A method and system for monitoring glacier flow velocity based on multi-source remote sensing data

Abstract

The application discloses a kind of glacier flow rate monitoring method and system based on multi-source remote sensing data.Method includes: obtaining optical image, radar data, height model, time series information and glacier surface image;Through data fusion analysis, generate glacier change trend, calculate glacier surface two-dimensional displacement vector and construct three-dimensional flow velocity vector;Extract feature point coordinates for gridding spatial analysis, combined with local interpolation method to improve three-dimensional motion description;Extract texture and crack features from surface image, analyze microstructure using convolutional neural network;Through space-time correlation and weighted fusion, generate preliminary dynamic information, correct and filter processed by historical data to determine the flow velocity change trend, finally output complete glacier dynamic information through dynamic prediction model.This method solves the problem of dimension loss and insufficient accuracy in traditional glacier monitoring, and obtains more comprehensive and accurate glacier dynamic information.

Description

Technical Field

[0001] This invention relates to the field of remote sensing monitoring technology, and in particular to a method and system for monitoring glacier flow velocity based on multi-source remote sensing data. Background Technology

[0002] Currently, glaciers serve as sensitive indicators of global climate change, and their dynamic evolution directly impacts the water cycle, sea-level changes, and ecosystem stability. Accurate monitoring of glacier movement is crucial for global change research. The current demand for monitoring glacier flow velocity has expanded from traditional local observations to wide-area, continuous, and multi-dimensional dynamic tracking. Especially against the backdrop of global warming, the accelerated retreat of glaciers requires monitoring technologies capable of capturing more subtle motion characteristics and faster spatiotemporal changes. With the development of remote sensing technology, multi-source satellite data provides a data foundation for large-scale dynamic monitoring of glaciers; however, constructing an accurate three-dimensional motion field of glaciers still faces significant challenges. Glacier movement exhibits multi-scale characteristics, with surface displacement displaying complex spatial heterogeneity, while processes such as internal deformation and bottom sliding of the ice body need to be derived through surface motion inversion. Existing technologies struggle to simultaneously meet the dual requirements of large-scale coverage and three-dimensional motion analysis. The limited observation perspective of a single satellite leads to insufficient accuracy in three-dimensional reconstruction, and the spatiotemporal resolution differences between optical and radar data also pose difficulties for multi-source data fusion. Furthermore, changes in the microstructures of glaciers, such as surface fissures and meltwater systems, are coupled with overall glacier motion, but current methods lack a systematic analysis of the dynamic response mechanisms of these micro-features. Therefore, there is an urgent need to develop a technical system that can coordinate remote sensing data from multiple platforms, integrate multi-scale features, and consider both three-dimensional motion reconstruction and surface micro-change analysis. This would overcome the bottlenecks of existing methods in terms of spatiotemporal continuity and motion resolution accuracy, providing a more reliable observational basis for glacier dynamic monitoring.

[0003] Existing technology is a two-dimensional glacier motion monitoring method based on a single synthetic aperture radar (SAR) satellite. This technology utilizes differential interferometry (D-InSAR) or offset tracking techniques to acquire the displacement components of the glacier surface along the radar line-of-sight using repeating orbit SAR images. These displacements are then projected onto a horizontal plane using satellite geometric parameters to derive the glacier's two-dimensional motion field. The specific steps are as follows: First, multi-temporal SAR image data of the same region and incident angle are selected. An interferometric phase map is generated through registration and interferometric processing. After phase unwrapping and terrain correction, the line-of-sight displacement is extracted. Next, a projection model is established based on the satellite orbital geometry, decomposing the line-of-sight displacement into east-west and north-south two-dimensional planar motion components. Finally, the short-term velocity field of the glacier is obtained through time series analysis.

[0004] Because it relies on data from only a single sensor, this method's 3D displacement reconstruction assumes that vertical motion is negligible or compensated for by empirical models, resulting in insufficient sensitivity to glacier surface subsidence or uplift. Furthermore, limited by the revisit period of SAR satellites (e.g., 6-12 days for Sentinel-1), it struggles to capture rapid transient processes in glacier changes. In addition, the method's identification of micro-surface deformations (such as crack propagation) depends on statistical thresholds of pixel offsets, lacking quantitative modeling of local strain fields. Therefore, existing technologies suffer from missing monitoring dimensions and insufficient accuracy. Summary of the Invention

[0005] This invention provides a method and system for monitoring glacier flow velocity based on multi-source remote sensing data, which solves the problems of missing dimensions and insufficient accuracy in traditional glacier monitoring.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a method for monitoring glacier flow velocity based on multi-source remote sensing data, comprising:

[0007] Acquire optical images, radar data, height models, time-series information, and glacier surface images;

[0008] By fusing and analyzing the optical images, radar data, height models, and time series information, the trend of glacier change can be obtained.

[0009] The two-dimensional displacement vector of the glacier surface is calculated based on the glacier change trend, and a three-dimensional velocity vector is constructed by combining it with the height model;

[0010] Feature point coordinates are extracted from the optical images and radar data and superimposed onto the boundaries of pre-established grid cells. The displacement vector within each grid is calculated based on the superposition result, and spatial analysis is performed to obtain spatial vector features.

[0011] The glacier's motion direction and velocity distribution are extracted based on the three-dimensional velocity vector. If the velocity change exceeds a preset velocity change threshold, the velocity distribution is supplemented by local region interpolation to obtain a continuous three-dimensional motion description.

[0012] Texture and crack features were extracted from the glacier surface images, and a convolutional neural network was used to analyze the surface microstructure to obtain microstructure features.

[0013] The spatiotemporal correlation analysis of the three-dimensional motion description and the microstructure features is performed, and the spatiotemporal correlation analysis results are integrated with the spatial vector features to obtain preliminary glacier dynamic information.

[0014] When abnormal fluctuations occur in the preliminary glacier dynamic information, the preliminary glacier dynamic information is corrected by pre-stored historical remote sensing data to determine the trend of glacier flow velocity changes;

[0015] The glacier flow velocity variation trend and the microstructural features are input into a pre-trained dynamic prediction model to obtain complete glacier dynamic information.

[0016] In one optional implementation, the step of fusing and analyzing the optical imagery, radar data, height model, and time series information to obtain the glacier change trend includes:

[0017] The optical image, radar data, and altitude model are spatiotemporally aligned with the time series information using a linear interpolation method to obtain spatiotemporally aligned data.

[0018] Principal component analysis was used to extract the main trends of change in the spatiotemporal aligned data to obtain the glacier change trends, which include the displacement and height changes of the glacier surface.

[0019] In one optional implementation, the step of calculating the two-dimensional displacement vector of the glacier surface based on the glacier change trend and constructing a three-dimensional velocity vector in conjunction with the height model includes:

[0020] Based on the glacier change trend, the Lucas-Kanade algorithm was used to calculate the two-dimensional displacement vector on the glacier surface;

[0021] Based on the two-dimensional displacement vector and the height model, a three-dimensional velocity vector is constructed using a rigid body transformation method, and then the three-dimensional velocity vector is smoothed using a Kalman filter algorithm to obtain a smoothed three-dimensional velocity vector.

[0022] In one optional implementation, the step of extracting feature point coordinates from the optical image and the radar data, superimposing them onto a pre-established grid cell boundary, calculating the displacement vector within each grid based on the superimposition result, and performing spatial analysis to obtain spatial vector features includes:

[0023] A rotation-invariant feature description method is used to extract feature points with invariant scale from optical images;

[0024] A permanent scatterer identification method is used to extract stable scattering feature point coordinates from radar data;

[0025] The coordinates of the invariant feature points and the scattering feature points are matched using the nearest neighbor search method. The coordinates of the successfully matched feature points are converted into the UTM projection coordinate system using a coordinate system transformation method. Then, the feature points are superimposed onto the pre-established grid cell boundary using a spatial buffer analysis method to obtain the superposition result.

[0026] Based on the superposition results, the displacement vector within each grid is obtained using a vector calculation method, and the displacement fluctuation intensity and directional disorder are extracted as spatial vector features based on the displacement vector.

[0027] In one optional implementation, the step of extracting texture and crack features from the glacier surface image and analyzing the surface microstructure using a convolutional neural network to obtain microstructure features includes:

[0028] Texture features were extracted from the glacier surface image using the gray-level co-occurrence matrix algorithm.

[0029] Based on the texture features, the Canny edge detection algorithm is used to extract crack features;

[0030] Based on the crack characteristics, a convolutional neural network was used to analyze the microstructure of the glacier surface to obtain the microstructure features.

[0031] In one optional implementation, the step of performing spatiotemporal correlation analysis on the three-dimensional motion description and the microstructural features, and integrating the spatiotemporal correlation analysis results with the spatial vector features to obtain preliminary glacier dynamic information, includes:

[0032] The spatiotemporal correlation analysis of the three-dimensional motion description and the microstructure features is performed by a dynamic time warping algorithm to obtain the spatiotemporal correlation results.

[0033] A weighted fusion algorithm is used to integrate the spatiotemporal correlation results and the spatial vector features to obtain preliminary glacier dynamic information.

[0034] In one optional implementation, the training process of the dynamic prediction model includes:

[0035] The historical glacier flow velocity variation trend and historical microstructure characteristics are obtained, and the data is cleaned and features are extracted from the historical glacier flow velocity variation trend and historical microstructure characteristics to obtain data features;

[0036] The data features are input into the input layer of the initially constructed neural network model for training, and the predicted values ​​output by the output layer of the neural network model are obtained.

[0037] Substitute the predicted value and the pre-stored actual value into the loss function to calculate the loss value;

[0038] The gradient of the output layer of the neural network model is calculated based on the loss value, and the gradient is passed forward layer by layer through the chain rule to calculate the gradient of the parameters of each layer and obtain the gradient data.

[0039] Update the parameters of each layer of the neural network model based on gradient data and a preset learning rate.

[0040] The parameters of each layer are iteratively updated until the number of training iterations of the neural network model is greater than a preset number of iterations, or the loss value of the neural network model is less than a preset loss threshold. At this point, the training is considered complete, and a dynamic prediction model is obtained.

[0041] Secondly, the present invention provides a glacier flow velocity monitoring system based on multi-source remote sensing data, comprising:

[0042] The data acquisition module is used to acquire optical images, radar data, height models, time-series information, and glacier surface images;

[0043] The data fusion module is used to perform data fusion analysis on the optical images, radar data, height model and time series information to obtain the glacier change trend;

[0044] The vector construction module is used to calculate the two-dimensional displacement vector of the glacier surface based on the glacier change trend, and to construct a three-dimensional flow velocity vector in combination with the height model;

[0045] The vector analysis module is used to extract feature point coordinates from the optical image and the radar data, and superimpose them onto the boundaries of a pre-established grid cell. Based on the superposition result, the displacement vector within each grid is calculated and spatial analysis is performed to obtain spatial vector features.

[0046] The three-dimensional motion analysis module is used to extract the glacier's motion direction and velocity distribution based on the three-dimensional velocity vector. If the velocity change exceeds a preset velocity change threshold, the velocity distribution is supplemented by local region interpolation to obtain a continuous three-dimensional motion description.

[0047] The microscopic feature analysis module is used to extract texture features and crack features from the glacier surface image, and to analyze the surface microstructure using a convolutional neural network to obtain microstructural features;

[0048] The preliminary dynamic analysis module is used to perform spatiotemporal correlation analysis on the three-dimensional motion description and the microstructure features, and to integrate the spatiotemporal correlation analysis results with the spatial vector features to obtain preliminary glacier dynamic information;

[0049] The data correction module is used to correct the preliminary glacier dynamic information by using pre-stored historical remote sensing data when abnormal fluctuations occur in the preliminary glacier dynamic information, and to determine the trend of glacier flow velocity changes.

[0050] A complete dynamic analysis module is used to input the glacier flow velocity change trend and the microstructural features into a pre-trained dynamic prediction model to obtain complete glacier dynamic information.

[0051] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the glacier flow velocity monitoring method based on multi-source remote sensing data as described above.

[0052] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the glacier flow velocity monitoring method based on multi-source remote sensing data as described above.

[0053] Compared with the prior art, the present invention has the following beneficial effects:

[0054] (1) By using a multi-source data fusion mechanism, optical images, radar data and height models are spatiotemporally aligned and principal component analysis is performed to construct a complete three-dimensional glacier change trend, which solves the problem of insufficient dimension of existing single sensor observation data and improves the three-dimensional dynamic analysis capability of glacier movement.

[0055] (2) The two-dimensional displacement vector of the glacier surface is calculated based on the Lucas-Kanade algorithm, and three-dimensional reconstruction is performed by combining the height model. Kalman filtering is used for optimization, which significantly improves the spatial accuracy and temporal continuity of the glacier motion field and reduces measurement error.

[0056] (3) By extracting stable feature points through rotation-invariant feature description and permanent scatterer identification technology, and constructing displacement vector field by combining gridded spatial analysis method, high-precision quantitative characterization of glacier non-uniform motion is realized, and micro-displacement at the millimeter / year scale can be detected.

[0057] (4) Convolutional neural networks were used to analyze the correlation between changes in the microstructure of the glacier surface and the macroscopic motion field. A multi-scale dynamic response model was established by combining dynamic time warping algorithm, which provides a new analytical dimension for understanding the mechanism of glacier movement.

[0058] In summary, this invention innovatively integrates multi-source remote sensing data to construct a three-dimensional dynamic glacier monitoring system encompassing both macroscopic motion fields and microscopic features. Firstly, multi-sensor data fusion technology overcomes the dimensionality limitations of single data sources. Secondly, advanced feature extraction and motion reconstruction algorithms improve measurement accuracy. Thirdly, neural network technology enhances the system's ability to analyze complex glacier features. Finally, time-series analysis methods enable the visual tracking of dynamic processes. This invention significantly improves the spatiotemporal resolution and accuracy of glacier monitoring. Attached Figure Description

[0059] Figure 1This is a schematic diagram of the glacier flow velocity monitoring method based on multi-source remote sensing data provided in the first embodiment of the present invention;

[0060] Figure 2 This is a schematic diagram of the structure of a glacier flow velocity monitoring system based on multi-source remote sensing data provided in the second embodiment of the present invention. Detailed Implementation

[0061] 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.

[0062] Reference Figure 1 The first embodiment of the present invention provides a method for monitoring glacier flow velocity based on multi-source remote sensing data, comprising the following steps:

[0063] S11 acquires optical images, radar data, height models, time-series information, and glacier surface images;

[0064] S12, perform data fusion analysis on the optical images, radar data, height model and time series information to obtain the glacier change trend;

[0065] S13, Calculate the two-dimensional displacement vector of the glacier surface based on the glacier change trend, and construct a three-dimensional flow velocity vector in combination with the height model;

[0066] S14, extract feature point coordinates from the optical image and radar data, and overlay them onto the pre-established grid cell boundary. Calculate the displacement vector within each grid based on the overlay result and perform spatial analysis to obtain spatial vector features.

[0067] S15, extract the glacier's motion direction and velocity distribution based on the three-dimensional velocity vector. If the velocity change exceeds a preset velocity change threshold, supplement the velocity distribution using local region interpolation to obtain a continuous three-dimensional motion description.

[0068] S16, Extract texture features and crack features from the glacier surface image, and use a convolutional neural network to analyze the surface microstructure to obtain microstructure features;

[0069] S17, Perform spatiotemporal correlation analysis on the three-dimensional motion description and the microstructure features, and integrate the spatiotemporal correlation analysis results with the spatial vector features to obtain preliminary glacier dynamic information;

[0070] S18, when the preliminary glacier dynamic information shows abnormal fluctuations, the preliminary glacier dynamic information is corrected by pre-stored historical remote sensing data to determine the trend of glacier flow velocity change;

[0071] S19, The glacier flow velocity change trend and the microstructure characteristics are input into a pre-trained dynamic prediction model to obtain complete glacier dynamic information.

[0072] In step S11, optical images, radar data, height models, time series information, and glacier surface images are acquired.

[0073] Specifically, the optical images are primarily acquired from multispectral satellite sensors (such as Sentinel-2 MSI and Landsat-8 OLI), containing reflectance data in the visible to short-wave infrared spectral band (400-2300 nm), with a spatial resolution of 10-30 meters. These images, through multispectral feature analysis, can identify snow cover distribution on the glacier surface (utilizing NIR band reflectance differences), crevasse development (through texture enhancement processing), and moraine coverage (combined with SWIR band absorption characteristics). The blue band (450-510 nm) exhibits significant spectral response characteristics to meltwater channels on the glacier surface and can be used to track seasonal ablation processes.

[0074] The radar data specifically refers to L-band (such as ALOS-2) and C-band (such as Sentinel-1) interferometric data acquired by synthetic aperture radar (SAR) satellites, containing amplitude and phase information. The phase information, obtained through differential interferometry (D-InSAR), can extract millimeter-level precision glacier surface deformation, while the amplitude information, obtained through offset tracking technology, can acquire sub-pixel-level displacement fields. Radar data has the ability to penetrate cloud layers, providing continuous observations even during polar nights, and is particularly suitable for monitoring minute movements (on the order of 0.1-1 m / year) caused by glacier base slippage.

[0075] The height model employs spaceborne laser altimetry data (such as ICESat-2 ATLAS) and radar interferometric digital elevation models (such as TanDEM-X DEM), achieving vertical accuracies of ±0.15m and ±2m, respectively. ICESat-2 uses a photon-counting lidar to acquire dense elevation points along its orbit (one measurement point every 70cm), enabling the creation of a three-dimensional point cloud model of the glacier surface and accurately capturing abrupt elevation changes (rapid ablation events >5m / week) caused by the collapse of the glacier tongue's terminus.

[0076] The time series information comes from a multi-temporal remote sensing observation dataset, including: an annual average deformation rate map (time baseline 6-12 months) generated by SBAS-InSAR processing, NDSI (normalized snow index) variation curves of optical image time series (daily scale), and radar backscattering coefficient time series (hourly scale). By fusing data with different time resolutions, a complete time evolution model from short-term pulse-like movements (such as glacier pulsation) to long-term trend changes can be constructed.

[0077] The glacier surface images specifically refer to ultra-high resolution images acquired by multispectral cameras (such as the MicaSense RedEdge-MX) and visible light cameras (with a resolution of up to 1 cm / pixel) mounted on UAVs. These images are used to generate digital surface models (DSMs) through the Structure of Motion (SfM) algorithm, which can clearly identify microscopic surface features such as millimeter-scale ice fissures (width > 5 cm) and the development of glacial lakes (area > 0.1 m²).

[0078] In step S12, the optical images, radar data, height model, and time series information are fused and analyzed to obtain the glacier change trend.

[0079] In one specific implementation, the step of fusing and analyzing the optical imagery, radar data, height model, and time series information to obtain the glacier change trend includes:

[0080] The optical image, radar data, and altitude model are spatiotemporally aligned with the time series information using a linear interpolation method to obtain spatiotemporally aligned data.

[0081] Principal component analysis was used to extract the main trends of change in the spatiotemporal aligned data to obtain the glacier change trends, which include the displacement and height changes of the glacier surface.

[0082] Specifically, the optical imagery, radar data, height models, and time-series information were first spatiotemporally aligned. Optical imagery provides information on glacier surface reflectivity, radar data includes the phase of surface deformation, height models record changes in terrain height, and time-series information marks the observation time points for each data point. Bilinear interpolation was used to unify all data onto the same spatial grid, ensuring that each pixel location corresponds to the same coordinates. Simultaneously, linear interpolation was used to synchronize all data to a unified time point, guaranteeing consistency across the temporal dimension.

[0083] After completing the spatiotemporal alignment of multi-source remote sensing data, an improved principal component analysis method was used to construct the glacier change characteristic field. Considering that the displacement field extracted from optical images and the height change field obtained from radar height models have different physical dimensions, the three components of eastward displacement, northward displacement, and vertical height change were first standardized to eliminate the influence of dimensional differences on the analysis results. By calculating the covariance matrix of the standardized data, the top three principal components with a cumulative contribution rate of 95% were extracted to characterize the overall glacier movement, seasonal ablation, and local anomalous events, respectively. The spatial distribution of the first principal component showed a high correlation with the movement along the main glacier line, and its time coefficient showed an accelerating trend. Comparison with the glacier dynamics model confirmed that this component reflects the overall movement mode dominated by the sliding at the bottom of the ice body. The second principal component showed a significant negative correlation in the ablation zone, and its time coefficient had an annual periodicity. Combined with meteorological station measured ablation data, it was confirmed that it characterizes the height loss caused by summer surface melting. The third principal component showed local extrema in the ice tongue shear zone region, and pulse signals synchronized with ice avalanche events were detected in the time coefficient. Comparison with the synthetic aperture radar phase discorrelation index confirmed that it corresponds to the local ice fracturing process. To address the multi-scale characteristics of the principal component time coefficients, a time-series decomposition method is employed to break down each component into long-term trends, interannual fluctuations, and short-term noise. The displacement trend represents the direction and velocity of movement at various points on the glacier surface, while the height trend specifically refers to the elevation change rate field derived from the long-term trend term. By standardizing the data to balance the contributions of displacement and height variables, the variance of the displacement component is reduced from 78% to 42% of the original data, while the contribution of height variation is increased to 38%, effectively avoiding the dominance of a single variable in the analysis results. The improved method solves the problems of dimensional interference and ambiguity in physical interpretation in traditional principal component analysis by fusing the physical characteristics and statistical regularities of multi-source data. This method can effectively integrate remote sensing data from different sensors and at different resolutions, improving monitoring accuracy through data fusion. Furthermore, this method can accurately identify seasonal patterns of glacier movement, detect millimeter-level interannual variations, and effectively distinguish deformation characteristics caused by different physical processes such as glacier ablation and ice flow movement. Compared to analysis methods based on a single data source, the fused results exhibit higher spatiotemporal consistency and reliability.

[0084] In step S13, the two-dimensional displacement vector of the glacier surface is calculated based on the glacier change trend, and a three-dimensional flow velocity vector is constructed by combining the height model.

[0085] In one specific implementation, the step of calculating the two-dimensional displacement vector of the glacier surface based on the glacier change trend and constructing a three-dimensional velocity vector in conjunction with the height model includes:

[0086] Based on the glacier change trend, the Lucas-Kanade algorithm was used to calculate the two-dimensional displacement vector on the glacier surface;

[0087] Based on the two-dimensional displacement vector and the height model, a three-dimensional velocity vector is constructed using a rigid body transformation method, and then the three-dimensional velocity vector is smoothed using a Kalman filter algorithm to obtain a smoothed three-dimensional velocity vector.

[0088] Specifically, the Lucas-Kanade optical flow algorithm is first used to extract two-dimensional displacement vectors from glacier surface image sequences. This algorithm analyzes the spatial changes of glacier surface features (such as crevasses and surface textures) within optical images over continuous time periods, selects a local image window (e.g., a 15-pixel × 15-pixel region), and calculates the temporal and spatial variations of the brightness gradient of pixels within the window. The algorithm assumes that the glacier surface motion within each window is consistent, and iteratively optimizes the optimal displacement direction and magnitude of pixels within each window to generate a two-dimensional planar motion field covering the entire glacier surface. For glacier areas with weaker texture features, the algorithm automatically increases the analysis window size to improve the reliability of displacement detection.

[0089] After obtaining the two-dimensional displacement field, the planar motion is extended to three-dimensional motion by combining it with height model data. By analyzing the time-series changes of the height model, the vertical height change rate of the glacier surface is extracted. The two-dimensional horizontal displacement vector is spatially registered with the vertical height change data to establish a three-dimensional motion projection model. This model considers the rigid body characteristics of glacier motion in three-dimensional space, and through optimized calculations of rotation and translation parameters, the planar displacement and height change are fused into a three-dimensional velocity vector. For example, if the two-dimensional planar displacement of a certain area shows eastward movement, while the corresponding height data indicates subsidence, the three-dimensional vector will combine these two components to form a downward-sloping three-dimensional motion trajectory.

[0090] Finally, the Kalman filter algorithm is used for time-series optimization of the three-dimensional velocity field. In the optimization of the three-dimensional velocity field, the Kalman filter algorithm improves the reliability of the results by fusing physical law predictions with real-time observation data. The algorithm constructs a prediction model based on the principle of glacier motion continuity, using the previous moment's velocity as a benchmark to estimate the current state; at the same time, it combines real-time observation data from optical and radar, setting data reliability weights according to sensor accuracy. When cloud cover causes local data loss, the algorithm automatically fills in the missing values ​​through historical motion trend predictions and performs probability corrections for abnormal fluctuations (such as single-day velocity mutations exceeding a threshold), ultimately generating a spatiotemporally continuous and error-controllable three-dimensional velocity field. This process combines the planar motion detection capability of optical imagery with the vertical change monitoring capability of height models, breaking through the limitations of traditional two-dimensional analysis. The dynamically adjusted analysis window solves the problem of motion estimation in low-texture glacier areas, while the multi-timescale fusion of Kalman filtering significantly improves the stability of data results under complex meteorological conditions. This method can accurately capture the abrupt motion characteristics in three-dimensional space when monitoring mountain glacier undulation events.

[0091] In step S14, feature point coordinates are extracted using the optical image and radar data and superimposed onto the pre-established grid cell boundary. Based on the superposition result, the displacement vector within each grid is calculated and spatial analysis is performed to obtain spatial vector features.

[0092] In one specific implementation, the step of extracting feature point coordinates from the optical image and the radar data, superimposing them onto a pre-established grid cell boundary, calculating the displacement vector within each grid based on the superimposition result, and performing spatial analysis to obtain spatial vector features includes:

[0093] A rotation-invariant feature description method is used to extract feature points with invariant scale from optical images;

[0094] A permanent scatterer identification method is used to extract stable scattering feature point coordinates from radar data;

[0095] The coordinates of the invariant feature points and the scattering feature points are matched using the nearest neighbor search method. The coordinates of the successfully matched feature points are converted into the UTM projection coordinate system using a coordinate system transformation method. Then, the feature points are superimposed onto the pre-established grid cell boundary using a spatial buffer analysis method to obtain the superposition result.

[0096] Based on the superposition results, the displacement vector within each grid is obtained using a vector calculation method, and the displacement fluctuation intensity and directional disorder are extracted as spatial vector features based on the displacement vector.

[0097] Specifically, stable feature points are first extracted from optical imagery and radar data. For optical imagery, rotation-invariant feature description methods are used to identify key points with scale invariance. This method analyzes the gradient direction histograms in different directions and at Gaussian blur scales in the image to screen for feature points that maintain their positional consistency after image rotation or scaling (such as fissure intersections or rock outcrops on the glacier surface). Radar data is processed using permanent scatterer identification technology. Based on long-term radar phase stability analysis, stable scattering points with high radar echo signal intensity and small phase changes (such as exposed bedrock areas or fixed glacial till) are screened.

[0098] For the problem of cross-sensor feature point matching, the key to the nearest neighbor search method lies in constructing a joint matching strategy that integrates spatial constraints and cross-modal feature consistency: After uniformly transforming optical image feature points and radar scattering points to a standard geographic coordinate system and completing spatial resolution calibration, a spatial candidate matching set for radar scattering points is first established based on a preset spatial search radius to limit the search range of optical feature points and control geometric deviations; then, optical feature points (such as high-dimensional SIFT features reduced in dimensionality by principal component analysis) and radar scattering features (such as multi-dimensional vectors after polarization parameter normalization) are projected onto a unified feature metric space to eliminate the incomparability caused by modal differences; in this fused space, a nearest neighbor search is performed on each radar scattering point within its corresponding spatial candidate set, the normalized feature similarity score is dynamically calculated, and a dual threshold condition of spatial distance and feature similarity is set for joint screening—only when the optical feature point simultaneously meets the spatial proximity constraint (not exceeding the preset maximum allowable spatial deviation) and the feature similarity meets the standard is it determined to be a valid matching pair of optical feature point and scattering feature point coordinates. The coordinates of successfully paired feature points are unified to the UTM projection coordinate system through geographic coordinate system transformation, eliminating positioning errors caused by differences in coordinate systems of different sensors.

[0099] When using spatial buffer analysis to overlay matching feature points onto a pre-generated regular grid system, the core of this approach lies in achieving multi-scale quantitative expression of glacier motion characteristics through spatial statistics and motion field modeling. Specifically, based on a pre-generated regular grid system (e.g., a 100m × 100m grid), a dynamic buffer zone is generated outwards from the boundary line of each grid cell (e.g., the buffer radius is set to 20% of the grid side length, i.e., 20 meters). Spatial overlay analysis extracts all matching feature point pairs falling within the buffer zone, forming a local motion observation sample set for each grid cell. For the sample set within each grid cell, the displacement vector (including horizontal displacement and azimuth) of each feature point pair is first calculated based on the coordinate difference after georegistration. Then, a weighted least squares method is used to fit the optimal solution of all displacement vectors, obtaining the average displacement direction (mean azimuth) and average motion rate (mean modulus) characterizing the overall motion state of the grid cell. Simultaneously, the spatial consistency of the motion direction (directional disorder) is quantified by statistically analyzing the standard deviation of the azimuth, and the coefficient of variation of the rate value characterizes the stability of the motion intensity (fluctuation intensity). For example, if the fitting results of a certain grid cell show an average displacement direction of 12°±5° east of southeast (standard deviation) and an average rate of 2.1 m±0.3 m per day (coefficient of variation 0.14), it indicates that the glacier in this area exhibits highly coordinated and stable movement. Conversely, if another grid cell shows a southeast direction of 15°±32° and a rate of 1.8 m±1.2 m (coefficient of variation 0.67), the directional disorder and wave intensity are significantly increased, suggesting the possible presence of ice fracturing or basement slip anomalies in this area. The overlay results are ultimately presented as a gridded layer, with each grid cell integrating four types of quantitative indicators: movement direction, rate, disorder, and wave intensity. Spatially, the dynamic zoning characteristics of the glacier's mainstream area (low disorder, low wave), shear zones (high disorder linear strips), and crevasse development zones (high wave patches) can be clearly identified, providing a spatially explicit analytical basis for the analysis of glacier movement mechanisms.

[0100] Based on the displacement vector dataset obtained by overlaying within grid cells, a motion field quantification analysis method is used to extract characteristic parameters of glacier motion. First, the displacement vectors of all feature points within the grid are spatially aggregated using a vector synthesis method to calculate the average displacement vector representing the overall motion trend, including the average motion direction (azimuth parameter) and the average motion rate (mean rate parameter). Then, the spatiotemporal heterogeneity of glacier motion is assessed based on the directional dispersion statistics (directional disorder index) and rate fluctuation statistics (fluctuation intensity index) of the displacement vectors. The directional disorder index is calculated by measuring the standard deviation of the displacement direction of each feature point within the grid relative to the average direction. When this index exceeds a preset directional consistency threshold, it indicates the existence of motion direction differentiation characteristics. The fluctuation intensity index is calculated using the coefficient of variation of the rate values ​​and is used to characterize the spatial stability of the motion rate. When this index exceeds a preset rate stability threshold, it reflects abnormal fluctuations in local motion. The final generated gridded spatial vector feature set contains four types of parameters: average motion direction, average rate, directional disorder, and fluctuation intensity, constituting a multi-dimensional quantitative characterization system of the glacier surface motion field, providing a basis for identifying unstable motion regions.

[0101] This process significantly improves the spatial resolution of the displacement field through multi-source feature fusion. Optical feature points provide high spatial density displacement observations, while radar scattering points ensure monitoring continuity in cloud-covered areas. This method can effectively identify the boundary features of ice flow shear zones and undulating glaciers in ice sheet edge monitoring.

[0102] In step S15, the glacier's motion direction and velocity distribution are extracted based on the three-dimensional velocity vector. If the velocity change exceeds a preset velocity change threshold, the velocity distribution is supplemented by local region interpolation to obtain a continuous three-dimensional motion description.

[0103] Specifically, spatial distribution characteristics of glacier motion are extracted based on three-dimensional velocity vector data. First, the three-dimensional velocity vector is decomposed into two physical quantities: direction of motion and magnitude of velocity. The direction is determined by the spatial orientation of the vector (e.g., a combination of east-west, south-north, and vertical components), and the magnitude of velocity is calculated using the vector's modulus. The glacier region is then gridded (e.g., divided into 100m × 100m units), and the average direction of motion and median velocity of all three-dimensional vectors within each grid are statistically analyzed to generate preliminary distribution maps of glacier motion direction and velocity.

[0104] When the velocity difference between adjacent grids exceeds a preset velocity change threshold (e.g., a velocity difference between adjacent grids exceeding 0.5 meters per year), the region is determined to have data discontinuity or anomalies. For such regions, a local interpolation method is used: a buffer zone (e.g., a radius of 300 meters) is defined centered on the anomalous grid. Three-dimensional velocity data from valid observation points within the buffer zone are collected, and the corrected velocity value for the central grid is calculated using an inverse distance-weighted interpolation algorithm. For example, if a grid has missing data due to cloud cover, the algorithm will reference the velocity values ​​of 10 valid observation points within a 200-meter radius, with closer observation points receiving higher weights, ultimately generating a continuous velocity field.

[0105] After interpolation, the corrected velocity field is spatially smoothed. A moving window mean filter (5×5 grid size) is used to eliminate local anomalous fluctuations while preserving the true motion boundaries of the glacier shear zone. The final output 3D motion description data includes the 3D motion direction (azimuth and pitch) and velocity value for each grid cell, as well as a data confidence index.

[0106] By processing three-dimensional velocity vectors, a complete description system of glacier motion is established. Based on a velocity change threshold detection mechanism, abnormal data regions are identified, and missing values ​​are filled using local interpolation techniques, effectively solving the data discontinuity problem caused by cloud cover or sensor limitations. Through gridded statistical processing and spatial filtering, the true characteristics of glacier motion (such as crack propagation and shear zone evolution) are preserved while eliminating random noise interference, generating a high-precision continuous motion field containing three-dimensional motion direction, velocity values, and confidence indices. This method significantly improves the spatial integrity and physical consistency of glacier dynamic monitoring data, providing a reliable three-dimensional motion parameter basis for analyzing glacier undulation mechanisms and assessing ice flow stability.

[0107] In step S16, texture features and crack features are extracted from the glacier surface image, and a convolutional neural network is used to analyze the surface microstructure to obtain microstructure features.

[0108] In one specific implementation, the step of extracting texture and crack features from the glacier surface image and analyzing the surface microstructure using a convolutional neural network to obtain microstructure features includes:

[0109] Texture features were extracted from the glacier surface image using the gray-level co-occurrence matrix algorithm.

[0110] Based on the texture features, the Canny edge detection algorithm is used to extract crack features;

[0111] Based on the crack characteristics, a convolutional neural network was used to analyze the microstructure of the glacier surface to obtain the microstructure features.

[0112] Specifically, the texture features of glacier surface images are first analyzed using a gray-level co-occurrence matrix (GLCM) algorithm. This algorithm quantifies the texture roughness, contrast, and uniformity of the glacier surface by statistically analyzing the frequency of combinations of pixel gray-level values ​​with different directions and spacings in the glacier surface image. For example, smooth ice surfaces exhibit high gray-level uniformity, while cracked areas display high-contrast directional texture patterns. The algorithm selects multiple feature parameters (such as energy, entropy, and correlation) to construct a multi-dimensional texture feature vector, effectively distinguishing different regions on the glacier surface, such as snow cover, bare ice, and moraine cover.

[0113] Based on the extracted texture features, the Canny edge detection algorithm is used to identify ice cracks. The algorithm first eliminates image noise through Gaussian filtering, then calculates the gradient magnitude and direction of each pixel, retaining the edge pixels with the largest gradient magnitude. A dual-threshold mechanism (high threshold captures significant edges, low threshold connects broken edges) generates continuous and fine ice crack contour lines. This process can accurately capture surface cracks with a width of millimeters and effectively distinguish between real ice cracks and false edges caused by image noise.

[0114] Finally, in constructing the ice crack depth analysis model, a multi-scale feature parsing architecture based on convolutional neural networks was adopted: the network takes local image patches containing ice cracks as input, and achieves hierarchical abstraction of features through cascaded operations of multiple convolutional kernels—the shallow network captures primary features such as crack contours and surface textures through edge detection kernels, while the deep network identifies complex structural features such as crack bifurcation topology and spatial clustering patterns through adaptive receptive field expansion; at the end of the network, a fully connected layer maps the multi-scale features to the microstructure classification space, outputting the probability distribution of tensile cracks, shear cracks, and healed cracks. In the post-processing module based on the classification results, the crack density parameter is calculated by statistically analyzing the pixel proportion of each type of crack per unit area, and the crack orientation distribution histogram is extracted based on the morphological analysis of the classification mask. During the training phase, a transfer learning strategy is introduced, loading a pre-trained glacier feature extraction network as the base model. The bottom convolutional layers are frozen to retain general feature extraction capabilities, while the parameters of higher-level networks are fine-tuned to adapt to the fine-grained feature requirements of the fracture classification task. Simultaneously, data augmentation techniques are employed to expand the small training set and improve the model's robustness to interference factors such as illumination changes and snow cover. This architecture, through the collaborative optimization of end-to-end feature learning and post-classification processing, achieves high-precision analysis and quantitative representation of the microstructural features of the glacier surface.

[0115] Finally, a convolutional neural network was constructed to perform in-depth analysis of ice fissure features. The network input consisted of local image patches containing ice fissures (e.g., a 256×256 pixel region). Multi-scale features were extracted through multi-layer convolution and pooling operations: shallow convolutional kernels captured edge and corner features; deep networks identified fissure bifurcation patterns and spatial distribution characteristics. At the network's end, fully connected layers mapped abstract features to microstructural classification results (e.g., tension fissures, shear fissures, and healed fissures). During training, a transfer learning method was employed, using a pre-trained glacier feature extraction model as the base network. Efficient learning on small sample data was achieved by freezing the bottom-level parameters and fine-tuning the high-level network.

[0116] High-precision analysis of the microstructure of glacier surfaces is achieved through multi-level feature analysis techniques. Combining texture quantification analysis with recognition algorithms overcomes the accuracy limitations of traditional visual interpretation, automatically identifying millimeter-level ice fissures and quantifying their spatial distribution patterns. By using deep learning models to mine features from the fissure network, the intrinsic correlation between microstructure and macroscopic motion is established, providing data support for revealing the stress distribution and ice fracturing mechanisms within glaciers.

[0117] In step S17, the three-dimensional motion description and the microstructure features are subjected to spatiotemporal correlation analysis, and the spatiotemporal correlation analysis results and the spatial vector features are integrated to obtain preliminary glacier dynamic information.

[0118] In one specific implementation, the step of performing spatiotemporal correlation analysis on the three-dimensional motion description and the microstructural features, and integrating the spatiotemporal correlation analysis results with the spatial vector features to obtain preliminary glacier dynamic information includes:

[0119] The spatiotemporal correlation analysis of the three-dimensional motion description and the microstructure features is performed by a dynamic time warping algorithm to obtain the spatiotemporal correlation results.

[0120] A weighted fusion algorithm is used to integrate the spatiotemporal correlation results and the spatial vector features to obtain preliminary glacier dynamic information.

[0121] Specifically, in implementing spatiotemporal correlation analysis, a spatiotemporal correlation result is generated through dynamic time warping algorithm and spatial coupling modeling. This result includes a dynamic matching matrix in the time dimension and a stress coupling coefficient matrix in the spatial dimension. Specifically, the dynamic time warping algorithm establishes the optimal time alignment relationship between three-dimensional motion time series data (such as monthly average velocity sequences) and microstructure time series data (such as fracture density change curves) through nonlinear path optimization, generating a time offset matrix—this matrix quantifies the time lag relationship between the dynamic events of the two types of data (such as the number of days of delay between glacier acceleration events and fracture density peaks). Simultaneously, a stress coupling model is constructed based on spatial correlation analysis to calculate the spatial coupling coefficient matrix between microfracture network characteristics (such as fracture orientation distribution entropy) and macroscopic motion field parameters (such as velocity gradient tensors). Each element value in this matrix represents the contribution intensity of microstructural changes to macroscopic motion at a specific spatial location (e.g., 0.8 indicates that fracture propagation at that location contributes 80% of the local deformation). The core elements of the spatiotemporal correlation results include: the event synchronicity index in the time dimension (reflecting the temporal correlation strength of motion-structure evolution), the stress transfer efficiency coefficient in the spatial dimension (quantifying the degree of micro-macro dynamic coupling), and the energy transfer spectrum across scales (characterizing the interaction mode of motion-structure at different spatial frequencies). For example, the spatiotemporal correlation results for a certain region show that the event synchronicity index reaches 0.92 (with a maximum value of 1), the stress transfer efficiency coefficient is 0.75, and the energy spectrum exhibits significant resonance peaks in the 10-100 meter wavelength range, indicating that there is strong spatiotemporal coupling between micro-fracture propagation and macro-ice flow motion in this region, and that fracture network evolution dominates the mesoscale motion characteristics.

[0122] Based on the spatiotemporal correlation results, an adaptive weighted fusion algorithm is used to integrate the spatiotemporal correlation results and the spatial vector features. The algorithm dynamically allocates fusion weights to each data source according to the confidence level of the 3D motion field, the spatial density of microstructural features, and the continuity index of the spatial vector features. For example, in areas with high sensor coverage, the weight of 3D motion data is increased; in areas with dense fractures, the weight of microstructural features is increased. Through spatial superposition and temporal synchronization of multi-dimensional features, preliminary glacier dynamic information including motion velocity, direction, and structural stability is generated.

[0123] This process breaks through the limitations of analysis based on a single data dimension, enabling the collaborative interpretation of the macroscopic motion and microscopic structural evolution of glaciers. Through spatiotemporal correlation analysis, it reveals the causal relationship between stress transmission within the ice body and surface deformation. The weighted fusion mechanism effectively unifies the differences in physical dimensions and resolution among multi-source data, constructing a dynamic information system that can reflect both instantaneous motion states and trace long-term evolutionary patterns.

[0124] In step S18, when abnormal fluctuations occur in the preliminary glacier dynamic information, the preliminary glacier dynamic information is corrected using pre-stored historical remote sensing data to determine the trend of glacier flow velocity changes.

[0125] Specifically, in step S18, when the preliminary dynamic information detects abnormal fluctuations (such as a sudden increase in velocity of more than 2 times in a local area or a sudden change in the direction of motion exceeding 30 degrees), the system automatically calls the historical remote sensing database for data verification. By comparing the current abnormal pattern with the historical event database (such as the motion characteristics of glacial surges and ice avalanche precursor events), the Dynamic Time Warping (DTW) algorithm is used to select the closest historical data pattern to determine the physical rationality of the current abnormal value. If it is determined to be a false anomaly caused by sensor noise or short-term interference, a correction coefficient is generated based on the spatiotemporal evolution law of historical data to normalize and calibrate the current data.

[0126] After data correction, the Kalman filter algorithm was applied for time series optimization. This algorithm establishes a state prediction model of glacier movement, predicting the current value based on the movement state at the previous moment, and then fusions the actual observations and predictions using probability weighting. During the filtering process, the model dynamically adjusts the prediction weights based on data quality: when the reliability of the observation data is high (such as optical images under clear weather), the observed values ​​are emphasized; when the data noise is high (such as during periods of cloud interference), the model's predictions are relied upon. Through iterative optimization over multiple time steps, short-term interference fluctuations are eliminated, and low-frequency signals reflecting the true movement trend of the glacier are extracted.

[0127] This step establishes a full-chain data processing framework of "historical verification - real-time correction - trend extraction," solving the problem of confusion between short-term interference and long-term trends in remote sensing monitoring. Through a historical data-driven anomaly detection mechanism, it effectively distinguishes between real glacier movement events and observational noise; the time-varying weight strategy of Kalman filtering improves the physical rationality of trend analysis while ensuring data continuity.

[0128] In step S19, the glacier flow velocity change trend and the microstructure characteristics are input into a pre-trained dynamic prediction model to obtain complete glacier dynamic information.

[0129] In one specific implementation, the training process of the dynamic prediction model includes:

[0130] The historical glacier flow velocity variation trend and historical microstructure characteristics are obtained, and the data is cleaned and features are extracted from the historical glacier flow velocity variation trend and historical microstructure characteristics to obtain data features;

[0131] The data features are input into the input layer of the initially constructed neural network model for training, and the predicted values ​​output by the output layer of the neural network model are obtained.

[0132] Substitute the predicted value and the pre-stored actual value into the loss function to calculate the loss value;

[0133] The gradient of the output layer of the neural network model is calculated based on the loss value, and the gradient is passed forward layer by layer through the chain rule to calculate the gradient of the parameters of each layer and obtain the gradient data.

[0134] Update the parameters of each layer of the neural network model based on gradient data and a preset learning rate.

[0135] The parameters of each layer are iteratively updated until the number of training iterations of the neural network model is greater than a preset number of iterations, or the loss value of the neural network model is less than a preset loss threshold. At this point, the training is considered complete, and a dynamic prediction model is obtained.

[0136] Specifically, the training of the dynamic prediction model is achieved through the following technical path: A multi-dimensional training dataset is constructed based on historical glacier flow velocity trends and historical microstructural characteristics. First, the original data is standardized to eliminate dimensional differences. Principal component analysis is used to extract characteristic parameters reflecting the essence of glacier movement (such as flow rate distribution, fracture density parameters, and directional consistency index) to form a dimensionality-reduced feature vector. The processed feature vector is then input into a deep neural network model consisting of an input layer, hidden layers, and an output layer. The number of nodes in the input layer corresponds to the dimension of the feature vector. The hidden layer uses the ReLU activation function to achieve nonlinear mapping, and the output layer uses a linear... The system generates predicted values ​​for future glacier dynamic parameters. During training, the Adam optimization algorithm is used, with mean squared error as the loss function. Forward propagation calculates the deviation between predicted and actual observations, while backpropagation calculates the gradient information of each layer's parameters. The chain rule is used to propagate gradient data layer by layer from the output layer to the input layer. The learning rate is dynamically adjusted based on the gradient magnitude, and L2 regularization is used to suppress overfitting. Training terminates when any of the following conditions are met: 1) The decrease in validation set loss is less than a preset convergence threshold (e.g., 1%) within a preset number of iterations (e.g., 10); 2) The total number of training iterations reaches a preset maximum number of iterations (e.g., 1000). The resulting dynamic prediction model can model the spatiotemporal coupling relationship between the glacier's dynamic field and microstructural features, achieving generalized prediction of the co-evolutionary laws of multiple physics fields. In the actual prediction stage, the acquired glacier flow velocity change trend and microstructural features are input into the trained model. The model analyzes the nonlinear coupling mechanism between macroscopic motion and microscopic evolution by learning the feature mapping relationship in the hidden layer, and outputs complete dynamic information including the glacier's motion speed, probability of direction change, and structural stability assessment in future periods.

[0137] By overcoming the limitations of linear assumptions in traditional statistical models through a deep learning framework, this model achieves end-to-end prediction of multi-scale dynamic characteristics of glaciers. The model autonomously mines hidden patterns of ice fracturing precursors and abrupt changes in motion within historical data. This addresses the issues of missing dimensions and insufficient precision in traditional glacier monitoring, providing more comprehensive and accurate information on glacier dynamics.

[0138] Reference Figure 2 The second embodiment of the present invention provides a glacier flow velocity monitoring system based on multi-source remote sensing data, comprising:

[0139] The data acquisition module is used to acquire optical images, radar data, height models, time-series information, and glacier surface images;

[0140] The data fusion module is used to perform data fusion analysis on the optical images, radar data, height model and time series information to obtain the glacier change trend;

[0141] The vector construction module is used to calculate the two-dimensional displacement vector of the glacier surface based on the glacier change trend, and to construct a three-dimensional flow velocity vector in combination with the height model;

[0142] The vector analysis module is used to extract feature point coordinates from the optical image and the radar data, and superimpose them onto the boundaries of a pre-established grid cell. Based on the superposition result, the displacement vector within each grid is calculated and spatial analysis is performed to obtain spatial vector features.

[0143] The three-dimensional motion analysis module is used to extract the glacier's motion direction and velocity distribution based on the three-dimensional velocity vector. If the velocity change exceeds a preset velocity change threshold, the velocity distribution is supplemented by local region interpolation to obtain a continuous three-dimensional motion description.

[0144] The microscopic feature analysis module is used to extract texture features and crack features from the glacier surface image, and to analyze the surface microstructure using a convolutional neural network to obtain microstructural features;

[0145] The preliminary dynamic analysis module is used to perform spatiotemporal correlation analysis on the three-dimensional motion description and the microstructure features, and to integrate the spatiotemporal correlation analysis results with the spatial vector features to obtain preliminary glacier dynamic information;

[0146] The data correction module is used to correct the preliminary glacier dynamic information by using pre-stored historical remote sensing data when abnormal fluctuations occur in the preliminary glacier dynamic information, and to determine the trend of glacier flow velocity changes.

[0147] A complete dynamic analysis module is used to input the glacier flow velocity change trend and the microstructural features into a pre-trained dynamic prediction model to obtain complete glacier dynamic information.

[0148] It should be noted that the glacier flow velocity monitoring device based on multi-source remote sensing data provided in this embodiment of the invention is used to execute all the process steps of the glacier flow velocity monitoring method based on multi-source remote sensing data in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0149] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a glacier flow velocity monitoring program based on multi-source remote sensing data. When the processor executes the computer program, it implements the steps described in the various embodiments of the glacier flow velocity monitoring method based on multi-source remote sensing data, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as a glacier flow velocity monitoring module based on multi-source remote sensing data.

[0150] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0151] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0152] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0153] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0154] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0155] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0156] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for monitoring glacier flow velocity based on multi-source remote sensing data, characterized in that, The method comprises the following steps: acquiring optical images, radar data, height models, time series information and glacier surface images; performing data fusion analysis on the optical images, radar data, height models and time series information to obtain a glacier change trend; calculating a two-dimensional displacement vector of the glacier surface according to the glacier change trend, and constructing a three-dimensional flow velocity vector in combination with the height model; extracting feature point coordinates from the optical images and the radar data, and superimposing the feature point coordinates onto a pre-established grid cell boundary, calculating a displacement vector in each grid according to the superimposition result, and performing spatial analysis to obtain spatial vector features; extracting a glacier movement direction and a speed distribution according to the three-dimensional flow velocity vector, and if the speed change exceeds a preset speed change threshold, supplementing the speed distribution by a local area interpolation method to obtain a continuous three-dimensional movement description; extracting texture features and crack features from the glacier surface images, and analyzing the surface microstructure by using a convolutional neural network to obtain microstructure features; performing spatio-temporal correlation analysis on the three-dimensional movement description and the microstructure features, integrating the spatio-temporal correlation analysis result with the spatial vector features, and obtaining preliminary glacier dynamic information; when the preliminary glacier dynamic information appears abnormal fluctuation, correcting the preliminary glacier dynamic information by using pre-stored historical remote sensing data to determine a glacier flow velocity change trend; inputting the glacier flow velocity change trend and the microstructure features into a pre-trained dynamic prediction model to obtain complete glacier dynamic information.

2. The method for monitoring glacier flow velocity based on multi-source remote sensing data according to claim 1, characterized in that, The data fusion analysis on the optical images, the radar data, the height models and the time series information to obtain the glacier change trend comprises the following steps: spatio-temporal alignment of the optical images, the radar data, the height models and the time series information by using a linear interpolation method to obtain spatio-temporal alignment data; extracting a main change trend from the spatio-temporal alignment data by using a principal component analysis method to obtain a glacier change trend, wherein the glacier change trend comprises a displacement change trend and a height change trend of the glacier surface.

3. The method for monitoring glacier flow velocity based on multi-source remote sensing data according to claim 1, characterized in that, The calculation of the two-dimensional displacement vector of the glacier surface according to the glacier change trend and the construction of the three-dimensional flow velocity vector in combination with the height model comprise the following steps: calculation of a two-dimensional displacement vector of the glacier surface by using a Lucas-Kanade algorithm according to the glacier change trend; construction of a three-dimensional flow velocity vector by using a rigid body transformation method according to the two-dimensional displacement vector and the height model, and smoothing of the three-dimensional flow velocity vector by using a Kalman filtering algorithm to obtain a smoothed three-dimensional flow velocity vector.

4. The method for monitoring glacier flow velocity based on multi-source remote sensing data according to claim 1, characterized in that, The extraction of feature point coordinates from the optical images and the radar data, the superimposition of the feature point coordinates onto a pre-established grid cell boundary, the calculation of a displacement vector in each grid according to the superimposition result, and the spatial analysis to obtain spatial vector features comprise the following steps: extraction of feature points with invariant scales from the optical images by using a rotation invariant feature description method; extraction of stable scattering feature point coordinates from the radar data by using a permanent scatterer identification method; The coordinates of the invariant feature points and the scattering feature points are matched by using a nearest neighbor search method, the coordinates of the feature points that are successfully matched are converted into UTM projection coordinates by using a coordinate system conversion method, and the feature points are superimposed on the boundaries of pre-established grid cells by using a spatial buffer analysis method to obtain a superimposition result; According to the superimposition result, a displacement vector in each grid is obtained by using a vector calculation method, and a displacement fluctuation intensity and a direction disorder degree are extracted as spatial vector features according to the displacement vector.

5. The method for monitoring glacier flow velocity based on multi-source remote sensing data according to claim 1, characterized in that, The texture features and the crack features are extracted from the glacier surface image, and a convolutional neural network is used to analyze the surface microstructure to obtain microstructure features, including: The texture features are extracted from the glacier surface image by using a gray level co-occurrence matrix algorithm; According to the texture features, a Canny edge detection algorithm is used to extract crack features; According to the crack features, a convolutional neural network is used to analyze the surface microstructure of the glacier to obtain microstructure features.

6. The method for monitoring glacier flow velocity based on multi-source remote sensing data according to claim 1, characterized in that, The three-dimensional motion description and the microstructure features are analyzed in space-time correlation, and the space-time correlation analysis result and the spatial vector features are integrated to obtain preliminary glacier dynamic information, including: The three-dimensional motion description and the microstructure features are analyzed in space-time correlation by using a dynamic time warping algorithm to obtain a space-time correlation result; The space-time correlation result and the spatial vector features are integrated by using a weighted fusion algorithm to obtain preliminary glacier dynamic information.

7. The method for monitoring glacier flow velocity based on multi-source remote sensing data according to claim 1, characterized in that, The training process of the dynamic prediction model includes: Obtaining historical glacier flow rate change trends and historical microstructure features, and performing data cleaning and feature extraction on the historical glacier flow rate change trends and the historical microstructure features to obtain data features; The data features are input into the input layer of the initially constructed neural network model for training, and the predicted values output by the output layer of the neural network model are obtained; The predicted values and pre-stored actual values are substituted into a loss function to calculate a loss value; The gradient output by the output layer of the neural network model is calculated according to the loss value, and the gradient is transmitted forward layer by layer by using a chain rule to calculate the gradient of each layer parameter to obtain gradient data; According to the gradient data and a preset learning rate, the parameters of each layer of the neural network model are updated; Each layer parameter is repeatedly iterated and updated until the training of the neural network model is completed when the number of training times of the neural network model is greater than a preset number, or when the loss value data of the neural network model is less than a preset loss threshold, and a dynamic prediction model is obtained.

8. A glacier flow velocity monitoring system based on multi-source remote sensing data, characterized in that, including: A data acquisition module is configured to acquire optical images, radar data, height models, time series information, and glacier surface images; A data fusion module is configured to perform data fusion analysis on the optical images, radar data, height models, and time series information to obtain a glacier change trend; A vector construction module is configured to calculate a two-dimensional displacement vector of a glacier surface according to the glacier change trend, and to construct a three-dimensional flow rate vector in combination with the height model; a vector analysis module configured to extract feature point coordinates from the optical image and the radar data, superimpose the feature point coordinates on a pre-established grid cell boundary, calculate a displacement vector in each grid cell based on a superimposition result, and perform spatial analysis to obtain a spatial vector feature; a three-dimensional motion analysis module configured to extract a glacier motion direction and a velocity distribution based on the three-dimensional flow velocity vector, supplement the velocity distribution by using a local area interpolation method if a velocity variation exceeds a preset velocity variation threshold, and obtain a continuous three-dimensional motion description; a micro-feature analysis module configured to extract texture features and crack features from the optical image of the glacier surface, analyze a surface microstructure by using a convolutional neural network, and obtain microstructure features; a preliminary dynamic analysis module configured to perform a spatio-temporal correlation analysis on the three-dimensional motion description and the microstructure features, integrate a spatio-temporal correlation analysis result and the spatial vector feature, and obtain preliminary glacier dynamic information; a data correction module configured to correct the preliminary glacier dynamic information by using pre-stored historical remote sensing data when the preliminary glacier dynamic information has an abnormal fluctuation, and determine a glacier flow velocity variation trend; a complete dynamic analysis module configured to input the glacier flow velocity variation trend and the microstructure features into a pre-trained dynamic prediction model, and obtain complete glacier dynamic information.

9. An electronic device, comprising: A computer readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the glacier flow velocity monitoring method based on multi-source remote sensing data according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the glacier flow velocity monitoring method based on multi-source remote sensing data according to any one of claims 1 to 7.

Images