.

[0053] 3. This invention introduces the advantage of regularization technology in extreme gradient boosting to avoid model overfitting, enabling the model to have better generalization ability. High-precision DEM fusion results can be obtained with simple parameter adjustments, thus lowering the threshold for using the method.

[0054] 4. This invention selects samples by comparing stratified sampling methods for different factors to obtain representative training samples. The model trained on this basis can be used for DEM fusion in other regions. Attached Figure Description

[0055] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] It should be understood that the step numbers used in the text are for ease of description only and are not intended to limit the order in which the steps are performed.

[0058] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0059] The terms “comprising” and “including” indicate the presence of the described feature, whole, step, operation, element and / or component, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or collections thereof.

[0060] The term “and / or” refers to any combination of one or more of the associated listed items, as well as all possible combinations, and includes these combinations.

[0061] Example 1:

[0062] See Figure 1 A DEM data fusion method based on multi-source data and machine learning includes the following steps:

[0063] Collect elevation model data from the Space Shuttle radar terrain mapping mission, ICESat-2 spaceborne lidar data, land cover data, and vegetation cover data, and preprocess the above data.

[0064] Based on the preprocessed data described above, an error factor analysis was performed on the elevation model data of the Space Shuttle radar terrain mapping mission.

[0065] Because the distribution of different land cover types and slope categories may be uneven in a specific area, and the distribution range of some types may be very small, using random sampling methods to divide the training and test samples may result in the training set failing to cover these types, leading to significant errors and reduced model generalization ability. Therefore, multiple stratified sampling methods targeting different factors are used to sample the preprocessed data and the elevation model data from the Space Shuttle radar topographic mapping mission. In this embodiment, for each land cover type and different slope division zones, the aforementioned stratified sampling methods targeting these two factors are used to sample the preprocessed data and the Space Shuttle radar data. The elevation model data of the topographic mapping mission was sampled to obtain the data of each laser point in the preprocessed ICESat-2 spaceborne lidar data and its corresponding elevation value, slope value in the preprocessed space shuttle radar topographic mapping mission elevation model data, the corresponding land cover type in the preprocessed land cover data, and the corresponding vegetation cover rate in the preprocessed vegetation cover rate data. 80% of the data sampled by each stratified sampling method was randomly selected as training samples and 20% was randomly selected as test samples. Training sets and test sets for different stratified sampling methods were constructed based on the training samples and test samples, respectively.

[0066] Based on the extreme gradient boosting and the error factor analysis results of the elevation model data of the above-mentioned space shuttle radar topographic mapping mission, a DEM data fusion model is constructed. The training sets of different hierarchical sampling methods are input into the DEM data fusion model, and the model parameters are updated by the grid search algorithm to obtain DEM data fusion models of different hierarchical sampling methods. The accuracy of the DEM data fusion model of the current hierarchical sampling method is verified by the test set of the corresponding hierarchical sampling method, and the one with the highest accuracy is taken as the final DEM data fusion model.

[0067] The final fused DEM data is obtained by fusing ICESat-2 spaceborne lidar data, land cover data, vegetation cover data, and elevation model data from the Space Shuttle radar topographic mapping mission, and by processing outliers in the fused DEM data.

[0068] As a preferred embodiment of this invention, the preprocessing steps for the elevation model data of the space shuttle radar terrain mapping mission are as follows:

[0069] Slope data was calculated using elevation model data from the Space Shuttle radar terrain mapping mission.

[0070] The preprocessing steps for the ICESat-2 satellite-borne lidar data are as follows:

[0071] The laser point data corresponding to the strong laser beams are filtered using the parameter `sc_orient` (the aircraft's flight orientation). During ATLAS detection, six laser beams are emitted, arranged in three pairs. To reduce duplicate data, this study selects the strong laser beams from each pair for subsequent analysis. When `sc_orient` equals 1, it indicates the aircraft is flying upright; in this case, the left side is a weak laser beam, and the right side is a strong laser beam. The opposite occurs when `sc_orient` equals 0. When `sc_orient` equals 2, it indicates the aircraft is turning; the data quality decreases in this case, and it is generally not used. Laser point data other than those corresponding to the strong laser beams are removed.

[0072] Since cloud cover has a significant impact on the accuracy of ICESat-2 data acquisition, laser point data with cloud_flag_atm > 2 in the laser point data corresponding to strong laser beams are removed.

[0073] Calculate the absolute value of the difference between the reference elevation value (dem_h) and the ICESat-2 interpolated land elevation value (h_te_interp) in each of the remaining laser point data. Retain the corresponding laser point data within three standard ranges of the absolute value of the difference from the mean (3 sigma principle), and filter out some laser point data with obvious deviations in elevation values.

[0074] Existing research shows that slope has a significant impact on the accuracy of ICESat-2 data points. Therefore, we removed the laser point data with an absolute slope value greater than 0.01 from the retained laser point data, and then removed the laser point data that could not correspond to land cover data and vegetation cover data.

[0075] Finally, remove laser point data from the remaining laser point data where the absolute value of the difference between the ICESat-2 interpolated land elevation value and the corresponding elevation value in the Space Shuttle Radar Topographic Mapping Mission elevation model data is greater than 100 meters, to obtain the final laser point data. Under normal circumstances, the error will not exceed 100 meters, while data points that exceed this range are often due to large deviations during satellite data acquisition and therefore cannot be used as research data.

[0076] The ICESat-2 interpolated land elevation values ​​of each laser point in the final laser point data are corrected to the same level as the elevation model data of the Space Shuttle Radar Topographic Mapping Mission, resulting in preprocessed ICESat-2 spaceborne lidar data. Since the geographic datum of both ICESat-2 and the elevation model of the Space Shuttle Radar Topographic Mapping Mission is WGS84, only the difference between the level and the datum needs to be considered.

[0077] In a preferred embodiment of this invention, the preprocessing steps for the land cover data and vegetation cover data are as follows:

[0078] The spatial coordinate system of the land cover data and vegetation cover data was transformed to be consistent with the elevation model data of the Space Shuttle Radar Topographic Mapping Mission. Then, the bilinear interpolation method was used to match all the pixels of the raster data with the elevation model data of the Space Shuttle Radar Topographic Mapping Mission to obtain the preprocessed land cover data and vegetation cover data.

[0079] As a preferred embodiment of this invention, the error factor analysis steps for the elevation model data of the space shuttle radar terrain mapping mission are specifically as follows:

[0080] The land cover was classified according to elevation, slope, vegetation cover, and land cover type. Elevation was divided into six categories: <500m, 500-1000m, 1000-1500m, 1500-2000m, 2000-2500m, and >2500m. Slope was divided into seven categories: 0-10°, 10°-20°, 20°-30°, 30°-40°, 40°-50°, 50°-60°, and above 60°. Vegetation cover was divided into seven categories: 0-20%, 20-40%, 40-60%, 60- The land cover types are classified into five categories: 80%, 80-100%, etc. Based on existing classifications, the elevation values ​​in the ICESat-2 satellite lidar data are used as reference values. The elevation, slope, vegetation coverage, and land cover type are used as error factors. The elevation error of the elevation model data of the Space Shuttle Radar Topographic Mapping Mission is calculated based on the reference values. The error factors of the elevation model data of the Space Shuttle Radar Topographic Mapping Mission are classified and analyzed by comparing the elevation error of the elevation model data of the Space Shuttle Radar Topographic Mapping Mission with the classification results of the above error factors.

[0081] The formula for calculating the elevation error of the space shuttle radar terrain mapping mission elevation model data based on reference values ​​is as follows:

[0082]

[0083] Where: error k The elevation error at point k; The elevation reference value for laser point k in the ICESat-2 spaceborne lidar data; z k This refers to the elevation value at point k in the elevation model data of the Space Shuttle radar terrain mapping mission.

[0084] Elevation errors typically have both positive and negative aspects. To avoid these two aspects canceling each other out and affecting the actual error assessment results, the absolute value of the elevation error between the elevation model data of the space shuttle radar topographic mapping mission and the corresponding reference value is taken. Based on this, the mean absolute error, root mean square error, and standard deviation of error are obtained respectively. These three indicators are used as indicators for error factor analysis.

[0085] In a preferred embodiment of this invention, the mean absolute error (MAE) is used to reflect the overall error of the elevation model data from the space shuttle radar topographic mapping mission, but it is not sensitive to outliers, as shown in the following formula:

[0086]

[0087] Where: n is the number of elevation reference values ​​in the ICESat-2 satellite-borne lidar data;

[0088] The root mean square error is highly sensitive to outliers (particularly large or small errors) in the data sequence and can more accurately reflect the error situation of the data, as shown in the following formula:

[0089]

[0090] The standard deviation of the error is used to reflect the dispersion of the elevation model data error in the space shuttle radar topographic mapping mission, and to reflect the stability at different locations, as shown in the following formula:

[0091]

[0092] Where μ is the average value of the error.

[0093] As a preferred embodiment of this example, the training sample consists of the corresponding elevation values ​​of the laser points in the ICESat-2 spaceborne lidar data sampled by the corresponding hierarchical sampling method, the corresponding slope data of the laser points, the corresponding land cover type data of the laser points, and the corresponding vegetation coverage rate of the laser points in the elevation model data of the Space Shuttle radar terrain mapping mission.

[0094] The training set is a training data matrix constructed by taking the neighborhood window value. That is, in the above training samples, each valid ICESat-2 satellite lidar laser point data is traversed, and the data corresponding to the surrounding 3×3 window is obtained with the laser point data as the center to construct the training data matrix, and the coordinates corresponding to the laser point data are added to the training data matrix.

[0095] In a preferred embodiment of this invention, the outlier handling step is as follows:

[0096] For each pixel in the fused DEM data, a 5×5 neighborhood window is taken around the center pixel. The difference between the elevation value of each pixel in the neighborhood window and the elevation value of the center pixel is calculated, and the standard deviation of the elevation difference vector is calculated. The number of pixels in the neighborhood whose absolute difference is greater than twice the standard deviation is collected. If more than 75% of the pixels in the neighborhood have an absolute difference greater than twice the standard deviation or the elevation value of the center pixel is obviously abnormal, the center pixel is judged as an outlier or noise point. Finally, the average elevation value of the neighborhood pixels is used to replace the elevation value of the center pixel.

[0097] Example 2:

[0098] A DEM data fusion system based on multi-source data and machine learning includes a data preprocessing module, an error factor analysis module, a training sample selection module, a DEM data fusion module, and an anomaly handling module.

[0099] The data preprocessing module is used to collect elevation model data from the Space Shuttle radar terrain mapping mission, ICESat-2 spaceborne lidar data, land cover data, and vegetation cover data, and to preprocess the above data.

[0100] The error factor analysis module is used to perform error factor analysis on the elevation model data of the space shuttle radar terrain mapping mission based on the preprocessed data.

[0101] The training sample selection module is used to sample the preprocessed data and the elevation model data of the Space Shuttle Radar Topographic Mapping Mission using various stratified sampling methods targeting different factors. This results in the data of each laser point in the preprocessed ICESat-2 spaceborne lidar data and its corresponding elevation value, slope value, land cover type, and vegetation coverage rate in the preprocessed land cover data. The data sampled by each stratified sampling method is divided into training samples and test samples, and training sets and test sets for different stratified sampling methods are constructed based on the training samples and test samples, respectively.

[0102] The DEM data fusion module is used to construct a DEM data fusion model based on the error factor analysis results of the elevation model data of the Space Shuttle Radar Topographic Mapping Mission. The training sets of different stratified sampling methods are input into the DEM data fusion model, and the model parameters are updated through a grid search algorithm to obtain DEM data fusion models of different stratified sampling methods. The accuracy of the DEM data fusion model of the current stratified sampling method is verified by the test set of the corresponding stratified sampling method. The model with the highest accuracy is taken as the final DEM data fusion model. The final DEM data fusion model is used to fuse ICESat-2 spaceborne lidar data, land cover data, vegetation cover data and Space Shuttle Radar Topographic Mapping Mission elevation model data to obtain fused DEM data.

[0103] The anomaly handling module is used to process the abnormal values ​​of the DEM data after fusion to obtain the final fused DEM data.

[0104] This embodiment is used to implement the functions in Embodiment 1, and will not be described again here.

[0105] Example 3:

[0106] This embodiment proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described in any embodiment of the present invention.

[0107] Example 4:

[0108] This embodiment proposes a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the method described in any embodiment of the present invention.

[0109] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, A and B simultaneously, or B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of singular or plural items. For example, at least one of a, b, and c can represent: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, and c can be single or multiple.

[0110] Those skilled in the art will recognize that the units and algorithm steps described in the embodiments disclosed herein can be implemented using electronic hardware, computer software, or a combination of electronic hardware and software. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0111] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0112] In the several embodiments provided in this application, any function, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0113] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A DEM data fusion method based on multi-source data and machine learning, characterized in that, Includes the following steps: Collect elevation model data from the Space Shuttle radar terrain mapping mission, ICESat-2 spaceborne lidar data, land cover data, and vegetation cover data, and preprocess the above data. Based on the preprocessed data described above, an error factor analysis was performed on the elevation model data of the Space Shuttle radar terrain mapping mission. By sampling the preprocessed data and the elevation model data of the Space Shuttle Radar Topographic Mapping Mission using various stratified sampling methods targeting different factors, the data of each laser point in the preprocessed ICESat-2 spaceborne lidar data and its corresponding elevation value, slope value, land cover type, and vegetation cover rate in the preprocessed land cover data and vegetation cover rate data were obtained. The data sampled by each of the above stratified sampling methods were divided into training samples and test samples, and training sets and test sets for different stratified sampling methods were constructed based on the training samples and test samples, respectively. Based on the extreme gradient boosting and the error factor analysis results of the elevation model data of the above-mentioned space shuttle radar topographic mapping mission, a DEM data fusion model is constructed. The training sets of different hierarchical sampling methods are input into the DEM data fusion model, and the model parameters are updated by the grid search algorithm to obtain DEM data fusion models of different hierarchical sampling methods. The accuracy of the DEM data fusion model of the current hierarchical sampling method is verified by the test set of the corresponding hierarchical sampling method, and the one with the highest accuracy is taken as the final DEM data fusion model. The final fused DEM data is obtained by fusing ICESat-2 spaceborne lidar data, land cover data, vegetation cover data, and elevation model data from the Space Shuttle radar topographic mapping mission, and by processing outliers in the fused DEM data.

2. The DEM data fusion method based on multi-source data and machine learning according to claim 1, characterized in that, The preprocessing steps for the elevation model data from the Space Shuttle radar terrain mapping mission are as follows: Slope data was calculated using elevation model data from the Space Shuttle radar terrain mapping mission. The preprocessing steps for the ICESat-2 satellite-borne lidar data are as follows: The laser point data corresponding to the strong laser beam is filtered by the flight direction of the aircraft, and the laser point data other than the laser point data corresponding to the strong laser beam is removed. Remove laser point data whose cloud cover value is greater than a preset value from the laser point data corresponding to a strong laser beam; Calculate the absolute value of the difference between the reference elevation value and the ICESat-2 interpolated land elevation value in each of the remaining laser point data, and retain the corresponding laser point data within three standard ranges of the absolute value of the difference from the mean; Remove the laser point data whose absolute slope value is greater than the preset value from the above-mentioned retained laser point data, and then remove the laser point data that cannot correspond to the land cover data and vegetation cover data. Finally, remove the laser point data in the remaining laser point data where the absolute value of the difference between the ICESat-2 interpolated land elevation value and the corresponding elevation value in the Space Shuttle Radar Topographic Mapping Mission elevation model data is greater than the preset value, and obtain the final laser point data. The ICESat-2 interpolated land elevation values ​​of each laser point in the final laser point data are corrected to the same level as the elevation model data of the Space Shuttle Radar Topographic Mapping Mission, thus obtaining the preprocessed ICESat-2 spaceborne lidar data.

3. The DEM data fusion method based on multi-source data and machine learning according to claim 1, characterized in that, The preprocessing steps for the land cover data and vegetation cover data are as follows: The spatial coordinate system of the land cover data and vegetation cover data was transformed to be consistent with the elevation model data of the Space Shuttle Radar Topographic Mapping Mission. Then, the bilinear interpolation method was used to match all the pixels of the raster data with the elevation model data of the Space Shuttle Radar Topographic Mapping Mission to obtain the preprocessed land cover data and vegetation cover data.

4. The DEM data fusion method based on multi-source data and machine learning according to claim 1, characterized in that, The specific steps for analyzing the error factors in the elevation model data of the space shuttle radar terrain mapping mission are as follows: Elevation, slope, vegetation coverage, and land cover type are classified. The elevation value in the ICESat-2 spaceborne lidar data is used as a reference value. The above-mentioned elevation, slope, vegetation coverage, and land cover type are used as error factors. The elevation error of the elevation model data of the space shuttle radar topographic mapping mission is calculated based on the reference value. The error factors of the elevation model data of the space shuttle radar topographic mapping mission are classified and analyzed by comparing the elevation error of the elevation model data of the space shuttle radar topographic mapping mission with the classification results of the above error factors. The formula for calculating the elevation error of the space shuttle radar terrain mapping mission elevation model data based on reference values ​​is as follows: Where: error k The elevation error at point k; The elevation reference value for laser point k in the ICESat-2 spaceborne lidar data; z k This refers to the elevation value at point k in the elevation model data of the Space Shuttle radar terrain mapping mission. The absolute value of the elevation error between the elevation model data of the space shuttle radar topographic mapping mission and the corresponding reference value is obtained. Based on this, the mean absolute error, root mean square error, and standard deviation of error are obtained respectively. These three indicators are used as indicators for error factor analysis.

5. The DEM data fusion method based on multi-source data and machine learning according to claim 4, characterized in that, The Mean Absolute Error (MAE) is used to reflect the overall error of the elevation model data from the Space Shuttle radar terrain mapping mission, as shown in the following formula: Where: n is the number of elevation reference values ​​in the ICESat-2 satellite-borne lidar data; The root mean square error (RMSE) is used to reflect outliers in the elevation model data of the space shuttle radar topographic mapping mission, as shown in the following formula: The standard deviation of error (STD) is used to reflect the dispersion of the elevation model data error in the space shuttle radar topographic mapping mission, and to reflect the stability at different locations, as shown in the following formula: Where μ is the average value of the error.

6. The DEM data fusion method based on multi-source data and machine learning according to claim 1, characterized in that, The training samples consist of the corresponding elevation values ​​of the laser points in the ICESat-2 spaceborne lidar data sampled by the corresponding stratified sampling method, the corresponding slope data of the laser points, the corresponding land cover type data of the laser points, and the corresponding vegetation coverage rate of the laser points in the elevation model data of the Space Shuttle radar terrain mapping mission. The training set is a training data matrix constructed by taking the neighborhood window value. That is, in the above training samples, each valid ICESat-2 satellite lidar laser point data is traversed, and the data corresponding to the i×i window around the laser point data is obtained as the center, the training data matrix is ​​constructed, and the coordinates corresponding to the laser point data are added to the training data matrix.

7. The DEM data fusion method based on multi-source data and machine learning according to claim 1, characterized in that, The outlier handling steps are as follows: For each pixel in the fused DEM data, take a j×j neighborhood window around the center pixel, calculate the difference between the elevation value of each pixel in the neighborhood window and the elevation value of the center pixel, and calculate the standard deviation of the elevation difference vector. Collect the number of pixels in the neighborhood whose absolute difference is greater than twice the standard deviation. If the number of pixels is greater than a preset value or the elevation value of the center pixel is obviously abnormal, the center pixel is judged as an abnormal point or noise point. Finally, replace the elevation value of the center pixel with the average elevation value of the neighborhood pixels.

8. A DEM data fusion system based on multi-source data and machine learning, characterized in that, It includes a data preprocessing module, an error factor analysis module, a training sample selection module, a DEM data fusion module, and an anomaly handling module; The data preprocessing module is used to collect elevation model data from the Space Shuttle radar terrain mapping mission, ICESat-2 spaceborne lidar data, land cover data, and vegetation cover data, and to preprocess the above data. The error factor analysis module is used to perform error factor analysis on the elevation model data of the space shuttle radar terrain mapping mission based on the preprocessed data. The training sample selection module is used to sample the preprocessed data and the elevation model data of the Space Shuttle Radar Topographic Mapping Mission using various stratified sampling methods targeting different factors. This results in the data of each laser point in the preprocessed ICESat-2 spaceborne lidar data and its corresponding elevation value, slope value, land cover type, and vegetation coverage rate in the preprocessed land cover data. The data sampled by each stratified sampling method is divided into training samples and test samples, and training sets and test sets for different stratified sampling methods are constructed based on the training samples and test samples, respectively. The DEM data fusion module is used to construct a DEM data fusion model based on the error factor analysis results of the elevation model data of the Space Shuttle Radar Topographic Mapping Mission. The training sets of different stratified sampling methods are input into the DEM data fusion model, and the model parameters are updated through a grid search algorithm to obtain DEM data fusion models of different stratified sampling methods. The accuracy of the DEM data fusion model of the current stratified sampling method is verified by the test set of the corresponding stratified sampling method. The model with the highest accuracy is taken as the final DEM data fusion model. The final DEM data fusion model is used to fuse ICESat-2 spaceborne lidar data, land cover data, vegetation cover data and Space Shuttle Radar Topographic Mapping Mission elevation model data to obtain fused DEM data. The anomaly handling module is used to process the abnormal values ​​of the DEM data after fusion to obtain the final fused DEM data.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 7.

Images