Three-dimensional display method applied to remote sensing data

Abstract

The invention discloses a three-dimensional display method applied to remote sensing data. The method comprises the following steps: acquiring multi-source remote sensing data; carrying out space alignment and time alignment on the multi-source remote sensing data; performing denoising and geometric correction on the aligned multi-source remote sensing data; generating an initial grid model according to the data after geometric correction; carrying out segmentation and topological optimization on the initial grid model to generate an optimized grid model; and performing rendering processing on the optimized grid model to obtain a visual three-dimensional model so as to effectively solve the problem of space-time isomerism of the multi-source remote sensing data and realize the effect of converting the original data into the high-quality three-dimensional model.

Description

Technical Field

[0001] The present invention belongs to the technical field of remote sensing information processing, and particularly relates to a three-dimensional display method applied to remote sensing data. Background Art

[0002] In the three-dimensional display method of remote sensing data, the core problem of the process display technology lies in how to dynamically present each stage of data processing and analysis without destroying the continuity of the three-dimensional model. First, the acquisition of remote sensing data usually involves the fusion of multi-source heterogeneous data, which may have large differences in spatial resolution and time series, resulting in difficulties in data alignment during the three-dimensional modeling process, thereby affecting the fluency of the process display. Secondly, although the noise filtering and geometric correction operations in the data preprocessing stage can improve the quality of the data, in the three-dimensional display, the real-time feedback of these operations may cause local distortion or time delay of the model, affecting the user's perception of the overall process.

[0003] In the three-dimensional modeling stage, how to achieve seamless connection of different processing steps while maintaining the model accuracy is another key problem. For example, in the process of generating a mesh model from point cloud data, operations such as mesh segmentation and topology optimization may introduce additional computational burdens, resulting in a decrease in the model rendering efficiency, thereby affecting the real-time performance of the process display. Finally, for the lighting and texture mapping processing in the three-dimensional rendering stage, how to avoid the loss of model details caused by excessive optimization while ensuring the visual effect is also a technical difficulty that needs to be solved. These problems need to be deeply explored from the perspectives of algorithms and computational resource allocation to ensure the integrity of the process display and the fluency of the user experience. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a three-dimensional display method applied to remote sensing data to solve the problems existing in the above prior art.

[0005] To achieve the above object, the present invention provides a three-dimensional display method applied to remote sensing data, including:

[0006] Obtain multi-source remote sensing data; perform spatial alignment and temporal alignment on the multi-source remote sensing data;

[0007] Denoise and geometrically correct the aligned multi-source remote sensing data;

[0008] Generate an initial mesh model according to the geometrically corrected data;

[0009] Segment and topologically optimize the initial mesh model to generate an optimized mesh model;

[0010] Render the optimized mesh model to obtain a visualized three-dimensional model.

[0011] Optionally, the multi-source remote sensing data includes optical remote sensing data, radar data, and hyperspectral data.

[0012] Optionally, the process of spatially aligning the multi-source remote sensing data includes:

[0013] Obtain the spatial resolution and time series information of the multi-source remote sensing data, and align the multi-source remote sensing data with different spatial resolutions through a spatial interpolation algorithm or resampling method to obtain a spatially aligned dataset.

[0014] Optionally, the process of obtaining subsequent multi-source remote sensing data includes:

[0015] Denoise the time series information, and synchronize the denoised time series information through a time series alignment algorithm to obtain the corresponding multi-source remote sensing data.

[0016] Optionally, the process of denoising and geometric correction of the aligned multi-source remote sensing data includes:

[0017] Denoise the aligned multi-source remote sensing data through a Gaussian filter, perform geometric correction on the denoised data through a geometric correction model, and verify and review the geometric correction results to obtain geometrically corrected data.

[0018] Optionally, the process of generating the initial mesh model includes:

[0019] Obtain the three-dimensional coordinate information of the geometrically corrected data, organize and manage the point cloud through an octree, construct a mesh model through a triangulation method, and optimize the constructed mesh model through a smoothing optimization method and a LOD method to obtain an initial mesh model.

[0020] Optionally, the process of segmenting and topologically optimizing the initial mesh model includes:

[0021] Extract and classify the features of different meshes of the initial mesh model to obtain a classification result. Segment the initial mesh model according to the classification result to obtain sub-meshes. Obtain the topological structure information of the sub-meshes. According to the topological structure information of the sub-meshes, adjust and combine the sub-meshes through a topological optimization algorithm to obtain an optimized mesh model.

[0022] Optionally, the process of obtaining the visualized three-dimensional model includes:

[0023] Perform ray reflection calculations on the optimized mesh model through a lighting model to obtain the lighting intensity analysis of each point on the mesh surface. According to the lighting intensity distribution results, use the texture mapping algorithm to map the preset texture information onto the mesh surface to generate a preliminary rendered 3D model. Extract the time series information from the preliminary rendered 3D model, and use the frequency domain analysis method to analyze the main frequency band characteristics of the time series. If the main frequency band characteristics meet the preset lighting effect threshold, output the final visual 3D model; if not, adjust the lighting model parameters and recalculate the ray reflection.

[0024] Compared with the prior art, the present invention has the following advantages and technical effects:

[0025] The present invention discloses a 3D display method applied to remote sensing data. The method first obtains multi-source remote sensing data, extracts the spatial resolution and time series information, realizes spatial alignment through the interpolation algorithm, and performs synchronous processing using the time series alignment algorithm. Subsequently, a Gaussian filter is used to eliminate high-frequency noise and geometric correction is performed. On this basis, the present invention uses a point cloud generation algorithm to convert the data into an initial mesh model, and generates an optimized mesh model through mesh segmentation and topological optimization. Finally, a lighting model and a texture mapping algorithm are applied for rendering to generate a 3D model with visual effects. This method effectively solves the spatio-temporal heterogeneity problem of multi-source remote sensing data, realizes the transformation from raw data to a high-quality 3D model, and provides a new way for the in-depth analysis and visualization of remote sensing data. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0027] Figure 1 is a flowchart of the 3D display method applied to remote sensing data according to an embodiment of the present invention;

[0028] Figure 2 is a structural schematic diagram of the 3D display system applied to remote sensing data according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0029] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0030] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0031] As Figure 1 , a three-dimensional display method and system for remote sensing data in this embodiment may specifically include:

[0032] S101. Obtain multi-source remote sensing data sources, and extract the spatial resolution and time series information of each data source.

[0033] Obtain multiple remote sensing data sources, extract the spatial resolution and time series information of each data source, analyze and process the extracted information, and generate a data set that meets business requirements. Adopt multi-source remote sensing data acquisition technology to obtain remote sensing data from different platforms, extract the spatial resolution and time series information of each data source, and generate a preliminary data set.

[0034] Exemplarily, multi-source remote sensing data acquisition technology is an important foundation in the field of remote sensing. Taking the monitoring of land cover change as an example, optical, radar, and hyperspectral data can be obtained simultaneously. Optical data such as Landsat series satellite images has a spatial resolution of 30 meters and a revisit period of 16 days; radar data such as Sentinel-1 satellite images has a resolution of 10 meters and a revisit period of 6 days; hyperspectral data such as HJ-1A satellite images has a resolution of 100 meters and a revisit period of 4 days. These data sources have their own characteristics and jointly constitute a preliminary data set.

[0035] S102. According to the spatial resolution and time series information, use an interpolation algorithm to achieve spatial alignment of different data sources, and generate a data set with a unified spatial resolution.

[0036] Obtain multiple remote sensing data sources, extract the spatial resolution and time series information of each data source, and generate a preliminary data set. According to the preliminary data set, use a spatial interpolation algorithm to align data with different spatial resolutions, and generate an aligned data set.

[0037] Exemplarily, the acquisition of remote sensing data sources is the basis for multi-source remote sensing data fusion. For example, different types of remote sensing images can be obtained from platforms such as optical satellites, radar satellites, and hyperspectral satellites. Each data source has its unique spatial resolution and time series characteristics. Optical satellites like the Landsat series can provide multi-spectral data with a resolution of 30 meters, and images are acquired every 16 days. Radar satellites such as Sentinel-1 can provide all-weather imaging data with a resolution of 10 meters, and the revisit period is 6 days. Spatial interpolation is a key technology for solving the problem of aligning data with different resolutions. Taking the fusion of optical and radar data as an example, the bilinear interpolation method can be used to resample the 30-meter resolution optical data to 10 meters to align it spatially with the radar data. This method can improve the spatial resolution while maintaining the characteristics of the original data, laying a foundation for subsequent analysis. The spatial alignment algorithm aims to solve the problem of inconsistent spatial resolutions of different data sources. Through methods such as interpolation or resampling, data with different resolutions can be unified onto the same spatial grid. In remote sensing image fusion, the spatial alignment of high-resolution panchromatic images and low-resolution multi-spectral images is a key step in achieving high-quality fusion. The bilinear interpolation method can be used to resample the low-resolution image to high resolution, or super-resolution reconstruction technology can be used to improve the spatial details of the low-resolution image.

[0038] S103. On the basis of spatial alignment, use the time series alignment algorithm to synchronize the time series of different data sources to obtain a multi-source data set with time alignment.

[0039] Use the interpolation algorithm to align the spatial resolutions of the multi-source data set to generate a spatially aligned data set. For the spatially aligned data set, use the denoising algorithm to denoise the time series information to obtain a denoised data set. For the possible deviations in the time series information in the denoised data set, use the time series alignment algorithm for synchronization processing to obtain a multi-source data set with time alignment.

[0040] Exemplarily, the spatial resolution alignment of the multi-source data set is a key step in remote sensing data processing. Taking the data of Landsat and MODIS satellites as an example, the former has a spatial resolution of 30 meters, and the latter is 250 meters. Spatial correspondence or the bilinear interpolation algorithm can be used to resample the MODIS data to a resolution of 30 meters to achieve spatial alignment. This process helps to directly compare the information obtained by different sensors in subsequent analysis. Denoising processing is crucial for improving data quality. Taking the monitoring of crop growth in remote sensing data as an example, the NDVI time series is often affected by clouds and generates outliers. The Savitzky-Golay filtering algorithm can effectively remove these noises and retain the true trend of vegetation growth. This method smooths the data through local polynomial fitting and can retain the higher-order moments of the signal while removing the noise.

[0041] Time series alignment is the key to ensuring the temporal consistency of multi-source data. Taking different satellite observation data as an example, due to differences in orbits and observation times, the observation times of the same area may be inconsistent. The dynamic time warping algorithm can be used to achieve the synchronization of time series. This method aligns time series with different lengths and sampling rates by finding the best correspondence, so that multi-source data is comparable in the time dimension.

[0042] S104. Use a Gaussian filter to filter the noise of the multi-source data set after time alignment to eliminate high-frequency noise interference in the data.

[0043] Use a Gaussian filter to filter the noise of the multi-source data set after time alignment to eliminate high-frequency noise interference.

[0044] Exemplarily, the Gaussian filter is a commonly used data smoothing method that can effectively eliminate high-frequency noise interference. In the processing of multi-source data sets, Gaussian filtering can retain the main features of the data while filtering out unnecessary detail fluctuations. For example, in meteorological data analysis, applying Gaussian filtering to the temperature time series can eliminate short-term fluctuations and highlight long-term change trends. Analyzing the denoised data through frequency domain analysis methods can reveal the periodic characteristics of time series data and analyze and verify its denoising effect. Through Fourier transform, the time-domain signal can be converted into a frequency-domain representation to identify the main frequency band characteristics. In the processing of seismic data in remote sensing data, for example, frequency domain analysis can help identify the main frequency components of seismic waves, providing an important basis for subsequent earthquake early warning and risk assessment. The time series alignment algorithm solves the problem of time synchronization between different data sources. There may be time deviations in remote sensing data from different sources. Algorithms such as dynamic time warping (DTW) can be used to align time series, enabling more accurate remote sensing data analysis.

[0045] S105. According to the pre-established geometric correction model, perform geometric correction on the data after noise filtering to obtain a multi-source data set after geometric correction.

[0046] In the geometric correction stage, based on the pre-built model, after ensuring the applicability of the model through coordinate system verification and DEM resolution matching, multi-source differential processing is implemented: high-resolution optical images use the RPC model and bicubic convolution resampling, combined with the DEM terrain compensation amount to correct terrain distortion; medium-resolution data uses second-order polynomial coordinate transformation and the LM algorithm to optimize parameters; SAR data applies the range-Doppler model to correct side-looking deviation, supplemented by thin plate spline functions to correct local deformation. After triple verification of RMS error, mutual information value, and Fourier frequency domain consistency, a geometrically refined data set supporting 12 types of sensors is generated, with high processing efficiency, meeting the 15-meter benchmark registration accuracy, and automatically marking and rechecking some areas.

[0047] Exemplarily, in the geometric correction stage, first, based on a pre-established geometric correction model (such as a second-order polynomial model integrating DEM data or an RPC orthorectification model), the applicability of the model is ensured through coordinate system consistency verification (error ≤ 0.1 pixel) and DEM resolution matching detection (resolution ratio ≤ 3 times). Subsequently, differential correction is implemented according to the characteristics of multi-source data: for high-resolution optical images, the RPC model and bicubic convolution resampling are used, and the terrain compensation amount is calculated in combination with DEM elevation data to correct the terrain undulation distortion; for medium-resolution data, a second-order polynomial coordinate transformation is performed, and the LM algorithm is used to iteratively optimize the coefficients; for SAR data, the range-Doppler model is applied to compensate for the side-looking geometric deviation, and at the same time, the thin plate spline function is used to correct local deformation. After correction, triple verification is carried out through RMS error analysis (plain ≤ 0.3 pixel / mountain area ≤ 0.8 pixel), mutual information value calculation (MI ≥ 0.85), and Fourier frequency domain consistency verification (cross-power spectrum mean > 0.9). Finally, a geometrically refined correction data set is generated to achieve cross-platform alignment of 12 types of sensor data. The processing efficiency reaches 1000×1000 pixels / 120 seconds (GPU acceleration), meeting the 15-meter registration accuracy of Google Earth reference data, and automatically marking areas with cloud cover > 30% for manual review.

[0048] S106. Based on the geometrically corrected data, use a point cloud generation algorithm to convert the point cloud data into an initial mesh model.

[0049] Based on the geometrically corrected remote sensing image, obtain the three-dimensional coordinate information of ground objects through feature extraction to generate a point cloud. Organize and manage the point cloud data using an octree, and use the Delaunay triangulation method to construct an initial mesh model. Improve the mesh quality through optimization means such as Laplacian smoothing, and generate meshes with different levels of detail as needed in combination with the LOD technology. Finally, a high-quality three-dimensional mesh model reflecting the true surface morphology is obtained.

[0050] Exemplarily, based on the geometrically corrected data, a point cloud generation algorithm is used to convert the point cloud data into an initial mesh model. Specifically, first, feature extraction is performed on the geometrically corrected multi-source remote sensing images to obtain the three-dimensional coordinate information of the ground objects, forming a dense point cloud data set. Geometric correction of remote sensing images refers to eliminating geometric distortions caused by factors such as sensor attitude and terrain undulation during the imaging process through a series of mathematical transformations, so that the images reach a unified reference framework in space. Common geometric correction methods include affine transformation, polynomial fitting, etc. After geometric correction, images acquired at different times and from different angles can be better registered together, laying a foundation for subsequent processing. Next, extracting the three-dimensional coordinate information of the ground objects from the geometrically corrected images is the key step in generating the point cloud. Usually, stereo matching technology is adopted, and the depth value is calculated using the parallax of the same ground object under different perspectives, and then the three-dimensional coordinates are obtained. To improve the accuracy and density, multiple images can be fused for joint matching, and sub-pixel interpolation and other means can be used to optimize the results. The finally output point cloud data set contains a large number of discrete surface sampling points of the ground objects, reflecting the morphological characteristics of the ground surface. However, the original point cloud often has problems such as noise and holes, and needs to be further preprocessed before being used for mesh construction. An effective method is to organize the point cloud based on the octree data structure. The octree is a hierarchical spatial index structure that recursively divides the entire space into eight sub-regions, and each node corresponds to a cubic spatial range. Through the top-down segmentation process, functions such as spatial query, duplicate removal, and filtering of the point cloud can be efficiently realized, significantly improving the subsequent processing efficiency. After having high-quality point cloud data, the construction of the triangular mesh can be started. The Delaunay triangulation method is a commonly used mesh generation algorithm, and its core idea is to maximize the minimum angle of the triangular elements as much as possible under the premise of satisfying the empty circle condition. The generated triangular mesh has better shape quality and can better approximate the true contour of the ground objects. In actual operation, a random incremental method can be used to quickly establish an initial triangular mesh skeleton, and then the remaining points are gradually inserted and the topological relationship is dynamically adjusted until all points are included. The initially generated triangular mesh may have some defects, such as sharp angles, too long or too short edges, etc., affecting the subsequent application effect. Therefore, grid optimization technology needs to be introduced to improve it. Laplacian smoothing is a simple and effective optimization strategy. It iteratively updates the position of each vertex to make it tend to the weighted average of the coordinates of the neighboring vertices, thereby achieving the purpose of smoothing the mesh. Of course, to avoid losing details due to excessive smoothing, the weight coefficient and the number of iterations should be reasonably set, and other constraint conditions (such as boundary preservation) should be combined when necessary to act together. Finally, considering that different application scenarios may have different requirements for the mesh resolution, the LOD (Level of Detail) technology can be used to generate mesh models with different levels of detail as needed.The basic idea is to dynamically adjust the refinement level of the grid according to parameters such as the viewing distance and viewing angle: when the user is far from the target, a rough grid is displayed to save resources; when approaching, the details are gradually increased to provide a more refined performance. This not only helps to balance the contradiction between performance and accuracy, but also enhances the realism of the interaction experience.

[0051] In summary, starting from the geometrically corrected remote sensing image, through a series of processes such as feature extraction, point cloud generation, grid construction and optimization, a three-dimensional grid model reflecting the true surface morphology can be finally obtained.

[0052] S107. For the initial grid model, use a grid segmentation algorithm to decompose the model into multiple sub-grids, and perform topological optimization among the sub-grids to generate an optimized grid model.

[0053] Use a grid segmentation algorithm to decompose the initial grid model into multiple sub-grids, and obtain the topological structure information of the sub-grids. According to the topological structure information of the sub-grids, use a topological optimization algorithm to adjust the geometric connection relationship of the sub-grids to obtain topologically optimized sub-grids. Reconstruct the topologically optimized sub-grids into an optimized grid model, and extract the geometric features of the optimized grid model. Use a frequency domain analysis method to extract time series information from the optimized grid model to obtain the main frequency band features of the time series. According to the main frequency band features, use a feature extraction algorithm to extract key feature values from the time series. Use a clustering algorithm to perform clustering analysis on the key feature values to obtain a preliminary classification result. If the preliminary classification result meets the preset threshold, output the final classification result; if not, readjust the parameters of the clustering algorithm.

[0054] Exemplarily, the grid segmentation algorithm is an important technology for processing three-dimensional models, which can decompose a complex grid model into more manageable sub-grids. For example, for a three-dimensional model constructed from complex remote sensing data, it can be segmented into sub-parts such as cities, rural areas, and cultivated land. This segmentation not only facilitates subsequent processing but also retains the topological structure information of the original model. The topological optimization algorithm further adjusts the connection relationship between the sub-grids to improve the model quality. For example, when optimizing a three-dimensional model, the connection between cities and rural areas may be adjusted to ensure they fit closely and eliminate possible gaps or overlaps. This process not only improves the geometric accuracy of the model but also enhances its physical property performance. Reconstructing the optimized grid model is the process of recombining the optimized sub-grids. In the example of a three-dimensional model, this is equivalent to reassembling the optimized components such as cities, rural areas, and cultivated land into a complete three-dimensional model. The purpose of this step is to obtain a model with more accurate geometric features and a more reasonable topological structure.

[0055] Extracting time series information from the optimized model is for analyzing the changes of the model in the time dimension. Taking the terrain model as an example, the changing trend of the terrain over time can be analyzed. Through frequency domain analysis, the main change cycles can be identified, such as seasonal changes or annual changes, etc. The feature extraction algorithm extracts key feature values from the time series. In the example of the terrain model, this may include the maximum value, minimum value, average value of the terrain height, as well as the change rate, etc. These feature values can generally describe the overall characteristics and changing trends of the terrain. Cluster analysis is the process of grouping data points with similar features. In terrain analysis, different regions may be classified according to the extracted feature values. For example, flat areas, hilly areas, and mountainous areas may be separated. The quality of the preliminary classification results directly affects the accuracy of subsequent analysis. If the preliminary classification results do not meet the preset threshold, the parameters of the clustering algorithm need to be adjusted. This may involve changing the number of cluster centers, adjusting the distance measurement method, etc. For example, if it is found that the classification of mountainous areas and hilly areas is not clear enough, the number of clusters may need to be increased or the feature weights may need to be adjusted to better distinguish these two types of terrain. The purpose of this series of steps is to extract valuable information from the complex three-dimensional model and understand and organize this information through classification. By optimizing the grid model, extracting time series features, and performing cluster analysis, the rich information contained in the three-dimensional model can be better understood and utilized, providing a solid foundation for subsequent decision-making and analysis.

[0056] On the other hand, subsequent clustering analysis can also be carried out. Feature extraction is the core link of data analysis. For multispectral data, the normalized difference vegetation index (NDVI) can be calculated as an indicator of vegetation coverage. For radar data, the backscatter coefficient can be used to characterize surface roughness and water content. By extracting these features, the physical properties and change processes of ground objects can be better described. Clustering analysis helps to identify the internal structure in the data. Taking land use classification as an example, the K-means algorithm can be used to cluster the extracted features. Suppose the data is divided into five categories: water body, urban area, forest, farmland, and bare land. The algorithm will assign pixels to the nearest category according to the similarity of features, thus forming a preliminary classification result. Data fusion is a key step in integrating multi-source information. The decision-level fusion method, such as Dempster-Shafer evidence theory, can be used to synthesize the classification results of optical and radar data. This method takes into account the reliability and uncertainty of different data sources and can generate a more accurate fusion result. The final classification process can adopt the random forest algorithm. This algorithm constructs multiple decision trees and comprehensively considers various features to finally classify the fused data. For example, in urban expansion monitoring, the changes in multi-temporal NDVI and radar backscatter coefficient can be used to identify newly added construction land, so as to realize the dynamic monitoring of urban development. This series of processing steps not only improves the quality and interpretability of remote sensing data, but also enhances the accuracy and reliability of the classification results. Through the above series of processing steps, different sub-grids are classified.

[0057] S108. Based on the optimized grid model, use the lighting model and texture mapping algorithm for rendering processing to generate a three-dimensional model with visual effects.

[0058] Use the lighting model to calculate the light reflection of the optimized grid model to obtain the light intensity distribution of each point on the grid surface. According to the light intensity distribution result, use the texture mapping algorithm to map the preset texture information to the grid surface to generate a preliminarily rendered three-dimensional model. Extract the time series information from the preliminarily rendered three-dimensional model, and use the frequency domain analysis method to analyze the main frequency band characteristics of the time series. If the main frequency band characteristics meet the preset lighting effect threshold, output the final rendering result; if not, adjust the lighting model parameters to recalculate the light reflection. According to the adjusted light intensity distribution, use the texture mapping algorithm to remap the texture information to generate a new three-dimensional rendering model. Extract the time series information from the new three-dimensional rendering model, and use the feature extraction algorithm to obtain the key feature values of the rendering effect. Use the clustering algorithm to perform clustering analysis on the key feature values to determine the final classification result of the rendering effect.

[0059] Exemplarily, a lighting model is a mathematical model in computer graphics that simulates the interaction between light and the surface of an object. Taking the Phong lighting model as an example, it considers three components: ambient light, diffuse reflection, and specular reflection. At each vertex of the mesh model, the combined effect of these three components can be calculated to obtain the lighting intensity at that point. For example, for a sphere model, the part of the vertex facing the light source will exhibit a higher lighting intensity, while the backlit side will be relatively darker. Texture mapping is a technique for applying a 2D image to the surface of a 3D object. For the optimized mesh model, the mapping method of the texture can be adjusted according to the lighting intensity distribution result. For example, for a mountain model, the visibility of the texture can be enhanced in areas with stronger lighting, while the contrast of the texture can be slightly reduced in shadow areas to simulate the light and shadow effects in the real world. Extracting time series information from the rendered model can help analyze the lighting changes in a dynamic scene. For example, in an animation of a sunrise scene, the lighting intensity at a fixed point over time can be recorded. By performing frequency domain analysis on this time series, a major low-frequency component corresponding to the gradual rise of the sun and some high-frequency components that may represent rapid lighting changes caused by clouds or other objects may be found. If the main frequency band characteristics do not meet the preset lighting effect threshold, the lighting model parameters need to be adjusted. For example, in an indoor scene, if the analysis result shows that the lighting change is too smooth, the intensity of the point light source may need to be increased or its position adjusted to create a more rich contrast between light and dark. After adjustment, the light reflection will be recalculated, and it may be found that the shadows in areas such as the corners become more obvious, adding a sense of depth to the scene. When remapping the texture information, the details of the texture can be adjusted according to the new lighting distribution. For example, for a rural area, the specular effect of the texture can be enhanced in the highlight area, while the roughness of the texture can be slightly increased in the dark part to simulate the microscopic changes on the surface of rural structures under different lighting conditions. This can make the rendering result more realistic and enhance the expressiveness of the material. Extracting time series information from the new 3D rendered model can help evaluate the dynamic characteristics of the rendering effect. For example, in a scene of a sparkling water surface, the temporal variation of the light intensity reflected by the water surface can be analyzed. Through the feature extraction algorithm, a periodic fluctuation pattern may be found, which represents the motion characteristics of the water waves. At the same time, some irregular high-frequency changes may also be found, which may represent the fine ripples caused by the wind blowing on the water surface. Finally, performing clustering analysis on the extracted key feature values can help classify different rendering effects. For example, it may be found that one class of feature values corresponds to the rendering effect of a calm water surface, while another class corresponds to the effect of a rough and turbulent water surface. Through this classification, the influence of the rendering parameters on the final effect can be better understood and controlled, so that the rendering settings can be adjusted more precisely to achieve the desired visual effect.

[0060] Such as Figure 2As shown in the figure, the present invention provides a three-dimensional display system for remote sensing data, mainly including:

[0061] A multi-source remote sensing data acquisition module, which is used to acquire multi-source remote sensing data sources and extract the spatial resolution and time series information of each data source;

[0062] A spatial alignment module, which is used to achieve the spatial alignment of different data sources by using an interpolation algorithm according to the spatial resolution and time series information, and generate a data set with a unified spatial resolution;

[0063] A time series alignment module, which is used to synchronize the time series of different data sources by using a time series alignment algorithm on the basis of spatial alignment, and obtain a multi-source data set with time alignment;

[0064] A noise filtering module, which is used to filter the noise of the multi-source data set after time alignment by using a Gaussian filter to eliminate the high-frequency noise interference in the data;

[0065] A geometric correction module, which is used to perform geometric correction on the data after noise filtering according to a pre-established geometric correction model to obtain a multi-source data set after geometric correction;

[0066] A point cloud conversion module, which is used to convert the point cloud data into an initial mesh model by using a point cloud generation algorithm on the basis of the data after geometric correction;

[0067] A mesh optimization module, which is used to decompose the model into multiple sub-meshes by using a mesh segmentation algorithm for the initial mesh model and perform topological optimization between the sub-meshes to generate an optimized mesh model;

[0068] A rendering processing module, which is used to perform rendering processing by using a lighting model and a texture mapping algorithm based on the optimized mesh model to generate a three-dimensional model with visual effects.

[0069] The above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A three-dimensional display method applied to remote sensing data, characterized in that, Including: Obtain multi-source remote sensing data; Perform spatial alignment and temporal alignment on the multi-source remote sensing data; Denoise and geometrically correct the aligned multi-source remote sensing data; Generate an initial mesh model based on the geometrically corrected data; Segment and topologically optimize the initial mesh model to generate an optimized mesh model; Perform rendering processing on the optimized mesh model to obtain a visualized three-dimensional model.

2. The method according to claim 1, wherein: The multi-source remote sensing data includes optical remote sensing data, radar data, and hyperspectral data.

3. The method according to claim 1, wherein: The process of performing spatial alignment on the multi-source remote sensing data includes: Obtain the spatial resolution and time series information of the multi-source remote sensing data, and align the multi-source remote sensing data with different spatial resolutions through a spatial interpolation algorithm or resampling method to obtain a spatially aligned data set.

4. The method according to claim 3, wherein: The process of obtaining the subsequent multi-source remote sensing data includes: Denoise the time series information, and synchronize the denoised time series information through a time series alignment algorithm to obtain the corresponding multi-source remote sensing data.

5. The method according to claim 1, wherein: The process of denoising and geometrically correcting the aligned multi-source remote sensing data includes: Denoise the aligned multi-source remote sensing data through a Gaussian filter, geometrically correct the denoised data through a geometric correction model, and verify and review the geometric correction result to obtain the geometrically corrected data.

6. The method according to claim 1, wherein: The process of generating the initial mesh model includes: Obtain the three-dimensional coordinate information of the geometrically corrected data, organize and manage the point cloud through an octree, construct a mesh model through a triangulation method, and optimize the constructed mesh model through a smoothing optimization method and a LOD method to obtain an initial mesh model.

7. The method according to claim 1, wherein: The process of segmenting and topologically optimizing the initial mesh model includes: Extract and classify the features of different meshes of the initial mesh model to obtain a classification result, segment the initial mesh model according to the classification result to obtain sub-meshes, obtain the topological structure information of the sub-meshes, and adjust and combine the sub-meshes through a topological optimization algorithm according to the topological structure information of the sub-meshes to obtain an optimized mesh model.

8. The method according to claim 1, wherein: The process of obtaining the visualized three-dimensional model includes: Perform ray reflection calculation on the optimized mesh model through a lighting model to obtain the lighting intensity analysis of each point on the mesh surface. According to the lighting intensity distribution result, map the preset texture information to the mesh surface through a texture mapping algorithm to generate a preliminarily rendered three-dimensional model. Extract the time series information from the preliminarily rendered three-dimensional model, and analyze the main frequency band characteristics of the time series by using a frequency domain analysis method. If the main frequency band characteristics meet the preset lighting effect threshold, output the final visualized three-dimensional model; if not, adjust the lighting model parameters and recalculate the ray reflection.

Images