High-precision topographic mapping system and method based on multi-source remote sensing data

Abstract

The invention discloses a high-precision topographic mapping system and method based on multi-source remote sensing data, and relates to the technical field of topographic mapping, and the method comprises the steps: obtaining an optical remote sensing image, a radar remote sensing image, LiDAR point cloud data and meteorological index data of a target region, and carrying out the data preprocessing of the optical remote sensing image, the radar remote sensing image and the LiDAR point cloud data; obtaining the surface feature types of the target area and the coverage area of each surface feature type, and carrying out the weighted fusion of the optical remote sensing image and the radar remote sensing image of the target area according to the surface feature types of the target area and the coverage area of each surface feature type, and obtaining a self-adaptive fusion remote sensing image; according to the method, points in LiDAR point cloud data are divided into distinguished ground points and non-ground points, a digital elevation model is constructed according to the ground points in the LiDAR point cloud data, three-dimensional modeling of the terrain is performed on the basis of the digital elevation model in combination with texture information of a self-adaptive fusion remote sensing image, a three-dimensional terrain model is generated, and the terrain surveying and mapping precision is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of topographic surveying and mapping, and specifically to a high-precision topographic surveying and mapping system and method based on multi-source remote sensing data. Background Art

[0002] CN118362102A "A topographic surveying and mapping system and method based on aerial surveying and mapping technology" includes: obtaining multiple image data and lidar point cloud data of the area to be surveyed, preprocessing the image data of the area to be surveyed, and each piece of image data of the area to be surveyed includes position information and elevation information; constructing a convolutional neural network model for surveying and mapping the terrain; obtaining the weather index of the area to be surveyed; obtaining the fitting curve of flight altitude with terrain complexity, weather index, and camera parameters through experiments; the unmanned aerial vehicle flies at a determined flight altitude and position in one-to-one matching to obtain new image data of the area to be surveyed; comparing, filling, and combining the new image data of the area to be surveyed with the image data of the area to be surveyed to obtain a surveyed image; importing the obtained lidar point cloud data; performing aerial triangulation on the surveyed image and establishing a three-dimensional topographic map in combination with the three-dimensional surface contour.

[0003] CN117848302B "A real-time terrain intelligent surveying and mapping method and system" includes: fusing multiple surveying and mapping data to generate initial contour map data, dividing sub-regions for each contour map, and selecting edge feature points of each initial contour map; using a GPS receiver and real-time kinematic differential positioning technology to obtain the real-time coordinates of the moving station, calculating the position deviation index to form the first terrain change index; obtaining the surface image of the sub-region through surveying equipment, comparing and calculating the area change index, and based on the surface image of each contour region and the selection method of edge feature points, selecting and calculating the curvature change index at the edge feature points, and then obtaining the second terrain change index; synthesizing the two indices to form a comprehensive terrain change index, comparing it with the re-surveying threshold, and flexibly adopting a re-surveying strategy.

[0004] Topographic surveying and mapping is of great significance in many fields such as urban planning, land management, geological exploration, and water conservancy projects. Traditional topographic surveying and mapping methods often rely on ground measurement means such as total stations and levels. Although the accuracy is high, the efficiency is low, and it is difficult to quickly obtain topographic information of large areas. With the development of remote sensing technology, using means such as satellite remote sensing and aerial remote sensing for topographic surveying and mapping has gradually become the mainstream. However, a single remote sensing data source has certain limitations. For example, optical remote sensing images are easily affected by weather, and radar remote sensing images may not be precise enough in terms of terrain details. Therefore, how to comprehensively utilize multiple remote sensing data, overcome their respective shortcomings, and improve the accuracy and reliability of topographic surveying and mapping has become an urgent problem to be solved. Summary of the Invention

[0005] To solve the above technical problems, the object of the present invention is to provide a high-precision topographic mapping method based on multi-source remote sensing data, including the following steps:

[0006] Step s1: Obtain optical remote sensing images, radar remote sensing images, LiDAR point cloud data, and meteorological index data of the target area. For example, select the WorldView-3 satellite to obtain optical remote sensing images, whose panchromatic image resolution can reach 0.31 meters, and the multi-spectral images cover multiple bands. At the same time, use the SAR data of the Sentinel-1 satellite to obtain radar remote sensing images to penetrate the clouds and vegetation that may exist in mountainous areas and obtain the general undulation information of the terrain. In addition, use an airborne LiDAR system to perform flight scanning to obtain the point cloud data of the mountainous area, ensuring that the elevation changes and detailed features of the terrain can be accurately measured, and perform data preprocessing on the optical remote sensing images, radar remote sensing images, and LiDAR point cloud data. The data preprocessing process includes geometric correction of the optical remote sensing images. Select obvious ground object markers (such as the peaks of mountains, road intersections at the bottom of valleys, etc.) in the mountainous area as ground control points, and use the polynomial fitting method for geometric correction, and use the 6S atmospheric correction model for radiometric correction to eliminate the atmospheric influence and improve the image quality. For radar remote sensing images, perform geometric correction according to their orbital parameters and imaging models, and perform calibration processing to convert the image gray values into backscattering coefficients. After correcting the installation angle deviation and time delay calibration of the LiDAR point cloud data, use a slope-based filtering algorithm to separate ground points and non-ground points, remove the point cloud data of non-ground objects such as trees and buildings, and retain accurate terrain point cloud information;

[0007] Step s2: Obtain the land cover types of the target area and the coverage areas of each land cover type. According to the land cover types of the target area and the coverage areas of each land cover type, perform weighted fusion on the optical remote sensing images and radar remote sensing images of the target area to obtain an adaptive fusion remote sensing image;

[0008] Step s3: Divide each point in the LiDAR point cloud data into ground points and non-ground points. According to each ground point in the LiDAR point cloud data, construct a digital elevation model. On the basis of the digital elevation model, combine the texture information of the adaptive fusion remote sensing image to perform three-dimensional modeling of the terrain and generate a three-dimensional terrain model.

[0009] Further, the process of obtaining the land cover types of the target area and the coverage areas of each land cover type includes:

[0010] Spectral feature extraction is performed on the pre-processed optical remote sensing image to obtain the spectral reflection data of different bands in the target area. Based on the spectral reflection data of different bands, the land cover types in the target area and the coverage area of each land cover type are obtained. By using the spectral reflectance differences of land covers in different bands, the land cover types and coverage areas can be identified more accurately. For example, in the near-infrared band, the reflectance of vegetation is significantly higher than that of other land covers. By analyzing the reflectance ratio of vegetation in the near-infrared and red bands (such as the normalized difference vegetation index NDVI), the vegetation coverage can be accurately estimated. The calculation formula is NDVI = (NIR - R) / (NIR + R), where NIR is the reflectance in the near-infrared band and R is the reflectance in the red band. When the NDVI value is close to 1, it indicates a high vegetation coverage; when the NDVI value is close to -1, it indicates a non-vegetation covered area.

[0011] Further, based on the land cover types in the target area and the coverage area of each land cover type, the weighted fusion of the optical remote sensing image and the radar remote sensing image of the target area is carried out. The process of obtaining the adaptive fusion remote sensing image includes:

[0012] Construct an adaptive weight adjustment model. Input the land cover types in the target area, the coverage area of each land cover type, and the meteorological index data into the adaptive weight adjustment model. According to the output of the adaptive weight adjustment model, obtain the first weight corresponding to the optical remote sensing image and the second weight corresponding to the radar remote sensing image. Perform weighted fusion on each pixel of the optical remote sensing image and the radar remote sensing image according to the first weight and the second weight to obtain the adaptive fusion remote sensing image.

[0013] The data-level fusion of the optical remote sensing image and the radar remote sensing image is carried out by using the weighted fusion method. Different weights are assigned to them according to the importance and reliability of the optical remote sensing image and the radar remote sensing image in topographic mapping. For example, in areas with less vegetation coverage, the optical image has a higher resolution and clear land cover details, and a higher weight can be given; while in areas with dense vegetation coverage or cloud and fog occlusion, the radar image can penetrate vegetation and cloud and fog, and a higher weight is given to the radar image at this time.

[0014] Further, the process of performing weighted fusion on each pixel of the optical remote sensing image and the radar remote sensing image according to the first weight and the second weight to obtain the adaptive fusion remote sensing image includes:

[0015] Let the pixel value of the optical remote sensing image be I q , and the pixel value of the radar remote sensing image be I r , the first weight be ω q , and the second weight be ω r . The calculation formula for the pixel value I f after weighted fusion is: I f = ωq I q + ω r I r ;

[0016] For example, if the weight ω q of the optical remote sensing image is 0.6, and the weight ω r of the radar remote sensing image is 0.4, the pixel value I q of a certain pixel in the optical remote sensing image is 100, and the corresponding pixel value I r of the radar remote sensing image is 80, then the pixel value I f after weighted fusion is 92.

[0017] Furthermore, the process of constructing the adaptive weight adjustment model includes:

[0018] Select multiple test areas in the target area in advance. Different test areas include different ground object types and the coverage areas of each ground object type. Collect the optical remote sensing images and radar remote sensing images of each test area under different meteorological index data conditions, perform data preprocessing on the optical remote sensing images and radar remote sensing images, and use different preset weight combinations (such as the weight of the optical remote sensing image being 0, 0.2, 0.4, 0.6, 0.8, 1, and the corresponding weight of the radar remote sensing image being 1, 0.8, 0.6, 0.4, 0.2, 0) to perform data fusion on the preprocessed optical remote sensing images and radar remote sensing images of each test area under different meteorological index data conditions, and obtain the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions;

[0019] Obtain the field measurement data of each test area (such as the terrain point data measured by a total station), compare the field measurement data of each test area with the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions, and obtain the accuracy errors of the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions;

[0020] Construct an adaptive weight adjustment model based on deep learning. Use the different ground object types included in each test area, the coverage areas of each ground object type, and the accuracy errors of the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions as the training set and the test set. Input the training set into the adaptive weight adjustment model for training until the loss function is trained stably, save the model parameters, test the adaptive weight adjustment model through the test set until it meets the preset requirements, and output the adaptive weight adjustment model;

[0021] Building an adaptive weight adjustment model based on deep learning is a complex process that involves multiple steps such as model selection, training, validation, and testing. In this embodiment, a deep belief network (DBN) is selected as the deep learning architecture, and the mean squared error loss function is chosen. It calculates the average of the squares of the differences between the predicted values and the true values. In the scenario of this adaptive weight adjustment model, the predicted values are the pixel values of the remotely sensed images (such as elevation values, land cover classification probabilities, etc.) after adaptive weight fusion, and the true values are the accurate values obtained through high-precision field measurements (such as terrain point data measured by total stations). Subsequently, the prepared training set is input into the selected deep learning model for training. During the training process, the weights are continuously updated through the backpropagation algorithm, causing the loss function to gradually decrease until it reaches a stable state. During this period, techniques such as Early Stopping are used to avoid overfitting. In addition to the basic training process, the various parameters of the model are optimized through Grid Search at the same time. The parameters include the learning rate, batch size, regularization coefficient, etc.;

[0022] When the model training is completed and the parameters are adjusted, the final evaluation is carried out through the test set to obtain the evaluation results of the model. The evaluation results include classification metrics such as accuracy, recall rate, F1 score, etc. According to the evaluation results on the test set, it is judged whether the model meets the expected standards. If the requirements are met, the model parameters are saved and preparation for deployment is made; if not ideal, it is necessary to return to a previous stage to re-examine issues such as data quality, model structure, or training strategy.

[0023] Further, the process of classifying each point in the LiDAR point cloud data into ground points and non-ground points includes:

[0024] Obtain the slope between each point and its adjacent points in the LiDAR point cloud data, preset the initial grid size of the target area, divide the target area into several identical initial grids according to the initial grid size, construct a two-dimensional coordinate system for the target area, set the starting point (Xmin, Ymin) of the target area, and divide it successively according to the selected grid size. For example, for a square grid, if the side length of the grid is L and the lower left corner coordinates of the first grid are the starting point (Xmin, Ymin), then the lower left corner coordinates of the second grid are (Xmin + L, Ymin), and the lower left corner coordinates of the third grid are (Xmin, Ymin + L), and so on until the entire target area is covered. At the same time, consider the boundary conditions during the division process to ensure that the grids in the last row and the last column can completely cover the boundary part of the target area; obtain the average slope within each initial grid, compare the slope between each point and its adjacent points within each initial grid with the average slope within each initial grid, obtain the slope difference of each point within each initial grid, compare the slope difference of each point with the preset slope difference threshold, mark the points with a slope difference greater than the slope difference threshold as non-ground points, and mark the points with a slope difference less than or equal to the slope difference threshold as ground points.

[0025] Further, the process of constructing a digital elevation model based on the ground points in the LiDAR point cloud data includes:

[0026] Obtain the elevation of the ground points within each initial grid, perform statistical analysis on the elevation of the ground points within each initial grid, obtain the elevation standard deviation within each initial grid, construct a grid size comparison table, where the grid size comparison table includes the grid sizes corresponding to different elevation standard deviations. It should be further noted that the initial grid size is greater than several grid sizes included in the grid size comparison table. Obtain the corresponding new grid size for each initial grid according to the elevation standard deviation within each initial grid and the grid size comparison table, divide each initial grid into several identical new grids according to the corresponding new grid size for each initial grid. If the new grid size is the same as the initial grid size, no division of the initial grid is performed, and the initial grid is marked as a new grid. Construct a digital elevation model for the new grid according to each new grid and the elevation of each point within each new grid.

[0027] Further, on the basis of the digital elevation model, the process of generating a three-dimensional terrain model by combining the texture information of the adaptive fusion remote sensing image for three-dimensional terrain modeling includes:

[0028] Constructing the coordinate matching relationship between the digital elevation model and the adaptive fusion remote sensing image. For the adaptively fused remote sensing image and the digital elevation model after data preprocessing, by determining their reference points (such as the corresponding relationship between the four corner points of the image and the boundary points of the digital elevation model) and adjusting the resolution differences (such as the comparison between the actual ground size represented by each pixel of the image and the grid size of the digital elevation model), the coordinate unification is achieved. Texture sampling is performed on the adaptively fused remote sensing image, and the sampled texture is attached to the corresponding position on the surface of the digital elevation model according to the coordinate matching relationship to generate a three-dimensional terrain model.

[0029] A high-precision terrain mapping system based on multi-source remote sensing data, including a cloud platform, which is communicatively connected to a data acquisition module, an image feature fusion module, and a three-dimensional modeling module;

[0030] The data acquisition module is used to obtain the optical remote sensing image, radar remote sensing image, LiDAR point cloud data, and meteorological index data of the target area, and perform data preprocessing on the optical remote sensing image, radar remote sensing image, and LiDAR point cloud data;

[0031] The image feature fusion module is used to obtain the land cover types of the target area and the coverage area of each land cover type, and perform weighted fusion on the optical remote sensing image and radar remote sensing image of the target area according to the land cover types of the target area and the coverage area of each land cover type to obtain an adaptively fused remote sensing image;

[0032] The three-dimensional modeling module is used to divide each point in the LiDAR point cloud data into ground points and non-ground points, construct a digital elevation model based on the ground points in the LiDAR point cloud data, and perform three-dimensional modeling of the terrain by combining the texture information of the adaptively fused remote sensing image on the basis of the digital elevation model to generate a three-dimensional terrain model.

[0033] Compared with the prior art, the beneficial effects of the present invention are:

[0034] 1. The data fusion advantage significantly improves the mapping accuracy:

[0035] Comprehensively utilize optical remote sensing images, radar remote sensing images and LiDAR point cloud data to give full play to the advantages of each data source. Optical images can clearly show the appearance details such as the color and texture of ground objects, which is conducive to identifying land use types and vegetation species; radar images are not restricted by light and weather, can penetrate clouds to obtain information, and accurately reflect the terrain undulation and ground object structure; LiDAR point cloud data provides high-precision three-dimensional terrain information. The fusion of the three overcomes the limitations of a single data source. For example, in mountainous area mapping, optical images present the appearance of vegetation and rocks, radar images penetrate the vegetation to reveal the terrain elevation difference, and LiDAR accurately measures the elevation, comprehensively capturing complex terrain and landforms. Compared with relying only on a single data source, the mapping accuracy is greatly improved, and the generated digital elevation model fits the actual terrain better.

[0036] Adopt adaptive weighted fusion of optical images and radar images, and dynamically adjust the weights according to ground object types, coverage areas and meteorological indicators. Increase the weight of radar images in densely vegetated areas to obtain the terrain, and increase the weight of optical images in sunny and open areas to show the details of ground objects, ensuring that the fused images accurately reflect the ground surface in different environments and avoiding information loss or deviation caused by fixed-weight fusion, further optimizing the mapping accuracy.

[0037] 2. Intelligent model realizes efficient and accurate weight allocation:

[0038] Build an adaptive weight adjustment model based on deep learning and train and optimize it using multi-region test data. The model learns the optimal weight combinations under different ground objects and meteorological conditions and automatically adapts to complex and changeable field conditions. For example, in the mapping of the urban-rural fringe area, there are diverse ground objects such as buildings, farmland, and forest land, and the meteorology varies in different seasons. The model gives the appropriate weights based on real-time data, reducing manual repeated debugging, improving both efficiency and ensuring the quality of the fused images, providing a reliable data basis for subsequent terrain modeling, etc.

[0039] 3. Fine point cloud processing ensures the high-precision construction of digital elevation models:

[0040] The method of LiDAR point cloud filtering to distinguish ground points from non-ground points is scientific and efficient. By comparing the slopes of points with adjacent points and combining the average slope of the initial grid with the threshold for judgment, it effectively separates non-ground points such as buildings and trees, and accurately extracts ground points for the construction of digital elevation models, avoiding terrain distortion caused by the interference of non-ground points. For example, in the area with high-rise buildings in the city, it accurately filters out the building point clouds and restores the real ground undulation, ensuring that the digital elevation model truthfully reflects the terrain elevation changes and providing accurate terrain data support for terrain analysis, engineering planning, etc.

[0041] Construct a digital elevation model by dynamically adjusting the grid size according to the standard deviation of the initial grid elevation, taking into account both accuracy and efficiency. Appropriately expand the grid in gently changing terrain areas to reduce data redundancy; refine the grid in complex terrain areas to accurately capture terrain details. For example, large grids in plain areas quickly outline the terrain contour, and small grids in mountainous and canyon areas meticulously depict the uneven terrain, enabling the digital elevation model to meet both the needs of macro planning and the accuracy requirements of micro engineering construction.

[0042] 4. Realistic 3D modeling enhances the terrain visualization effect:

[0043] Use adaptive fusion of remote sensing image textures and digital elevation models for 3D modeling. Texture mapping enables the model to have both accurate terrain geometry and rich feature appearances. For example, when simulating mountainous areas, the digital elevation model shapes the mountain undulations, and the fused image textures endow the mountains with vegetation and rock colors and textures. Combined with lighting rendering, a realistic 3D terrain model is generated, intuitively showing the terrain and feature distributions, assisting professionals such as urban planners and geological explorers to efficiently understand the on-site situation, and improving the scientificity and accuracy of decision-making. Brief Description of the Drawings

[0044] Figure 1 This is the schematic diagram of a high-precision terrain mapping method based on multi-source remote sensing data according to an embodiment of the present application.

[0045] Figure 2 This is the schematic diagram of a high-precision terrain mapping system based on multi-source remote sensing data according to an embodiment of the present application. Detailed Implementation Manner

[0046] Next, in combination with the drawings in the embodiments of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts fall within the protection scope of the present application.

[0047] As Figure 1 shown, a high-precision terrain mapping method based on multi-source remote sensing data includes the following steps:

[0048] Step s1: Obtain the optical remote sensing image, radar remote sensing image, LiDAR point cloud data, and meteorological index data of the target area. For example, select the WorldView-3 satellite to obtain the optical remote sensing image, whose panchromatic image resolution can reach 0.31 meters, and the multispectral image covers multiple bands. At the same time, use the SAR data of the Sentinel-1 satellite to obtain the radar remote sensing image to penetrate the clouds and vegetation that may exist in the mountainous area and obtain the general undulation information of the terrain. In addition, use an airborne LiDAR system to perform flight scanning to obtain the point cloud data of the mountainous area, ensuring that the elevation changes and detailed features of the terrain can be accurately measured, and perform data preprocessing on the optical remote sensing image, radar remote sensing image, and LiDAR point cloud data. The data preprocessing process includes geometric correction of the optical remote sensing image. Select obvious ground object markers (such as the peak of a mountain, the intersection of roads at the bottom of a valley, etc.) within the mountainous area as ground control points, and use the polynomial fitting method for geometric correction, and use the 6S atmospheric correction model for radiometric correction to eliminate the atmospheric influence and improve the image quality. For the radar remote sensing image, perform geometric correction according to its orbital parameters and imaging model, and perform calibration processing to convert the image gray value into the backscattering coefficient. After performing installation angle deviation correction and time delay calibration on the LiDAR point cloud data, use a slope-based filtering algorithm to separate ground points and non-ground points, remove the point cloud data of non-ground objects such as trees and buildings, and retain the accurate terrain point cloud information;

[0049] Step s2: Obtain the land cover types of the target area and the coverage area of each land cover type. According to the land cover types of the target area and the coverage area of each land cover type, perform weighted fusion on the optical remote sensing image and radar remote sensing image of the target area to obtain an adaptive fusion remote sensing image;

[0050] Step s3: Divide each point in the LiDAR point cloud data into ground points and non-ground points. According to each ground point in the LiDAR point cloud data, construct a digital elevation model. On the basis of the digital elevation model, combine the texture information of the adaptive fusion remote sensing image to perform three-dimensional modeling of the terrain and generate a three-dimensional terrain model.

[0051] It should be further noted that in the specific implementation process, the process of obtaining the land cover types of the target area and the coverage area of each land cover type includes:

[0052] Spectral feature extraction is performed on the preprocessed optical remote sensing image to obtain the spectral reflection data of different bands in the target area. Based on the spectral reflection data of different bands, the land cover types in the target area and the coverage area of each land cover type are obtained. By using the spectral reflectance differences of land covers in different bands, the land cover types and coverage areas can be identified more accurately. For example, in the near-infrared band, the reflectance of vegetation is significantly higher than that of other land covers. By analyzing the reflectance ratio of vegetation in the near-infrared and red bands (such as the normalized difference vegetation index NDVI), the vegetation coverage can be accurately estimated. The calculation formula is NDVI = (NIR - R) / (NIR + R), where NIR is the reflectance in the near-infrared band and R is the reflectance in the red band. When the NDVI value is close to 1, it indicates a high vegetation coverage; when the NDVI value is close to -1, it indicates a non-vegetated area.

[0053] It should be further noted that in the specific implementation process, according to the land cover types in the target area and the coverage area of each land cover type, the weighted fusion of the optical remote sensing image and the radar remote sensing image of the target area is carried out. The process of obtaining the adaptive fusion remote sensing image includes:

[0054] Construct an adaptive weight adjustment model, input the land cover types in the target area, the coverage area of each land cover type, and the meteorological index data into the adaptive weight adjustment model. According to the output of the adaptive weight adjustment model, obtain the first weight corresponding to the optical remote sensing image and the second weight corresponding to the radar remote sensing image. Perform weighted fusion on each pixel of the optical remote sensing image and the radar remote sensing image according to the first weight and the second weight to obtain the adaptive fusion remote sensing image.

[0055] The data-level fusion of the optical remote sensing image and the radar remote sensing image is carried out by using the weighted fusion method. Different weights are assigned to them according to their importance and reliability in topographic mapping. For example, in areas with less vegetation coverage, the optical image has a higher resolution and clear land cover details, so a higher weight can be given; while in areas with dense vegetation coverage or cloud and fog occlusion, the radar image can penetrate vegetation and cloud and fog, and a higher weight is given to the radar image at this time.

[0056] It should be further noted that in the specific implementation process, the process of performing weighted fusion on each pixel of the optical remote sensing image and the radar remote sensing image according to the first weight and the second weight to obtain the adaptive fusion remote sensing image includes:

[0057] Let the pixel value of the optical remote sensing image be I q , and the pixel value of the radar remote sensing image be I r , the first weight be ω q , and the second weight be ω r , and the pixel value I after weighted fusion fThe calculation formula for I is: f = ω q I q + ω r I r ;

[0058] For example, if the weight ω q of the optical remote sensing image = 0.6, and the weight ω r of the radar remote sensing image = 0.4, the pixel value I q of a certain pixel in the optical remote sensing image = 100, and the corresponding pixel value I r of the radar remote sensing image = 80, then the pixel value I f after weighted fusion = 92.

[0059] It should be further noted that in the specific implementation process, the process of constructing the adaptive weight adjustment model includes:

[0060] Select multiple test areas in the target area in advance. Different test areas include different land cover types and the coverage area of each land cover type. Collect the optical remote sensing images and radar remote sensing images of each test area under different meteorological index data conditions. Perform data preprocessing on the optical remote sensing images and radar remote sensing images. Use different preset weight combinations (such as the weight of the optical remote sensing image is 0, 0.2, 0.4, 0.6, 0.8, 1, and the corresponding weight of the radar remote sensing image is 1, 0.8, 0.6, 0.4, 0.2, 0) to perform data fusion on the preprocessed optical remote sensing images and radar remote sensing images of each test area under different meteorological index data conditions, and obtain the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions;

[0061] Obtain the field measurement data of each test area (such as the terrain point data measured by a total station). Compare the field measurement data of each test area with the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions, and obtain the accuracy error of the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions;

[0062] Build an adaptive weight adjustment model based on deep learning. Use the different land cover types included in each test area, the coverage area of each land cover type, and the accuracy error of the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions as the training set and the test set. Input the training set into the adaptive weight adjustment model for training until the loss function is trained stably, and save the model parameters. Test the adaptive weight adjustment model through the test set until it meets the preset requirements, and output the adaptive weight adjustment model;

[0063] Building an adaptive weight adjustment model based on deep learning is a complex process that involves multiple steps such as model selection, training, validation, and testing. In this embodiment, a deep belief network (DBN) is selected as the deep learning architecture, and the mean squared error loss function is chosen. It calculates the average of the squares of the differences between the predicted values and the true values. In the scenario of this adaptive weight adjustment model, the predicted values are the pixel values of the remotely sensed images (such as elevation values, land cover classification probabilities, etc.) after adaptive weight fusion, and the true values are the accurate values obtained through high-precision field measurements (such as topographic point data measured by total stations). Subsequently, the prepared training set is input into the selected deep learning model for training. During the training process, the weights are continuously updated through the backpropagation algorithm, causing the loss function to gradually decrease until it reaches a stable state. During this period, techniques such as Early Stopping are used to avoid overfitting. In addition to the basic training process, various parameters of the model are tuned through Grid Search at the same time. The parameters include the learning rate, batch size, regularization coefficient, etc.;

[0064] When the model training is completed and the parameters are adjusted, the final evaluation is carried out through the test set to obtain the evaluation results of the model. The evaluation results include classification metrics such as accuracy, recall rate, F1 score, etc. According to the evaluation results on the test set, it is judged whether the model meets the expected standards. If the requirements are met, the model parameters are saved and deployment is prepared; if not, it is necessary to return to a previous stage to re-examine issues such as data quality, model structure, or training strategy.

[0065] It should be further noted that in the specific implementation process, the process of dividing each point in the LiDAR point cloud data into ground points and non-ground points includes:

[0066] Obtain the slope between each point and its adjacent points in the LiDAR point cloud data, preset the initial grid size of the target area, divide the target area into several identical initial grids according to the initial grid size, construct a two-dimensional coordinate system for the target area, set the starting point (Xmin, Ymin) of the target area, and divide it successively according to the selected grid size. For example, for a square grid, if the side length of the grid is L and the lower left corner coordinates of the first grid are the starting point (Xmin, Ymin), then the lower left corner coordinates of the second grid are (Xmin + L, Ymin), and the lower left corner coordinates of the third grid are (Xmin, Ymin + L), and so on until the entire target area is covered. At the same time, consider the boundary conditions during the division process to ensure that the grids in the last row and the last column can completely cover the boundary part of the target area; obtain the average slope within each initial grid, compare the slope between each point and its adjacent points within each initial grid with the average slope within each initial grid, obtain the slope difference of each point within each initial grid, compare the slope difference of each point with the preset slope difference threshold, mark the points with a slope difference greater than the slope difference threshold as non-ground points, and mark the points with a slope difference less than or equal to the slope difference threshold as ground points.

[0067] It should be further noted that in the specific implementation process, the process of constructing a digital elevation model based on each ground point in the LiDAR point cloud data includes:

[0068] Obtain the elevation of the ground points within each initial grid, perform statistical analysis on the elevation of the ground points within each initial grid, obtain the elevation standard deviation within each initial grid, construct a grid size comparison table, where the grid size comparison table includes the grid sizes corresponding to different elevation standard deviations. It should be further noted that the initial grid size is greater than several grid sizes included in the grid size comparison table. Obtain the new grid size corresponding to each initial grid according to the elevation standard deviation within each initial grid and the grid size comparison table, divide each initial grid into several identical new grids according to the new grid size corresponding to each initial grid. If the new grid size is the same as the initial grid size, no division of the initial grid is performed, and the initial grid is marked as a new grid. Construct a digital elevation model of the new grid according to each new grid and the elevation of each point within each new grid.

[0069] It should be further noted that in the specific implementation process, on the basis of the digital elevation model, the process of generating a three-dimensional terrain model by combining the texture information of the adaptive fusion remote sensing image for three-dimensional terrain modeling includes:

[0070] Constructing the coordinate matching relationship between the digital elevation model and the adaptive fusion remote sensing image. For the adaptive fusion remote sensing image and the digital elevation model after data preprocessing, the coordinate unification is achieved by determining their reference points (such as the corresponding relationship between the four corner points of the image and the boundary points of the digital elevation model) and adjusting the resolution differences (such as the comparison between the actual ground size represented by each pixel of the image and the grid size of the digital elevation model). Texture sampling is performed on the adaptive fusion remote sensing image, and the sampled texture is attached to the corresponding position on the surface of the digital elevation model according to the coordinate matching relationship. For example, for the position of a mountaintop in the digital elevation model, the texture of the corresponding mountaintop area is found in the optical image, and its color and pattern information are given to the geometric surface of the mountaintop in the digital elevation model, so that the mountaintop presents the real appearance seen in the optical image in the three-dimensional model, such as the color of the rock and the form of vegetation coverage, etc., to generate a three-dimensional terrain model.

[0071] As Figure 2 shown, a high-precision terrain mapping system based on multi-source remote sensing data includes a cloud, and the cloud is communicatively connected to a data acquisition module, an image feature fusion module, and a three-dimensional modeling module;

[0072] The data acquisition module is used to obtain the optical remote sensing image, radar remote sensing image, LiDAR point cloud data, and meteorological index data of the target area, and perform data preprocessing on the optical remote sensing image, radar remote sensing image, and LiDAR point cloud data;

[0073] The image feature fusion module is used to obtain the land cover types of the target area and the coverage area of each land cover type, and perform weighted fusion on the optical remote sensing image and the radar remote sensing image of the target area according to the land cover types of the target area and the coverage area of each land cover type to obtain an adaptive fusion remote sensing image;

[0074] The three-dimensional modeling module is used to divide each point in the LiDAR point cloud data into ground points and non-ground points, construct a digital elevation model according to each ground point in the LiDAR point cloud data, and perform three-dimensional modeling of the terrain on the basis of the digital elevation model in combination with the texture information of the adaptive fusion remote sensing image to generate a three-dimensional terrain model.

[0075] The above embodiments are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A high-precision terrain mapping method based on multi-source remote sensing data, characterized in that: The following steps are involved: Step s1: obtaining optical remote sensing images, radar remote sensing images, LiDAR point cloud data and meteorological index data of the target area, and performing data preprocessing on the optical remote sensing images, radar remote sensing images and LiDAR point cloud data; Step s2: obtaining the type of ground objects in the target area and the coverage area of ​​each ground object type, and performing weighted fusion on the optical remote sensing image and the radar remote sensing image of the target area according to the type of ground objects in the target area and the coverage area of ​​each ground object type to obtain an adaptive fused remote sensing image; Step s3: Divide each point in the LiDAR point cloud data into ground points and non-ground points, and build a digital elevation model based on each ground point in the LiDAR point cloud data. On the basis of the digital elevation model, combine the texture information of the adaptive fusion remote sensing image to perform three-dimensional modeling of the terrain and generate a three-dimensional terrain model.

2. The high-precision terrain mapping method based on multi-source remote sensing data according to claim 1, characterized in that: The process of obtaining the object types in the target area and the coverage area of ​​each object type includes: The spectral features of the optical remote sensing images after data preprocessing are extracted to obtain the spectral reflectance data of different bands in the target area. The ground object type in the target area and the coverage area of ​​each ground object type are obtained based on the spectral reflectance data of different bands.

3. The high-precision terrain mapping method based on multi-source remote sensing data according to claim 2 is characterized in that: According to the type of objects in the target area and the coverage area of ​​each object type, the optical remote sensing image and the radar remote sensing image of the target area are weighted fused to obtain the adaptive fused remote sensing image. The process includes: An adaptive weight adjustment model is constructed, and the land feature type of the target area, the coverage area of ​​each land feature type, and meteorological index data are input into the adaptive weight adjustment model. The first weight corresponding to the optical remote sensing image and the second weight corresponding to the radar remote sensing image are output according to the adaptive weight adjustment model. Each pixel point in the optical remote sensing image and the radar remote sensing image is weightedly fused according to the first weight and the second weight to obtain an adaptive fused remote sensing image.

4. The high-precision terrain mapping method based on multi-source remote sensing data according to claim 3 is characterized in that: The process of building an adaptive weight adjustment model includes: Select multiple test areas in the target area in advance, where different test areas include different types of land objects and coverage areas of each type of land objects, collect optical remote sensing images and radar remote sensing images of each test area under different meteorological index data conditions, perform data preprocessing on the optical remote sensing images and radar remote sensing images, use different preset weight combinations to perform data fusion of the optical remote sensing images and radar remote sensing images of each test area after data preprocessing under different meteorological index data conditions, and obtain adaptive fused remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions; Obtain the field measurement data of each test area, compare the field measurement data of each test area with the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions, and obtain the accuracy error of the adaptive fusion remote sensing images of each test area under different weight combination conditions under different meteorological index data conditions; An adaptive weight adjustment model is constructed based on deep learning. The different types of land features in each test area, the coverage area of ​​each land feature type, and the accuracy error of the adaptively fused remote sensing images under different weight combinations of each test area under different meteorological indicator data conditions are used as training sets and test sets. The training set is input into the adaptive weight adjustment model for training until the loss function training is stable, and the model parameters are saved. The adaptive weight adjustment model is tested by the test set until it meets the preset requirements, and the adaptive weight adjustment model is output.

5. The high-precision terrain mapping method based on multi-source remote sensing data according to claim 4 is characterized in that: The process of classifying each point in the LiDAR point cloud data into ground points and non-ground points includes: Obtain the slope between each point and adjacent points in the LiDAR point cloud data, preset the initial grid size of the target area, divide the target area into several identical initial grids according to the initial grid size, obtain the average slope in each initial grid, compare the slope between each point and adjacent points in each initial grid with the average slope in each initial grid, obtain the slope difference of each point in each initial grid, compare the slope difference of each point with the preset slope difference threshold, mark the points with slope difference greater than the slope difference threshold as non-ground points, and mark the points with slope difference less than or equal to the slope difference threshold as ground points.

6. The high-precision terrain mapping method based on multi-source remote sensing data according to claim 5, characterized in that: Based on the ground points in the LiDAR point cloud data, the process of building a digital elevation model includes: The elevations of the ground points in each initial grid are obtained, and statistical analysis is performed on the elevations of the ground points in each initial grid, and the elevation standard deviations in each initial grid are obtained. A grid size comparison table is constructed, wherein the grid size comparison table includes grid sizes corresponding to different elevation standard deviations. A new grid size corresponding to each initial grid is obtained according to the elevation standard deviations in each initial grid and the grid size comparison table. According to the new grid size corresponding to each initial grid, each initial grid is divided into a number of identical new grids. If the new grid size is consistent with the initial grid size, the initial grid is not divided, and the initial grid is marked as a new grid. According to the elevations of each new grid and each point in each new grid, a digital elevation model of the new grid is constructed.

7. The high-precision terrain mapping method based on multi-source remote sensing data according to claim 6, characterized in that: Based on the digital elevation model, the three-dimensional terrain modeling is carried out by combining the texture information of the adaptive fusion remote sensing image. The process of generating the three-dimensional terrain model includes: A coordinate matching relationship between the digital elevation model and the adaptive fused remote sensing image is constructed, texture sampling is performed on the adaptive fused remote sensing image, and the sampled texture is attached to the corresponding position on the surface of the digital elevation model according to the coordinate matching relationship.

8. A high-precision terrain mapping system based on multi-source remote sensing data, specifically applied to a high-precision terrain mapping method based on multi-source remote sensing data as claimed in any one of claims 1 to 7, characterized in that: It includes a cloud, wherein the cloud is communicatively connected with a data acquisition module, an image feature fusion module and a three-dimensional modeling module; The data acquisition module is used to obtain optical remote sensing images, radar remote sensing images, LiDAR point cloud data and meteorological index data of the target area, and perform data preprocessing on the optical remote sensing images, radar remote sensing images and LiDAR point cloud data; The image feature fusion module is used to obtain the type of ground objects in the target area and the coverage area of ​​each ground object type. According to the type of ground objects in the target area and the coverage area of ​​each ground object type, the optical remote sensing image and the radar remote sensing image of the target area are weightedly fused to obtain an adaptive fused remote sensing image. The 3D modeling module is used to divide each point in the LiDAR point cloud data into ground points and non-ground points, and build a digital elevation model based on each ground point in the LiDAR point cloud data. On the basis of the digital elevation model, the 3D modeling of the terrain is performed in combination with the texture information of the adaptive fusion remote sensing image to generate a 3D terrain model.

Images