Unmanned aerial vehicle refined surveying and mapping method for digital elevation on natural ice and snow surface

Abstract

The invention relates to the technical field of ice and snow surveying and mapping, and discloses an unmanned aerial vehicle refined surveying and mapping method for natural ice and snow surface digital elevation. According to the method, a surveying and mapping area is deconstructed into an airspace flight layer, an ice and snow covering layer and a bedrock terrain layer; unmanned aerial vehicle flight attitude parameters and illumination intensity distribution data of an airspace flight layer are obtained, and a multispectral image sequence and a laser point cloud data set of an ice and snow covering layer are collected according to the data; performing multi-resolution adaptive decomposition on the laser point cloud data set to generate an ice and snow layer point cloud data component set, and extracting ice and rock junction spectral features in combination with a multi-spectral image sequence; predicting a reference elevation model based on the features and bedrock terrain layer historical data, calculating an elevation residual value of an ice and snow layer point cloud data component set and the reference elevation model, identifying and eliminating abnormal point cloud data according to a preset terrain anomaly criterion, reconstructing an ice and snow surface point cloud topological structure, and generating a digital elevation surface model in combination with a multispectral image sequence. And high-precision and refined surveying and mapping of natural ice and snow surfaces can be realized.

Description

Technical Field

[0001] This invention relates to the field of ice and snow surveying technology, specifically to a method for high-precision UAV surveying of digital elevation of natural ice and snow surfaces. Background Technology

[0002] In surveys of snow and ice resources, monitoring of glacier dynamics, and engineering projects in cold regions, obtaining accurate digital elevation information of snow and ice surfaces is crucial. Currently, the main technical means used for snow and ice surface mapping include satellite remote sensing mapping, ground-based manual mapping, and traditional UAV mapping. Satellite remote sensing mapping technology can achieve large-scale observation of snow and ice coverage areas, but it is limited by satellite imaging resolution and revisit period, making it difficult to obtain detailed elevation data of snow and ice surfaces on a small scale. Furthermore, in areas with heavy cloud cover, data acquisition quality is easily affected, failing to meet the high-precision monitoring requirements for local snow and ice topographic changes. While ground-based manual mapping technology can obtain high-precision elevation data in specific areas, snow and ice covered areas are often characterized by harsh environments, complex terrain, and difficulty for personnel to access. This not only results in low operational efficiency but also poses significant safety risks. Additionally, it cannot achieve rapid mapping of large-scale snow and ice areas, making it unsuitable for large-scale snow and ice resource surveys. Traditional UAV mapping technology has, to some extent, compensated for the shortcomings of satellite remote sensing and ground-based manual mapping, enabling mapping of medium-sized ice and snow areas. However, it still faces numerous challenges in practical applications. Traditional UAV mapping typically focuses only on a single flight observation level, failing to fully consider the relationship between the ice and snow cover layer and the underlying bedrock topography layer. This results in insufficient accuracy when dealing with variations in ice and snow thickness and identifying ice-rock boundaries. Furthermore, traditional methods often employ a uniform, fixed-resolution decomposition approach for laser point cloud data processing after acquisition. This fails to adapt to the topographic features of different areas on the ice and snow surface, easily leading to redundant point cloud data or loss of crucial information, thus affecting the accuracy of subsequent digital elevation model (DEM) generation. Simultaneously, in the point cloud data denoising process, traditional methods often use simple threshold filtering, which struggles to effectively identify and eliminate abnormal point cloud data caused by factors such as surface reflection and abrupt topographic changes. This further reduces the reliability of the final DEM surface model, failing to meet the practical needs for refined, high-precision mapping of natural ice and snow surfaces. Summary of the Invention

[0003] The purpose of this invention is to provide a method for high-resolution UAV mapping of digital elevation of natural ice and snow surfaces, in order to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, this invention provides a method for high-resolution UAV mapping of digital elevation data for natural ice and snow surfaces, the method comprising: The survey area is deconstructed into three observation levels; wherein, the observation levels include the airspace flight level, the snow and ice cover level, and the bedrock topography level; Acquire the UAV flight attitude parameters and illumination intensity distribution data of the airspace flight layer; Based on the UAV flight attitude parameters and light intensity distribution data, collect multispectral image sequences and laser point cloud datasets of the snow and ice cover layer; The laser point cloud dataset is subjected to multi-resolution adaptive decomposition to generate a snow and ice layer point cloud data component set; Based on the multispectral image sequence and the snow and ice layer point cloud data component set, extract the spectral features of the ice-rock interface; Based on the spectral characteristics of the ice-rock interface and historical data of the bedrock topographic layer, a benchmark elevation model for the bedrock topographic layer is predicted. Calculate the elevation residual between the snow and ice layer point cloud data component set and the benchmark elevation model; Anomaly point cloud data is identified based on the elevation residual value and the preset terrain anomaly criteria. Remove the abnormal point cloud data and reconstruct the topology of the point cloud on the ice and snow surface; A digital elevation surface model is generated based on the reconstructed point cloud topology of the ice and snow surface and the multispectral image sequence.

[0005] Preferably, the step of acquiring the UAV flight attitude parameters and illumination intensity distribution data of the airspace flight layer includes: Collect real-time pitch angle data, roll angle data, heading angle data, and positioning data of the UAV; Simultaneously acquire solar elevation angle data, atmospheric transmittance data, and cloud reflectance data; Light intensity distribution data are generated using the real-time pitch angle data, roll angle data, heading angle data, positioning data, solar altitude angle data, atmospheric transmittance data, and cloud reflectance data.

[0006] Preferably, the step of performing multi-resolution adaptive decomposition on the laser point cloud dataset to generate a snow and ice layer point cloud data component set includes: Obtain the point density distribution characteristics and echo intensity distribution characteristics of the laser point cloud dataset; The optimal decomposition scale parameters are determined based on the point density distribution characteristics and echo intensity distribution characteristics. The laser point cloud dataset is decomposed into multiple resolutions using a spatial spectral decomposition algorithm based on the optimal decomposition scale parameter. Generate a set of point cloud data components for the snow and ice layer containing different spatial frequency characteristics.

[0007] Preferably, the step of extracting the spectral features of the ice-rock interface includes: The multispectral image sequence is subjected to snow reflectance correction processing; Obtain the ice-rock boundary curve characteristics of the corrected multispectral image sequence in the near-infrared band; Extract the gradient abrupt change location data and reflectance transition data of the ice-rock boundary curve features; The gradient abrupt change location data and reflectance transition data are fused to generate the spectral characteristics of the ice-rock boundary.

[0008] Preferably, the step of predicting the baseline elevation model of the bedrock topographic layer includes: Obtain geological structure type data and lithological distribution data of historical bedrock topographic layers; The spectral characteristics of the ice-rock boundary, along with geological structure type data and lithological distribution data, are input into the terrain generation network model. Output the baseline elevation model prediction value of the bedrock topographic layer.

[0009] Preferably, the step of calculating the elevation residual between the snow and ice layer point cloud data component set and the reference elevation model includes: Project the snow and ice layer point cloud data component set onto the reference elevation model coordinate system; Calculate the elevation difference between each point in the snow and ice layer point cloud data component set and the corresponding position in the benchmark elevation model; An elevation residual distribution matrix is ​​generated based on the elevation difference.

[0010] Preferably, the step of identifying abnormal point cloud data includes: Obtain the local variance and spatial autocorrelation characteristics of the elevation residual distribution matrix; Determine whether the local variance features and spatial autocorrelation features meet the preset terrain anomaly criteria; If the preset terrain anomaly criteria are met, the corresponding point cloud data is marked as anomalous point cloud data.

[0011] Preferably, the step of removing the abnormal point cloud data and reconstructing the topology of the ice and snow surface point cloud includes: Establish a spatial index for the snow and ice layer point cloud data after removing outlier point cloud data; The inverse distance-weighted interpolation algorithm was used to fill in the elevation data of the missing point cloud locations. The point cloud data of the filled ice and snow layer is processed by Delaunay triangulation to generate the topology of the point cloud on the ice and snow surface.

[0012] Preferably, the step of generating a digital elevation surface model based on the reconstructed ice and snow surface point cloud topology and multispectral image sequence includes: The reconstructed ice and snow surface point cloud topology is subjected to Kriging space interpolation. The texture feature data of the multispectral image sequence are fused to generate rasterized elevation data; The local terrain undulation features of the rasterized elevation data are optimized using a multi-fractal filtering algorithm. Output a digital elevation surface model of natural ice and snow surfaces.

[0013] Preferably, the method further includes: When the void ratio of the point cloud topology on the ice and snow surface exceeds a preset threshold, the acquisition of the multispectral image sequence and laser point cloud dataset of the ice and snow cover layer is re-triggered. The digital elevation surface model is updated based on the newly acquired multispectral image sequence and laser point cloud dataset.

[0014] Compared with the prior art, the beneficial effects of the present invention are: This UAV-based refined mapping method for digital elevation of natural ice and snow surfaces breaks through the limitations of traditional mapping techniques that focus only on a single observation level by deconstructing the mapping area into three observation layers: the airspace flight layer, the ice and snow cover layer, and the bedrock topography layer. It enables comprehensive perception and analysis of the ice and snow mapping area from multiple dimensions, laying a solid foundation for obtaining accurate digital elevation information of the ice and snow surface. In the data acquisition phase, by combining UAV flight attitude parameters and illumination intensity distribution data from the airspace flight layer, multispectral image sequences and laser point cloud datasets from the ice and snow cover layer are collected. This makes the data acquisition process more targeted, effectively reducing the impact of factors such as unstable flight attitude and poor lighting conditions on data quality, and improving the reliability and validity of the raw data. A multi-resolution adaptive decomposition method is employed to generate snow and ice layer point cloud data component sets from laser point cloud datasets. This method can flexibly adjust based on the topographic complexity and data characteristics of different regions on the snow and ice surface. It retains more detailed point cloud information in complex terrain areas and reduces redundant data in flat terrain areas, ensuring the integrity of data in key areas while improving data processing efficiency. This avoids the information loss or data redundancy problems caused by traditional fixed-resolution decomposition methods. Based on multispectral image sequences and snow and ice layer point cloud data component sets, spectral features of the ice-rock boundary are extracted. This accurately identifies the boundary between the snow and ice cover layer and the bedrock topographic layer, providing accurate feature data for subsequent prediction of the bedrock topographic layer benchmark elevation model. This solves the problem of traditional methods failing to accurately distinguish ice-rock boundary areas. Based on the spectral characteristics of the ice-rock boundary and historical data of the bedrock topographic layer, a benchmark elevation model is predicted. This model fully integrates current surveying data and historical topographic information, effectively reducing the impact of difficulties in bedrock topographic observation caused by snow and ice cover. The predicted benchmark elevation model better reflects the actual bedrock topography, providing a reliable reference for subsequent elevation residual calculations. By calculating the elevation residual values ​​between the snow and ice layer point cloud data component sets and the benchmark elevation model, and combining this with preset topographic anomaly criteria to identify anomalous point cloud data, it is possible to accurately locate anomalous data caused by factors such as snow and ice surface reflection, abrupt topographic changes, and data acquisition errors. Compared to traditional simple threshold filtering methods, the accuracy of anomalous point cloud identification is higher, effectively reducing the interference of anomalous data on subsequent modeling. Removing outlier point cloud data and reconstructing the topology of ice and snow surface point clouds can further optimize the quality of point cloud data, making the reconstructed point cloud topology more realistically reflect the terrain features of the ice and snow surface, providing excellent data support for generating high-quality digital elevation surface models. Finally, a digital elevation surface model is generated based on the reconstructed point cloud topology and multispectral image sequences. This model fully integrates the 3D terrain information of the point cloud data and the texture and spectral information of the multispectral images. The generated digital elevation surface model not only has high elevation accuracy but also reflects additional information such as the material and coverage of the ice and snow surface. It can meet the refined needs for digital elevation information of ice and snow surfaces in various scenarios such as ice and snow resource surveys in polar and high-altitude regions, glacier dynamic monitoring, and cold-region engineering construction, thus broadening the application scope of UAV mapping technology in the field of natural ice and snow surface mapping. Attached Figure Description

[0015] Figure 1 This is a schematic diagram illustrating the working principle of the UAV-based refined mapping method for digital elevation of natural ice and snow surfaces as described in this invention. Figure 2 A flowchart for multi-resolution adaptive decomposition of laser point cloud datasets; Figure 3 A flowchart for extracting spectral features from the ice-rock interface; Figure 4 This is a flowchart for calculating elevation residuals. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Please see Figure 1This invention provides a method for high-resolution UAV mapping of digital elevation data for natural ice and snow surfaces, the method comprising: High-precision terrain modeling is achieved by deconstructing the surveyed area into three observation levels: the airspace flight layer, the snow and ice cover layer, and the bedrock topography layer. Specifically, the process involves acquiring UAV flight attitude parameters and illumination intensity distribution data for the airspace flight layer; based on this data, multispectral image sequences and laser point cloud datasets for the snow and ice cover layer are collected; the laser point cloud dataset undergoes multi-resolution adaptive decomposition to generate snow and ice layer point cloud data component sets; spectral features of the ice-rock interface are extracted using the multispectral image sequences and the snow and ice layer point cloud data component sets; a baseline elevation model is predicted by combining historical data from the bedrock topography layer; the elevation residual between the snow and ice layer point cloud data component sets and the baseline elevation model is calculated; anomalous point cloud data is identified based on the elevation residual values ​​and preset terrain anomaly criteria; anomalous point cloud data is removed, and the topology of the snow and ice surface point cloud is reconstructed; finally, a digital elevation surface model is generated based on the reconstructed topology and the multispectral image sequences. The specific implementation methods are described in detail below through several examples.

[0018] Example 1: See Figure 2 This involves acquiring UAV flight attitude parameters and illumination intensity distribution data in the airspace flight layer, as well as performing multi-resolution adaptive decomposition of laser point cloud datasets to generate snow and ice layer point cloud data component sets. When acquiring UAV flight attitude parameters, the onboard inertial measurement unit (IMU) and Global Navigation Satellite System (GNSS) receiver work together. The IMU continuously records pitch, roll, and yaw angle data at a frequency of 100 times per second. Noise is eliminated using a Kalman filter algorithm to output stable attitude information. For example, in mountainous terrain, the UAV may encounter airflow disturbances, causing instantaneous changes in the roll angle. The filtering algorithm smooths these fluctuations, ensuring the reliability of the attitude data. The GNSS receiver provides centimeter-level accuracy positioning data. Combined with real-time kinematic (RTK) technology, atmospheric errors and multipath effects are eliminated, ensuring high accuracy of the UAV's position information even in complex environments. This data is packaged in real-time by an onboard computer, timestamped, and then transmitted to a ground station or cloud storage system.

[0019] The generation of illumination intensity distribution data relies on the fusion of multi-source data. Solar altitude angle data is calculated using astronomical algorithms based on the latitude, longitude, and time information of the surveyed area. This data dynamically changes over time, reflecting the angle of the sun relative to the horizon. Atmospheric transmittance data is measured by airborne meteorological sensors or obtained from local weather stations. This parameter describes the degree of atmospheric attenuation of solar radiation and is affected by factors such as humidity and aerosol concentration. Cloud reflectance data is obtained through analysis of images captured by multispectral cameras. By calculating the reflectance intensity in the visible light band, cloud cover and thickness are inferred. These data are input into a radiative transfer model, which simulates the propagation of light in the atmosphere based on physical laws, outputting a spatially distributed illumination intensity map. This distribution map is represented in a grid format, with each cell containing an illumination intensity value, used for illumination correction in subsequent image acquisition to eliminate the effects of shadows and overexposure.

[0020] Point density distribution characteristics are obtained by statistically analyzing the number of points per unit area. For example, point density is higher in flat, snowy areas and lower in steep or shady areas. Echo intensity distribution characteristics are based on the analysis of signal intensity received by lidar. The intensity value is affected by surface material and incident angle; snowy surfaces typically exhibit high reflectivity, while rocky or bare areas have lower intensity. These characteristics are used to determine the optimal decomposition scale parameter, which is automatically calculated using a clustering algorithm (such as K-means) to divide the point cloud into homogeneous regions, each corresponding to a specific scale value. For example, in detailed ice fissure regions, a smaller decomposition scale is chosen to preserve high-frequency information; in flat areas, a larger scale is chosen to improve processing efficiency.

[0021] The spatial spectral decomposition algorithm performs multi-resolution decomposition based on optimal decomposition scale parameters. This algorithm employs wavelet transform principles, treating point cloud data as signals and separating different frequency components through convolution operations. The decomposition process includes low-pass and high-pass filtering. Low-pass filtering extracts low-frequency background information, such as large-scale terrain trends; high-pass filtering captures high-frequency details, such as subtle undulations and textures. A component set of the snow and ice layer point cloud data is generated, containing multiple subsets, each corresponding to a specific spatial frequency range. For example, the low-frequency component set represents the overall terrain outline, the mid-frequency component set contains medium-scale features such as snow hills, and the high-frequency component set preserves fine structures such as ice crystal textures. These component sets are used in subsequent steps to support feature extraction and anomaly detection without losing the multi-scale information of the original data.

[0022] Emphasis is placed on precise data acquisition and intelligent processing. By integrating sensor data, environmental parameters, and advanced algorithms, the quality and applicability of the output data are ensured. The fusion of UAV attitude and illumination data enhances the reliability of subsequent imagery, while point cloud decomposition preserves multi-scale terrain features, laying a solid data foundation for the overall mapping methodology.

[0023] Example 2: See Figure 3 This involves extracting spectral features of the ice-rock boundary and developing a baseline elevation model for predicting bedrock topographic layers. This process, through fine processing of multispectral images and fusion with geological data, enables accurate inference of hidden bedrock topography. Extracting spectral features of the ice-rock boundary begins with snow reflectance correction processing of the multispectral image sequence. This correction process utilizes the illumination intensity distribution data and atmospheric transmittance parameters generated in Example 1, employing a physics-based radiative transfer model to adjust the original images pixel-by-pixel. Taking the mapping of a glacier area in the Qilian Mountains as an example, in the original multispectral images acquired by UAVs, the snow reflectance on sunny slopes exhibits saturation due to strong sunlight, while shaded areas display lower values ​​due to shadows. The correction model, by incorporating solar altitude angle data and atmospheric parameters, calculates the theoretical reflectance of each pixel, eliminating the effects of uneven illumination and atmospheric scattering, and outputting a corrected image whose reflectance value depends only on surface characteristics. In the processed image sequence, the differences in spectral characteristics of different surface covers become more pronounced.

[0024] Obtaining the ice-rock boundary curve characteristics in the near-infrared band of the corrected multispectral image sequence is crucial for subsequent analysis. The near-infrared band is extremely sensitive to moisture content, and there are significant differences in the spectral responses of snow / ice and bedrock in this band. By scanning the images line by line using spectral analysis tools, the reflectance value of each pixel in the near-infrared band is extracted, generating a reflectance variation curve along the scan line. In the area where the glacier terminus meets the rock wall, this curve typically exhibits a clear inflection point: the reflectance remains high and relatively stable in the snow / ice-covered area, while it drops sharply when transitioning to the rocky area. This pattern of change constitutes the core content of the ice-rock boundary curve characteristics.

[0025] Extracting gradient abrupt change location data and reflectance transition data from the ice-rock boundary curve requires digital image processing methods. First-order differential calculations are performed on the near-infrared reflectance curve to obtain its gradient change information. Sudden increases in gradient values ​​typically indicate boundary locations; these abrupt change points can be automatically detected by setting appropriate thresholds. Simultaneously, the reflectance difference between the regions on either side of the boundary is calculated as transition data, reflecting the difference in optical properties between ice / snow and bedrock. In a practical application in a periglacial region of the Altai Mountains, the system detected 17 locations where gradient values ​​exceeded a preset threshold. The average reflectance difference at these points reached 0.35, significantly higher than in other areas.

[0026] Generating spectral features of the ice-rock boundary by fusing gradient abrupt change location data and reflectance transition data is a data integration process. It involves spatially registering all detected boundary location information with corresponding reflectance difference values ​​to create a vector layer containing location coordinates and spectral feature values. Each feature point not only contains geographic coordinate information but also stores attributes such as the magnitude of reflectance transitions and boundary orientation. This fused data provides dual information on the spatial distribution and spectral characteristics of the ice-rock boundary, offering crucial input for bedrock topography prediction.

[0027] Predicting the baseline elevation model for bedrock topographic layers requires integrating multi-source data. First, it's essential to acquire historical data on the geological structure types and lithological distribution of the bedrock topographic layers. This data typically comes from regional geological survey reports, previous geophysical exploration results, or publicly available geological databases. Geological structure type data includes spatial distribution information of tectonic elements such as fault lines and fold traces, while lithological distribution data describes the outcrop range of different rock types (e.g., granite, limestone, shale). For example, in a valley in the Himalayas, historical geological maps show the presence of a major thrust fault and distribution areas of three different lithologies.

[0028] The spectral characteristics of the ice-rock boundary, along with geological structural type data and lithological distribution data, are input into a terrain generation network model for comprehensive processing. The terrain generation network model employs a machine learning-based architecture, establishing a mapping relationship between input features and terrain elevation through training on historical terrain data. The network first extracts and fuses features from the input data, spatially associating the boundary locations indicated by spectral features with geological structural lines, while considering the influence of different lithologies on terrain morphology. The model outputs a baseline elevation model prediction of the bedrock topographic layer. This output is elevation data in a regular grid format, with each grid cell containing the predicted bedrock surface elevation value. Throughout the processing, the system maintains spatial reference consistency across different data sources, ensuring all analyses are performed in a unified coordinate system. The final generated baseline elevation model provides an important reference benchmark for subsequent ice and snow thickness calculations and anomaly detection.

[0029] Example 3: See Figure 4 This involves calculating the elevation residuals between the snow and ice layer point cloud data component sets and the benchmark elevation model, as well as identifying anomalous point cloud data. The first step in calculating the elevation residuals is to project the snow and ice layer point cloud data component sets onto the benchmark elevation model coordinate system. This projection transformation requires a unified spatial reference system, typically using a seven-parameter coordinate transformation model to convert the point cloud data from the local coordinate system acquired by lidar to the geodetic coordinate system or projected coordinate system used by the benchmark elevation model. During the transformation, coordinate rotation, translation, and scale factors need to be considered to ensure that the point cloud data and the benchmark model correspond precisely in spatial location. The projected point cloud data retains its original three-dimensional coordinate attributes, but the coordinate values ​​have been transformed to the target coordinate system.

[0030] The elevation difference between each point in the snow and ice layer point cloud data component set and its corresponding location in the benchmark elevation model is calculated point by point. This calculation process is achieved through spatial interpolation. First, the planar coordinate position of each point cloud data point is determined on the surface of the benchmark elevation model. Then, the benchmark elevation value corresponding to that location is obtained through bilinear interpolation or nearest-neighbor interpolation. The actual elevation value of the point cloud data point is subtracted from the interpolated benchmark elevation value to obtain the elevation difference for each point. This difference reflects the deviation between the actual measured elevation of the snow and ice surface and the estimated bedrock elevation, including snow and ice thickness information as well as possible measurement errors or anomalous terrain features.

[0031] An elevation residual distribution matrix is ​​generated based on the elevation differences. This matrix is ​​a regular grid structure, with the grid size determined by the point cloud density and terrain complexity. Each grid cell contains the statistical characteristic value of all point cloud elevation differences within that region, typically using the mean or median as the representative value. The row and column indices of the matrix correspond to planar coordinate positions, and the matrix element values ​​represent the elevation residuals at those locations. This matrix representation is beneficial for subsequent spatial analysis and anomaly detection.

[0032] When identifying anomalous point cloud data, the local variance characteristics of the elevation residual value distribution matrix are first obtained. These local variance characteristics are calculated using a sliding window method, with the window size typically set to a 3×3 or 5×5 grid cell. For each location in the matrix, the variance of all values ​​within its surrounding window is calculated: in: Let N represent the local variance at position (i,j), and N be the number of grid cells within the window. Let K be the elevation residual value of the k-th grid cell. This is the arithmetic mean of all values ​​within the window. Local variance characteristics reflect the spatial variability of elevation residual values; high variance areas typically correspond to abrupt topographic changes or anomalies.

[0033] Simultaneously, the spatial autocorrelation characteristics of the elevation residual distribution matrix are obtained, calculated using global and local Moran indices. The global Moran index measures the spatial clustering of elevation residual values ​​across the entire region, while the local Moran index identifies local spatial clustering patterns. The calculation requires defining a spatial weight matrix, typically constructed based on a distance decay function or adjacency relationships. Spatial autocorrelation characteristics help identify elevation residual regions exhibiting anomalous spatial distribution patterns.

[0034] The determination of whether local variance and spatial autocorrelation characteristics meet preset terrain anomaly criteria is based on statistical principles and topographic knowledge, and typically includes a combination of variance and autocorrelation coefficient thresholds. For example, a terrain anomaly might be considered to exist in an area when the local variance exceeds twice the standard deviation and the local Moran's index shows significant spatial clustering. The specific values ​​of these thresholds need to be adjusted according to the actual terrain characteristics and measurement accuracy.

[0035] If a preset terrain anomaly criterion is met, the corresponding point cloud data is marked as anomalous. This marking process is achieved through spatial mapping, mapping grid cells in the matrix that meet the anomaly conditions back to the original point cloud data, thus marking all point cloud data points within that area. The marked anomalous point cloud data will be removed in subsequent processing steps to improve the accuracy and reliability of the digital elevation model. The entire anomaly identification process enables the automatic detection of special terrain features (such as ice fissures, ice caves, and deposits) within the snow and ice layer, providing data quality assurance for refined terrain modeling.

[0036] Example 4: This involves the process of removing outlier point cloud data and reconstructing the topology of the ice and snow surface point cloud, as well as generating a digital elevation surface model based on the reconstruction results. The core objective of this stage is to construct an accurate and reliable digital elevation model of the ice and snow surface through data repair and fusion techniques after identifying outlier data points.

[0037] After removing outlier point cloud data, the first step is to establish a spatial index for the snow and ice layer point cloud data. This step uses a quadtree index structure to organize and manage the remaining point cloud data. The quadtree index recursively divides the two-dimensional space into four quadrants, with each node storing the location information and elevation values ​​of all point cloud data within that region. Taking the point cloud data processing of a glacier area in the Tianshan Mountains as an example, after removing approximately 5% of the point cloud data marked as outliers, the remaining point cloud data is organized using a quadtree index with an index depth of 8 levels, ultimately forming an index structure containing 256 leaf nodes. Each leaf node stores the spatial extent and elevation statistics of the point cloud data within its corresponding region. This index structure significantly improves the efficiency of subsequent data processing.

[0038] An inverse distance weighted interpolation algorithm was used to fill in the elevation data of missing point cloud locations. This algorithm is based on the principle of spatial correlation, using the elevation values ​​of surrounding known points to estimate the elevation of the missing location. For each missing point location, the algorithm searches for known points within a certain range around it, calculating weights based on the distances between these known points and the missing point; the closer the known points are, the greater their weight. During the interpolation process, the search radius was set to three times the average point spacing, and the power parameter was set to 2. This parameter setting ensures that the interpolation results reflect local terrain features without excessively smoothing terrain details. Through inverse distance weighted interpolation, the elevation information of the void areas caused by outlier removal was successfully recovered.

[0039] The filled snow and ice layer point cloud data is subjected to Delaunay triangulation to generate the topological structure of the snow and ice surface point cloud. Delaunay triangulation is a method that connects scattered point sets into a continuous triangular network, characterized by maximizing the minimum interior angle and avoiding the formation of elongated triangles. In the process, the point cloud data is first projected onto a plane, and then a triangular network is constructed on a two-dimensional plane. Each data point becomes a vertex of the triangular network, and the connections between points form the topological structure. This triangular network structure accurately represents the geometry and continuity of the snow and ice surface, providing a geometric foundation for subsequent elevation model generation.

[0040] When generating a digital elevation surface model based on the reconstructed ice and snow surface point cloud topology and multispectral image sequences, the reconstructed ice and snow surface point cloud topology is first processed by Kriging spatial interpolation. Kriging interpolation is a statistical spatial interpolation method that considers not only the distance relationship between known points and points to be estimated, but also the spatial autocorrelation characteristics of the data. During the interpolation process, the spatial correlation range of the point cloud data is determined through variogram analysis, and then this information is used to perform optimal unbiased estimation. The interpolation result is elevation data in a regular grid, with the grid resolution determined based on the point cloud density and application requirements.

[0041] Rasterized elevation data is generated by fusing texture feature data from multispectral image sequences. The texture information provided by multispectral imagery includes features such as surface color and reflectance characteristics, which are correlated with topographic elevation. Multispectral imagery is aligned with an elevation grid using image registration techniques, and then texture features, including gray-level co-occurrence matrix features and edge density features, are extracted from the images. These texture features are then fused with the elevation data to generate a rasterized elevation dataset rich in surface information.

[0042] A multifractal filtering algorithm was employed to optimize the local topographic relief features of rasterized elevation data. Based on fractal geometry theory, multifractal filtering effectively enhances detailed topographic features while suppressing noise. During the filtering process, the statistical characteristics of the elevation data at different scales were analyzed to identify and enhance realistic topographic features, such as ice ripples and micro-cracks. The filtered elevation data better preserves the natural undulations of the terrain, improving the realism and accuracy of the digital elevation model.

[0043] Table 1: Processing parameters for point cloud data on ice and snow surfaces.

[0044]

[0045] Referring to Table 1, the final output digital elevation surface model (DEM) of natural ice and snow is a comprehensive dataset containing spatial geometric information and surface attribute information. The model is stored in a regular grid format, with each grid cell containing elevation values, accuracy indicators, and texture information from multispectral imagery. This DEM accurately reflects the microscopic topographic features and macroscopic morphological structure of the ice and snow surface, providing a reliable data foundation for applications such as glacier change monitoring and ice and snow hydrological research. The entire processing achieves the transformation from defective point cloud data to a complete and accurate DEM, demonstrating the effectiveness and practicality of UAV-based refined mapping methods.

[0046] Example 5: This example involves a quality monitoring and dynamic update mechanism for the topology of point clouds on snow and ice surfaces. The core of this implementation lies in establishing a feedback adjustment process. When the point cloud quality is found to be unsatisfactory during data processing, a re-acquisition process is automatically initiated, and the final digital elevation model is updated through data fusion technology. The implementation process begins with a quality assessment of the generated point cloud topology on the snow and ice surface. The core indicator for quality assessment is the void ratio, which is the proportion of missing areas in the point cloud data to the total area. The void ratio is calculated through spatial statistical analysis. The entire mapping area is divided into regular grids, and the presence of point cloud data is detected in each grid cell. Grid cells lacking data are marked as voids, and the void ratio is defined as the ratio of the number of void grids to the total number of grids. This calculation process is fully automated, and the system can monitor quality changes in real time during data processing.

[0047] The preset threshold is set based on a comprehensive consideration of surveying accuracy requirements and terrain features. The threshold is usually determined according to glacier type, terrain complexity, and application needs. For relatively flat ice sheet areas, the threshold may be set higher; while for glacier tongue areas with complex terrain and well-developed fissures, the threshold is set lower. When the system detects that the void ratio of the point cloud topology on the ice and snow surface exceeds this preset threshold, it automatically triggers a re-acquisition process.

[0048] Re-triggered acquisition of multispectral image sequences and laser point cloud datasets of the snow and ice cover is an intelligent process. The system first analyzes the spatial distribution characteristics of the cavity areas to determine the specific areas requiring re-acquisition. Then, it automatically generates an optimized UAV flight path, ensuring coverage of all cavity areas while avoiding duplicate acquisition. Flight path planning considers environmental factors such as terrain undulations, wind direction and speed, and the effective working range of the sensors. During the re-acquisition process, the multispectral camera and lidar on the UAV work synchronously to acquire new imagery and point cloud data. The newly acquired data maintains strict consistency with the original data in terms of timestamps and spatial reference, facilitating subsequent data fusion processing.

[0049] Updating the digital elevation surface model (DEM) based on newly acquired multispectral image sequences and laser point cloud datasets is an iterative optimization process. The new data first undergoes the same processing steps as the original data, including point cloud decomposition, feature extraction, and residual calculation. Then, the new and old data are fused, taking into account both temporal consistency and spatial continuity. For overlapping areas, a weighted average method is used to integrate data from different periods, with weights determined based on data quality and acquisition time. For existing voids, newly acquired data is used directly to fill them. The resulting updated DEM not only fills in the original data gaps but also improves overall data quality and timeliness.

[0050] The entire implementation process embodies an intelligent and adaptive data processing concept. The system can autonomously monitor data quality, automatically initiate remedial measures when deficiencies are detected, and generate more complete products through data fusion technology. This implementation method is particularly suitable for long-term glacier monitoring projects, capable of addressing data quality issues caused by uncertainties such as weather changes and equipment failures, ensuring the continuity and reliability of the digital elevation model sequence. Through this dynamic update mechanism, the final output digital elevation surface model can more accurately and completely reflect the true topographic features of the snow and ice surface, providing high-quality data support for applications such as glaciological research and water resource assessment.

[0051] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0052] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for refined UAV mapping of digital elevation of natural ice and snow surfaces, characterized in that, include: The survey area is deconstructed into three observation levels; wherein, the observation levels include the airspace flight level, the snow and ice cover level, and the bedrock topography level; Acquire the UAV flight attitude parameters and illumination intensity distribution data of the airspace flight layer; Based on the UAV flight attitude parameters and light intensity distribution data, collect multispectral image sequences and laser point cloud datasets of the snow and ice cover layer; The laser point cloud dataset is subjected to multi-resolution adaptive decomposition to generate a snow and ice layer point cloud data component set; Based on the multispectral image sequence and the snow and ice layer point cloud data component set, extract the spectral features of the ice-rock interface; Based on the spectral characteristics of the ice-rock interface and historical data of the bedrock topographic layer, a benchmark elevation model for the bedrock topographic layer is predicted. Calculate the elevation residual between the snow and ice layer point cloud data component set and the benchmark elevation model; Anomaly point cloud data is identified based on the elevation residual value and the preset terrain anomaly criteria. Remove the abnormal point cloud data and reconstruct the topology of the point cloud on the ice and snow surface; A digital elevation surface model is generated based on the reconstructed point cloud topology of the ice and snow surface and the multispectral image sequence.

2. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 1, characterized in that, The steps of acquiring the UAV flight attitude parameters and illumination intensity distribution data of the airspace flight layer include: Collect real-time pitch angle data, roll angle data, heading angle data, and positioning data of the UAV; Simultaneously acquire solar elevation angle data, atmospheric transmittance data, and cloud reflectance data; Light intensity distribution data are generated using the real-time pitch angle data, roll angle data, heading angle data, positioning data, solar altitude angle data, atmospheric transmittance data, and cloud reflectance data.

3. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 1, characterized in that, The step of performing multi-resolution adaptive decomposition on the laser point cloud dataset to generate ice and snow layer point cloud data component sets includes: Obtain the point density distribution characteristics and echo intensity distribution characteristics of the laser point cloud dataset; The optimal decomposition scale parameters are determined based on the point density distribution characteristics and echo intensity distribution characteristics. The laser point cloud dataset is decomposed into multiple resolutions using a spatial spectral decomposition algorithm based on the optimal decomposition scale parameter. Generate a set of point cloud data components for the snow and ice layer containing different spatial frequency characteristics.

4. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 1, characterized in that, The step of extracting spectral features of the ice-rock interface includes: The multispectral image sequence is subjected to snow reflectance correction processing; Obtain the ice-rock boundary curve characteristics of the corrected multispectral image sequence in the near-infrared band; Extract the gradient abrupt change location data and reflectance transition data of the ice-rock boundary curve features; The gradient abrupt change location data and reflectance transition data are fused to generate the spectral characteristics of the ice-rock boundary.

5. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 4, characterized in that, The step of predicting the baseline elevation model of the bedrock topographic layer includes: Obtain geological structure type data and lithological distribution data of historical bedrock topographic layers; The spectral characteristics of the ice-rock boundary, along with geological structure type data and lithological distribution data, are input into the terrain generation network model. Output the baseline elevation model prediction value of the bedrock topographic layer.

6. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 1, characterized in that, The step of calculating the elevation residual between the snow and ice layer point cloud data component set and the benchmark elevation model includes: Project the snow and ice layer point cloud data component set onto the reference elevation model coordinate system; Calculate the elevation difference between each point in the snow and ice layer point cloud data component set and the corresponding position in the benchmark elevation model; An elevation residual distribution matrix is ​​generated based on the elevation difference.

7. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 6, characterized in that, The step of identifying abnormal point cloud data includes: Obtain the local variance and spatial autocorrelation characteristics of the elevation residual distribution matrix; Determine whether the local variance features and spatial autocorrelation features meet the preset terrain anomaly criteria; If the preset terrain anomaly criteria are met, the corresponding point cloud data is marked as anomalous point cloud data.

8. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 7, characterized in that, The step of removing the abnormal point cloud data and reconstructing the topology of the ice and snow surface point cloud includes: Establish a spatial index for the snow and ice layer point cloud data after removing outlier point cloud data; The inverse distance-weighted interpolation algorithm was used to fill in the elevation data of the missing point cloud locations. The point cloud data of the filled ice and snow layer is processed by Delaunay triangulation to generate the topology of the point cloud on the ice and snow surface.

9. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 1, characterized in that, The step of generating a digital elevation surface model based on the reconstructed ice and snow surface point cloud topology and multispectral image sequence includes: The reconstructed ice and snow surface point cloud topology is subjected to Kriging space interpolation. The texture feature data of the multispectral image sequence are fused to generate rasterized elevation data; The local terrain undulation features of the rasterized elevation data are optimized using a multi-fractal filtering algorithm. Output a digital elevation surface model of natural ice and snow surfaces.

10. The UAV-based refined mapping method for digital elevation of natural ice and snow surfaces according to claim 9, characterized in that, Also includes: When the void ratio of the point cloud topology on the ice and snow surface exceeds a preset threshold, the acquisition of the multispectral image sequence and laser point cloud dataset of the ice and snow cover layer is re-triggered. The digital elevation surface model is updated based on the newly acquired multispectral image sequence and laser point cloud dataset.

Images