A remote sensing data multi-scale spatio-temporal fusion method and system and a storage medium

Abstract

This application relates to a method, system, and storage medium for multi-scale spatiotemporal fusion of remote sensing data. The method includes: acquiring a first remote sensing dataset with a first spatial resolution, a second remote sensing dataset with a second spatial resolution, and background field data with a second resolution; establishing a spatiotemporal scale transformation model based on the data from both datasets during overlapping periods, generating a parameter set containing scale transformation parameters and quality evaluation parameters; aligning the parameter set with the background field data through coordinate orientation correction and spatial interpolation; dynamically determining continuous fusion weights based on the quality evaluation parameters, and performing weighted fusion of the model output and the background field data to generate preliminary fused data; and finally, performing independent smoothing and denoising processing on the time-series fused data. This method can solve problems such as mismatched parameter spatial matching, spatial discontinuity of fusion results, and poor engineering adaptability caused by high process coupling, achieving spatially continuous, temporally smooth, and easily scaled-up remote sensing data fusion.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing data processing and geoscience information engineering technology, and in particular relates to a method, system and storage medium for multi-scale spatiotemporal fusion of remote sensing data. Background Technology

[0002] With the development of remote sensing observation technology, multi-scale spatiotemporal fusion technology has emerged. This technology aims to solve the problem that one type of dataset has a long time coverage and continuous time series but low spatial resolution, while another type of dataset has high spatial resolution but short time coverage. This leads to the traditional fusion method of scale transformation based on statistical regression.

[0003] Traditional techniques typically involve establishing a scale transformation model at a low spatial resolution scale and then applying the obtained model parameters to a high spatial resolution grid through interpolation or resampling. When using regression models for fusion, a hard threshold judgment strategy is adopted based on the statistical significance test results. Some methods also introduce complex time series smoothing or filtering operations during the fusion process.

[0004] However, the current statistical regression-based scaling method has the following problems: First, there is the problem of parameter space matching error. Direct resampling can easily lead to misalignment or incomplete coverage between model parameters and the target high-resolution grid, resulting in invalid or erroneous fusion results. Second, there is the problem of spatial discontinuity in the fusion results. Hard threshold judgment strategies can easily form discontinuous boundaries and block artifacts when the reliability of local areas is insufficient, which can destroy the spatial consistency of the fusion results. Third, there is the problem of high coupling in the computational process and insufficient engineering adaptability. The highly coupled processing steps lead to a complex computational process and high memory consumption, which is not conducive to the engineering processing of large-scale, long-term remote sensing data. Summary of the Invention

[0005] Therefore, it is necessary to provide a method, system, and storage medium for multi-scale spatiotemporal fusion of remote sensing data that can effectively solve parameter space matching errors, improve the spatial continuity of fusion results, and take into account computational efficiency and engineering adaptability, in order to address the above-mentioned technical problems.

[0006] Firstly, this application provides a method for multi-scale spatiotemporal fusion of remote sensing data, including:

[0007] S1. Acquire the first remote sensing dataset with the first spatial resolution, the second remote sensing dataset with the second spatial resolution, and the background field data with the second spatial resolution;

[0008] S2. Based on the aggregated data of the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, a spatiotemporal scale transformation model is constructed, resulting in a parameter set containing the first spatial resolution. The parameter set includes scale transformation parameters and quality evaluation parameters, which are used to characterize the reliability of the spatiotemporal scale transformation model at the corresponding spatial location.

[0009] S3. Perform spatial adaptive alignment on the parameter set and background field data to make the spatial coordinate grid of the parameter set consistent with the spatial coordinate grid of the background field data, and obtain the aligned parameter set and the aligned background field data.

[0010] S4. For the target spatiotemporal unit, determine the fusion weight based on the quality evaluation parameters in the aligned parameter set, and perform weighted fusion on the aligned parameter set and the aligned background field data based on the fusion weight to generate preliminary fused data containing the second spatial resolution.

[0011] S5. After generating the initial fusion data of multiple target spatiotemporal units, the initial fusion data corresponding to the multiple target spatiotemporal units are smoothed and filtered for noise reduction according to the time series to obtain fused remote sensing data.

[0012] In one embodiment, a spatiotemporal scale transformation model is constructed based on aggregated data from the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, resulting in a parameter set containing a first spatial resolution, including:

[0013] S11. Extract the time series values ​​of the first remote sensing dataset within the overlapping time period on each grid cell of the first spatial resolution.

[0014] S12. Based on the geographical range of the first spatial resolution grid cell, spatially average all high-resolution pixel values ​​of the second remote sensing dataset that fall within the geographical range to generate a resolution aggregated time series corresponding to the time points of the first remote sensing dataset.

[0015] S13. For each first spatial resolution grid cell, the time series value of the first remote sensing dataset is used as the independent variable, and the corresponding resolution aggregated time series value is used as the dependent variable for fitting to obtain a linear regression model.

[0016] S14. Extract regression coefficients from the linear regression model as scaling parameters, and calculate the coefficient of determination of the linear regression model as a quality evaluation parameter.

[0017] S15. Collect the scale transformation parameters and quality evaluation parameters corresponding to all grid cells of the first spatial resolution to generate a parameter set containing the first spatial resolution.

[0018] In one embodiment, spatial adaptive alignment is performed on the parameter set and background field data to make the spatial coordinate grid of the parameter set consistent with the spatial coordinate grid of the background field data, resulting in aligned parameter set and aligned background field data, including:

[0019] S21. Obtain background field data containing the second spatial resolution, and determine the spatial coordinate axis direction of the grid metadata of the parameter set and the grid metadata of the background field data to obtain the direction determination result.

[0020] S22. When the direction judgment result is that the parameter set is inconsistent with the background field data in at least one coordinate axis direction, perform a matrix flip operation on the parameter set along the corresponding coordinate axis to make the coordinate axis direction of the parameter set consistent with the coordinate axis direction of the background field data, and obtain the parameter set after direction correction.

[0021] S23. When the direction judgment result is that the parameter set is consistent with the background field data in the coordinate axis direction, perform spatial interpolation mapping on the parameter set and the parameter set after direction correction, and resample the parameter set and the parameter set after direction correction from the first spatial resolution to the second spatial resolution to obtain the parameter set after interpolation mapping.

[0022] S24. Output the orientation-corrected parameter set, the interpolated parameter set, and the background field data as aligned parameter sets and aligned background field data with the same spatial coordinate grid.

[0023] In one embodiment, for a target spatiotemporal unit, a fusion weight is determined based on quality evaluation parameters in the aligned parameter set, and the aligned parameter set and the aligned background field data are weighted and fused based on the fusion weight to generate preliminary fused data containing a second spatial resolution, including:

[0024] S31. For the target pixels and corresponding target time points on the second spatial resolution grid, extract the scale transformation parameters and quality evaluation parameters corresponding to the target pixel positions from the aligned parameter set.

[0025] S32. Obtain the first remote sensing data value of the low-resolution pixel covering the area where the target pixel is located at the target time point;

[0026] S33. Using the scale transformation parameters and the first remote sensing data value, calculate the initial prediction value of the spatiotemporal scale transformation model at the target pixel;

[0027] S34. Obtain the aligned background field data values ​​at the target time point and target pixel location;

[0028] S35. Based on the values ​​of the quality evaluation parameters, the fusion weight is calculated through a preset weight mapping function;

[0029] S36. Use fusion weights to linearly weight the initial predicted value and the background field data value, calculate the preliminary fusion value of the target pixel at the target time point, and obtain the preliminary fusion data.

[0030] In one embodiment, the fusion weight is calculated based on the values ​​of the quality evaluation parameters using a preset weight mapping function, including:

[0031] S41. Based on the numerical distribution characteristics of the quality evaluation parameters, dynamically determine the first threshold and the second threshold; wherein the first threshold is less than the second threshold;

[0032] S42. When the quality evaluation parameter value is less than the first threshold, the fusion weight is 0.

[0033] S43. When the quality evaluation parameter value is greater than or equal to the first threshold and less than or equal to the second threshold, the fusion weight is calculated using a linear interpolation function; wherein, the expression of the linear interpolation function is:

[0034]

[0035] In the formula, To integrate weights, These are quality evaluation parameter values. The first threshold, The second threshold;

[0036] S44. When the quality evaluation parameter value is greater than the second threshold, the fusion weight is 1.

[0037] In one embodiment, the method further includes:

[0038] S51. When processing remote sensing data over a large area, the parameter set and background field data are spatially divided into multiple sub-region blocks.

[0039] S52. Based on the parameter set subsets and background field data subsets corresponding to multiple sub-region blocks, perform coordinate direction consistency judgment, coordinate flip correction, and bilinear interpolation spatial resampling mapping to obtain the parameter subsets after alignment of each sub-region block.

[0040] S53. After all sub-region blocks have been processed, the parameter subsets after the sub-region blocks are aligned are spliced ​​together according to the spatial correspondence to generate the aligned parameter set.

[0041] In one embodiment, after generating preliminary fused data for multiple target spatiotemporal units, the preliminary fused data corresponding to the multiple target spatiotemporal units is smoothed and denoised according to the time series to obtain fused remote sensing data, including:

[0042] S61. Obtain preliminary fusion data of multiple target spatiotemporal units to form a preliminary fusion time series;

[0043] S62. Perform convolution smoothing on the preliminary fused time series to obtain the smoothed time series;

[0044] S63. Perform outlier detection and correction on the smoothed time series to obtain the time series with outliers removed.

[0045] S64. Arrange the time series data after anomaly removal according to the original spatial grid structure and time order to generate fused remote sensing data.

[0046] Secondly, this application also provides a multi-scale spatiotemporal fusion system for remote sensing data, comprising:

[0047] The multidimensional data acquisition module is used to acquire a first remote sensing dataset with a first spatial resolution, a second remote sensing dataset with a second spatial resolution, and background field data with a second spatial resolution.

[0048] The parameter modeling module is used to construct a spatiotemporal scale transformation model based on aggregated data from the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, and to obtain a parameter set containing the first spatial resolution. The parameter set includes scale transformation parameters and quality evaluation parameters, which are used to characterize the reliability of the spatiotemporal scale transformation model at the corresponding spatial location.

[0049] The spatial adaptive alignment module is used to perform spatial adaptive alignment of the parameter set and the background field data, so that the spatial coordinate grid of the parameter set is consistent with the spatial coordinate grid of the background field data, and thus obtain the aligned parameter set and the aligned background field data.

[0050] The adaptive weighted fusion module is used to determine the fusion weight based on the quality evaluation parameters in the aligned parameter set for the target spatiotemporal unit, and to perform weighted fusion on the aligned parameter set and the aligned background field data based on the fusion weight to generate preliminary fused data containing the second spatial resolution.

[0051] The fused remote sensing data generation module is used to smooth and filter the preliminary fused data corresponding to multiple target spatiotemporal units according to the time series after completing the initial fused data generation of multiple target spatiotemporal units, so as to obtain fused remote sensing data.

[0052] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the method described in the first aspect.

[0053] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0054] The aforementioned method, system, and storage medium for multi-scale spatiotemporal fusion of remote sensing data effectively eliminates parameter spatial matching errors caused by differences in different remote sensing data grid systems through spatial adaptive alignment processing, ensuring that the parameter set is consistent with the spatial coordinate grid of the background field data and avoiding invalid or erroneous fusion results. By employing a continuous weight fusion strategy based on quality evaluation parameters and dynamic threshold adjustment, replacing traditional hard threshold judgment, it achieves a smooth spatial transition of the fusion results, improves spatial continuity, and eliminates block artifacts and discontinuous boundaries. A block processing mechanism addresses the memory bottleneck of large-scale data processing, while decoupling temporal post-processing from the core fusion steps, reducing computational coupling and enhancing engineering adaptability. This method combines good versatility and regional robustness, adaptable to the fusion of various remote sensing parameters, without relying on specific sensors or data products, ultimately generating fused remote sensing data with high spatial consistency, temporal continuity, and strong practicality. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0056] Figure 1 This is a flowchart illustrating a multi-scale spatiotemporal fusion method for remote sensing data in one embodiment;

[0057] Figure 2 This is a schematic diagram of the process of fusion processing based on multi-source remote sensing data in one embodiment;

[0058] Figure 3 This is a schematic diagram of the structure of a multi-scale spatiotemporal fusion system for remote sensing data in one embodiment. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0060] In one embodiment, reference Figure 1The document presents a flowchart illustrating a multi-scale spatiotemporal fusion method for remote sensing data provided in this application. This embodiment uses the application of this method to a remote sensing data fusion terminal (hereinafter referred to as the terminal) as an example. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0061] S1. Obtain the first remote sensing dataset with the first spatial resolution, the second remote sensing dataset with the second spatial resolution, and the background field data with the second spatial resolution.

[0062] For example, the first remote sensing dataset acquired by the remote sensing data fusion terminal comes from platforms with continuous observation capabilities, such as long-term satellite observation systems and meteorological satellite monitoring networks. Its core function is to provide complete time-series information. The first spatial resolution is relatively low, capable of covering long-term dynamic changes over a large area. The second remote sensing dataset comes from high-precision observation methods such as high-resolution remote sensing satellites and airborne remote sensing. The second spatial resolution is higher than the first spatial resolution, accurately capturing subtle spatial heterogeneity features of the Earth's surface. It also overlaps with the first remote sensing dataset in time, providing a matching basis for subsequent model construction. Background field data is acquired from sources such as historical remote sensing data product libraries, geospatial data cloud platforms, and reanalysis datasets. It must have a second spatial resolution and meet the requirements of spatial continuity and temporal stability, serving as a reference benchmark in the fusion process. All acquired datasets must undergo format standardization to ensure data compatibility and usability.

[0063] S2. Based on the aggregated data of the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, a spatiotemporal scale transformation model is constructed to obtain a parameter set containing the first spatial resolution.

[0064] The parameter set includes scale transformation parameters and quality evaluation parameters. The quality evaluation parameters are used to characterize the reliability of the spatiotemporal scale transformation model at the corresponding spatial location.

[0065] For example, the remote sensing data fusion terminal first extracts the overlapping time interval between the first and second remote sensing datasets, and performs data aggregation processing based on this interval. The aggregation process follows the principle of spatial scale matching, that is, according to the geographical range of the grid units of the first remote sensing dataset, the high-resolution pixels of the second remote sensing dataset falling within this range are statistically aggregated to form aggregated data with the same spatial resolution as the first remote sensing dataset. Subsequently, a spatiotemporal scale conversion model is constructed based on the principle of statistical regression analysis. The core logic of this model is to establish a quantitative correspondence between low spatial resolution data and high spatial resolution aggregated data. In the model construction process, the time series values ​​of the first remote sensing dataset are used as independent variables, and the corresponding aggregated data are used as dependent variables. The scale conversion parameters of the model are obtained through data fitting operations. These parameters are the core quantitative indicators for realizing the conversion of data at different scales. At the same time, quality evaluation parameters are calculated, including the coefficient of determination (R², an indicator reflecting the goodness of fit of the model) and fitting error indicators. Their core function is to quantitatively characterize the predictive reliability of the model at various spatial locations and avoid fusion errors caused by insufficient local reliability of the model. Finally, the scale transformation parameters and corresponding quality evaluation parameters of all grid cells are integrated to form a parameter set with first spatial resolution, providing basic parameter support for the subsequent fusion process.

[0066] S3. Perform spatial adaptive alignment on the parameter set and background field data to make the spatial coordinate grid of the parameter set consistent with the spatial coordinate grid of the background field data, thus obtaining the aligned parameter set and the aligned background field data.

[0067] For example, the remote sensing data fusion terminal first extracts grid metadata from the parameter set and background field data, including core information such as coordinate system, grid origin, pixel size, and coordinate axis orientation. Based on geospatial matching theory, it performs a consistency judgment on the spatial coordinate axis orientations of the two data sets. When the judgment result shows that the parameter set is inconsistent with the background field data in one or more coordinate axis orientations, a matrix flipping algorithm is used to perform flipping correction on the parameter set along the corresponding coordinate axes to ensure that the coordinate axis orientations of the two data sets are completely consistent. If the coordinate axis orientations are consistent, spatial interpolation mapping is directly performed. The interpolation method uses high-precision interpolation such as bilinear interpolation or spline interpolation to resample the parameter set from the first spatial resolution to the second spatial resolution, achieving spatial scale matching with the background field data. For large-scale data processing scenarios, a block processing mechanism is adopted. According to preset spatial division rules, the parameter set and background field data are divided into multiple independent sub-region blocks. Orientation judgment, correction, and interpolation mapping operations are performed block by block. After processing, the sub-region blocks are stitched together according to the spatial location correspondence, finally outputting an aligned parameter set and aligned background field data with completely consistent spatial coordinate grids, eliminating parameter misalignment problems caused by differences in the grid system.

[0068] S4. For the target spatiotemporal unit, determine the fusion weight based on the quality evaluation parameters in the aligned parameter set, and perform weighted fusion on the aligned parameter set and the aligned background field data based on the fusion weight to generate preliminary fused data containing the second spatial resolution.

[0069] For example, the remote sensing data fusion terminal first clarifies the specific range of the target spatiotemporal unit, i.e., a specific combination of spatial location and time node. Then, it extracts the scale transformation parameters and quality assessment parameters corresponding to the target spatiotemporal unit from the aligned parameter set. The determination of the fusion weight is based on a dynamic threshold adjustment strategy. The core principle of this strategy is to adaptively determine the first and second thresholds according to the overall distribution statistical characteristics of the quality assessment parameters, avoiding the problem of insufficient fusion adaptability caused by fixed thresholds. When the quality assessment parameters are lower than the first threshold, it indicates that the model's prediction reliability in that spatiotemporal unit is extremely low, and the fusion weight is set to 0. At this time, the fusion process is mainly based on the background field data. When the quality assessment parameters are higher than the second threshold, it indicates that the model's prediction results are highly reliable, and the fusion weight is set to 1. The fusion result is mainly based on the model output. When the quality assessment parameters are between the two thresholds, the fusion weight changes continuously and linearly with the quality assessment parameters, realizing a smooth transition between the model output and the background field data. After determining the fusion weights, based on the principle of combining multi-source data with different weights, the predicted values ​​obtained by scaling the aligned parameter set are linearly weighted with the aligned background field data to generate preliminary fusion data with a second spatial resolution, ensuring the spatial continuity and reliability of the fusion results.

[0070] S5. After generating the initial fusion data of multiple target spatiotemporal units, the initial fusion data corresponding to the multiple target spatiotemporal units are smoothed and filtered for noise reduction according to the time series to obtain fused remote sensing data.

[0071] For example, the remote sensing data fusion terminal first organizes the preliminary fusion data of all target spatiotemporal units in chronological order to construct a complete preliminary fusion time series. This series contains fusion data of various spatial locations within the study area at different time points. Subsequently, the time series data undergoes smoothing and denoising processes. Filtering methods can include moving average filtering, Savitzky-Golay filtering, or wavelet denoising, with the core objective of eliminating high-frequency noise caused by observation and model errors in the preliminary fusion data. After smoothing, an outlier detection algorithm is used to identify outliers deviating from the normal range. For detected outliers, reasonable corrections are made based on the normal data characteristics of adjacent time points to avoid the impact of outliers on the overall data quality. Finally, the smoothed, denoised, and outlier-corrected time series data is rearranged and integrated according to the original spatial grid structure and chronological order to generate fused remote sensing data with high spatial resolution, long-term continuity, and high data reliability, meeting the needs of subsequent geoscientific analysis, resource surveys, and other applications.

[0072] In the aforementioned multi-scale spatiotemporal fusion method for remote sensing data, spatial adaptive alignment processing is used to eliminate parameter spatial matching errors caused by differences in different remote sensing data grid systems, ensuring accurate matching between parameters and the target high-resolution grid and avoiding invalid or erroneous fusion results. A continuous weight fusion strategy based on quality evaluation parameters replaces the traditional hard threshold judgment, achieving a smooth spatial transition of the fusion result, effectively solving the problems of spatial discontinuity and block artifacts in the fusion result, and ensuring spatial consistency. By decoupling the core fusion steps from temporal post-processing and introducing a block processing mechanism, the coupling degree of the computational process is reduced, overcoming the memory bottleneck of large-scale, long-term data processing and improving engineering adaptability. Simultaneously, the dynamic threshold adjustment strategy allows the method to adapt to different data statistical characteristics, enhancing regional adaptability and robustness, and possessing good versatility, suitable for the fusion of various remote sensing parameters, providing high-quality data support for related applications.

[0073] To further illustrate the solutions of this application, a specific embodiment is described below in conjunction with the above-mentioned multi-scale spatiotemporal fusion method for remote sensing data. (Reference) Figure 2 The document presents a flowchart illustrating the fusion processing of multi-source remote sensing data provided in this application, including the following main steps:

[0074] 1. Input Data: First remote sensing data and second remote sensing data are used as initial inputs. Specifically, a first remote sensing dataset and a second remote sensing dataset are acquired, wherein the first remote sensing dataset has a first spatial resolution and is used to provide long-term series information; the second remote sensing dataset has a second spatial resolution higher than the first spatial resolution and is used to provide spatial detail information, and the first remote sensing dataset and the second remote sensing dataset have at least partial overlap in the time dimension;

[0075] 2. Model Building and Parameter Generation (S10): Establish a spatiotemporal scale transformation model and generate a parameter set for subsequent processing. Specifically, based on the spatiotemporal characteristics of the first and second remote sensing data, a scale transformation model is constructed to generate a parameter set suitable for different resolutions and time frequencies, including spatial resampling coefficients and temporal interpolation functions. Based on the aggregated data of the first remote sensing dataset and its corresponding second remote sensing dataset within the overlapping time period, a spatiotemporal scale transformation model is established to obtain a parameter set containing at least one scale transformation parameter and its corresponding quality evaluation parameter. The parameter set has a first spatial resolution.

[0076] 3. Background Field Data Acquisition (S30): Acquire auxiliary background field data for fusion correction. Specifically, select background field data with high spatial or temporal consistency (such as historical remote sensing products, reanalysis data, etc.) as a reference benchmark in the fusion process. Acquire background field data with a second spatial resolution.

[0077] 4. Spatial Alignment Processing (S40): Perform spatial adaptive alignment processing on the parameter set and background field data to ensure that the spatial coordinate grid of the parameter set is consistent with that of the background field data; determine whether the coordinate directions are consistent: if they are inconsistent, perform coordinate flip correction; if they are consistent, perform spatial interpolation mapping. For example, spatial matching is performed on the input data and the background field, and the coordinate system directions are determined to be consistent: if they are inconsistent, perform coordinate flip or reprojection correction; if they are consistent, use bilinear interpolation or spline interpolation methods to map the data to a unified grid. Specifically, this includes:

[0078] (1) Determine the consistency between the parameter set and the background field data in at least one spatial coordinate axis direction;

[0079] (2) When an inconsistency is determined, a flip correction is performed on the parameter set in the corresponding coordinate axis direction;

[0080] (3) Based on the spatial coordinate grid of the background field data, perform spatial interpolation mapping on the parameter set;

[0081] (4) Optionally, when processing large-scale data, a block processing mechanism is adopted to divide the parameter set and the background field data into multiple spatial blocks, and perform the alignment and mapping operations of steps (1)-(3) block by block to reduce memory usage.

[0082] 5. Adaptive Weighted Fusion (S50): For the target spatiotemporal unit, based on the aligned parameter set and background field data, an adaptive weighted fusion algorithm generates fused data with a second spatial resolution. The adaptive weighted fusion algorithm determines the fusion weights based on quality evaluation parameters. For example, the fusion weights are determined according to the goodness of fit (R²): R² < first threshold: weight = 0, background field data is dominant; first threshold ≤ R² ≤ second threshold: weights change linearly, weighted fusion is performed; R² > second threshold: weight = 1, model output is dominant. Specifically, the goodness of fit (R²) between the model output and the background field data is calculated, and the fusion weights are dynamically allocated according to a preset threshold: if R² is lower than the first threshold, the model output has low reliability, and the background field data is used entirely; if R² is between the two thresholds, linear weighted fusion is used, and the weights increase with R²; if R² is higher than the second threshold, the model output is reliable, and the model results are directly used. The quality evaluation parameters are used to characterize the reliability of the spatiotemporal scale conversion model at the corresponding spatial location. The quality evaluation parameters include, but are not limited to, regression determination coefficients, fitting error indices, or significance test indices. Specifically, this includes: dynamically determining the first and second thresholds based on the distributional statistical characteristics of quality evaluation parameters, and employing a three-region fusion strategy.

[0083] When the quality evaluation parameter is lower than the first threshold, the fusion weight is 0 or close to 0, and the fusion result is mainly based on the background field data.

[0084] When the quality evaluation parameter is higher than the second threshold, the fusion weight is taken as 1 or close to 1, and the fusion result is mainly based on the output of the spatiotemporal scale conversion model.

[0085] When the quality evaluation parameter is between the first threshold and the second threshold, the fusion weight changes continuously with the quality evaluation parameter.

[0086] 6. Generate fused data: Output preliminary fusion results. This includes:

[0087] (1) Determine the fusion weight based on the quality evaluation parameters. The fusion weight is a function of the quality evaluation parameters and its value ranges from 0 to 1.

[0088] (2) Based on the fusion weight, the initial output value of the spatiotemporal scale conversion model and the background field data are weighted and combined to obtain fused data.

[0089] 7. Time Series Post-processing (S60): After generating fused data from multiple target spatiotemporal units, time series post-processing operations are performed on the resulting fused time series dataset. These operations include smoothing filtering or denoising for the time series signal corresponding to each spatial location, and are independent of the adaptive weighted fusion algorithm. Specifically, moving average, Savitzky-Golay filtering, or wavelet denoising methods are applied to the fused time series data to eliminate high-frequency noise and improve data continuity.

[0090] Output fused product: Generate the final usable fused remote sensing product. Specifically, the processed data is integrated into a standardized format (such as GeoTIFF, NetCDF) to generate a fused remote sensing product that can be directly used for analysis and applications.

[0091] The aforementioned fusion processing method based on multi-source remote sensing data eliminates the problem of scale transformation parameter misalignment caused by differences in different remote sensing data grid systems by introducing parameter space adaptive alignment processing, ensuring the spatial integrity of the fusion process. The continuous weight fusion strategy based on quality evaluation parameters avoids spatial abrupt changes caused by hard threshold judgments, achieving a smooth spatial transition of the fusion results and significantly improving spatial continuity. By decoupling the core fusion steps from the time-series post-processing steps, the coupling degree of the computational process is reduced, improving the method's engineering adaptability in large-scale, long-term remote sensing data processing. The introduction of a block processing mechanism effectively solves the memory bottleneck problem in large-scale remote sensing data processing, supporting efficient fusion of TB-level data and enhancing engineering practicality. The adoption of a dynamic threshold adjustment strategy enables the fusion method to adapt to the statistical characteristics of data in different regions and seasons, improving the regional adaptability and robustness of the fusion results. It is applicable to data fusion of various remote sensing parameters (such as vegetation index, land surface temperature, etc.) and does not depend on specific sensors or specific data products.

[0092] In an optional embodiment, a spatiotemporal scale transformation model is constructed based on aggregated data from the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, resulting in a parameter set containing a first spatial resolution. This includes the following steps:

[0093] S11. Extract the time series values ​​of the first remote sensing dataset within the overlapping time period on each grid cell of the first spatial resolution.

[0094] Optionally, the remote sensing data fusion terminal first analyzes the spatial grid structure of the first remote sensing dataset, clarifying the geographic coordinate range and boundary information of each first spatial resolution grid unit, and extracts target data based on the geographic range to locate the spatial position corresponding to each grid unit. Then, it determines the overlapping time interval between the first and second remote sensing datasets. This interval is obtained by comparing the time coverage of the two datasets and extracting the temporal intersection to ensure temporal consistency in subsequent data processing. Based on the determined overlapping time interval, continuous data sequences are extracted from the time dimension of the datasets. All remote sensing observations within this time interval are extracted one by one for each grid unit and arranged in chronological order to form the time series values ​​corresponding to each grid unit. During the extraction process, data validity verification is required to remove invalid data caused by observation failures, atmospheric interference, or other factors, ensuring the integrity and reliability of the time series.

[0095] S12. Based on the geographical range of the first spatial resolution grid cell, spatially average all high-resolution pixel values ​​of the second remote sensing dataset that fall within the geographical range to generate a resolution aggregated time series corresponding to the time points of the first remote sensing dataset.

[0096] Optionally, the remote sensing data fusion terminal first acquires the precise geographic extent of each grid cell at the first spatial resolution, including latitude and longitude boundaries, projected coordinate ranges, and other geographic information. It then analyzes the overlay relationship of different spatial data layers to determine whether the high-resolution pixels of the second remote sensing dataset fall within this geographic extent. For all high-resolution pixels falling within the geographic extent of the target grid cell, their observation values ​​at the corresponding time points are extracted. These time points must be strictly consistent with the observation time points of the first remote sensing dataset to ensure temporal matching. Subsequently, a spatial averaging algorithm is used to statistically average the values ​​of all high-resolution pixels within the target grid cell at each time point. The aim is to aggregate the high spatial resolution data into data with the same spatial resolution as the first remote sensing dataset, eliminating the impact of spatial scale differences. The spatial averaging results at each time point are arranged chronologically to generate the resolution aggregated time series corresponding to that grid cell. This ensures that the aggregated data retains the spatial information corresponding to the high-resolution features of the second remote sensing dataset while achieving spatial scale matching with the first remote sensing dataset, providing a suitable data foundation for subsequent model construction.

[0097] S13. For each grid cell with the first spatial resolution, the time series value of the first remote sensing dataset is used as the independent variable, and the corresponding resolution aggregated time series value is used as the dependent variable for fitting to obtain a linear regression model.

[0098] Optionally, for each first spatial resolution grid cell, the remote sensing data fusion terminal extracts the time series values ​​of the first remote sensing dataset and the resolution aggregated time series values ​​corresponding to that cell, constructing paired sample data sets. The number of sample data sets is consistent with the number of observation time points within the overlapping time period. Based on linear regression theory, the time series values ​​of the first remote sensing dataset are set as independent variables (i.e., explanatory variables), and the resolution aggregated time series values ​​are set as dependent variables (i.e., explained variables). The coefficients of the linear regression equation are solved using the least squares method. The core expression of the linear regression model is: ,in Represents the aggregated time series values ​​at the specified resolution. Represents the time series values ​​of the first remote sensing dataset. The regression slope, The intercept and the two parameters together constitute the core parameters of the model. During model fitting, data preprocessing is required, including outlier removal and data standardization, to ensure the reliability and standardization of the sample data and avoid the impact of outliers on the model fitting accuracy. Through this fitting process, a quantitative correspondence is established between low-spatial-resolution data and high-spatial-resolution aggregated data within each grid cell.

[0099] S14. Extract the regression coefficients from the linear regression model as scaling parameters, and calculate the coefficient of determination of the linear regression model as a quality evaluation parameter.

[0100] Optionally, after completing the linear regression model fitting, the remote sensing data fusion terminal first extracts regression coefficients from the model results, including the regression slope and intercept. These two parameters together constitute the scale transformation parameter. The core function of the scale transformation parameter is to establish a conversion bridge between low-spatial-resolution data and high-spatial-resolution data. Subsequently, this parameter can be used to convert the low-resolution data of the first remote sensing dataset into predicted values ​​at the corresponding high-resolution scale. Its value reflects the quantitative mapping relationship between the two types of data. Then, the coefficient of determination of the linear regression model is calculated as a quality evaluation parameter. The calculation principle of this parameter is based on the degree of deviation between the model's predicted values ​​and the actual observed values ​​(i.e., the resolution aggregated time series values). Its value ranges from 0 to 1. The closer it is to 1, the better the model fit, and the higher the reliability of the model's prediction at the corresponding spatial location; conversely, a lower value indicates a poorer model fit and insufficient prediction reliability. In addition to the coefficient of determination, fitting error indicators and significance test indicators can also be calculated as supplementary quality evaluation parameters to comprehensively and multidimensionally characterize the reliability of the model at various spatial locations, providing a scientific basis for determining the subsequent fusion weights.

[0101] S15. Collect the scale transformation parameters and quality evaluation parameters corresponding to all grid cells of the first spatial resolution to generate a parameter set containing the first spatial resolution.

[0102] Optionally, the remote sensing data fusion terminal first establishes a spatial indexing system for the parameter set. This indexing system is based on the grid cells of the first remote sensing dataset, assigning a unique spatial identifier to each grid cell to ensure accurate correspondence between parameters and spatial locations. Then, according to this spatial index, the scale transformation parameters and quality evaluation parameters corresponding to each first spatial resolution grid cell are extracted one by one, and these parameters are associated and stored with the corresponding grid cell spatial identifiers. During the parameter set process, different types of parameters are standardized to a standard geographic data format to ensure that the data format and storage structure of all parameters remain consistent, facilitating subsequent data retrieval and processing. Simultaneously, a parameter quality check mechanism is established to verify the completeness and rationality of the aggregated parameters, removing grid cell data with missing parameters or obviously unreasonable parameter values ​​to ensure the completeness and reliability of the parameter set. Finally, all verified parameters are organized and arranged according to the spatial grid structure of the first remote sensing dataset to generate a parameter set with the first spatial resolution. This parameter set fully preserves the scale transformation capability and corresponding reliability evaluation information for each spatial location.

[0103] In an optional embodiment, spatial adaptive alignment is performed on the parameter set and background field data to make the spatial coordinate grid of the parameter set consistent with the spatial coordinate grid of the background field data, resulting in aligned parameter set and aligned background field data, including the following steps:

[0104] S21. Obtain background field data containing the second spatial resolution, and determine the spatial coordinate axis direction of the grid metadata of the parameter set and the grid metadata of the background field data to obtain the direction determination result.

[0105] Optionally, the background field data acquired by the remote sensing data fusion terminal must meet the second spatial resolution requirement. Its data sources include historically validated remote sensing products, authoritative geospatial databases, and reanalysis datasets generated from multi-source data fusion. The data must possess good spatial continuity and temporal stability. After acquiring the background field data, its grid metadata is first extracted, including key information such as coordinate system, grid origin coordinates, pixel size, and spatial coordinate axis orientation. Simultaneously, the grid metadata of the parameter set is extracted, using the same metadata parsing standard to ensure consistent extraction dimensions for both types of data metadata. The consistency of different data coordinate systems and related attributes is verified, focusing on a comparative analysis of the spatial coordinate axis orientations of both. This determines whether the orientation definitions of the parameter set's main coordinate axes, such as the x-axis and y-axis, are completely consistent with the background field data. Finally, a clear orientation judgment result is output, providing a basis for subsequent coordinate correction. The coordinate system includes the geographic coordinate system and the projected coordinate system.

[0106] S22. When the direction judgment result is that the parameter set is inconsistent with the background field data in at least one coordinate axis direction, perform a matrix flip operation on the parameter set along the corresponding coordinate axis to make the coordinate axis direction of the parameter set consistent with the coordinate axis direction of the background field data, and obtain the parameter set after direction correction.

[0107] Optionally, after obtaining the orientation determination result, if the remote sensing data fusion terminal determines that the parameter set is inconsistent with the background field data in one or more coordinate axis directions, it performs a matrix flipping operation based on matrix transformation theory. First, the specific directions of the inconsistent coordinate axes are identified, such as opposite x-axis directions or opposite y-axis directions. Then, a corresponding flipping strategy is formulated for that coordinate axis. The flipping operation uses a matrix row and column transformation algorithm. For a two-dimensional parameter matrix, if the x-axis direction is inconsistent, the columns of the matrix are reversed; if the y-axis direction is inconsistent, the rows of the matrix are reversed. This algorithm achieves the flipping correction of the parameter set's coordinate axis orientation. During the flipping process, the correspondence between the scale transformation parameters and quality evaluation parameters of each grid cell in the parameter set must remain unchanged to ensure that the spatial attributes of the parameters are not destroyed. After the flipping is completed, the grid metadata of the corrected parameter set is re-extracted, and the consistency with the coordinate axis orientation of the background field data is verified again to confirm that the coordinate axis orientations of the two are completely consistent. Finally, the orientation-corrected parameter set is obtained, which is unified with the background field data in terms of spatial coordinate axis orientation.

[0108] S23. When the direction judgment result is that the parameter set is consistent with the background field data in the coordinate axis direction, perform spatial interpolation mapping on the parameter set and the parameter set after direction correction, and resample the parameter set and the parameter set after direction correction from the first spatial resolution to the second spatial resolution to obtain the parameter set after interpolation mapping.

[0109] Optionally, after confirming that the coordinate axes of the parameter set and the background field data are consistent, the remote sensing data fusion terminal performs spatial interpolation mapping based on the principle of spatial scale transformation. High-precision spatial interpolation methods are selected, such as bilinear interpolation, which determines the interpolation result by calculating the weighted average of the four neighboring pixels around the target pixel, achieving a balance between computational efficiency and interpolation accuracy; or spline interpolation, which ensures the spatial continuity of the interpolation result by constructing a smooth spline function, suitable for scenarios with high requirements for data smoothness. For both the parameter set after orientation correction and the original parameter set without orientation correction, the same interpolation parameter settings are used, including the interpolation window size and weight calculation rules, to ensure consistency in the interpolation process. The core objective of interpolation mapping is to resample the parameter set at the first spatial resolution to the second spatial resolution, so that the pixel size and grid number of the parameter set perfectly match the background field data. During the interpolation process, the correspondence between the scale transformation parameters and the quality evaluation parameters must be maintained; that is, the scale transformation parameters and quality evaluation parameters corresponding to each newly generated high-resolution pixel are calculated based on the parameters of its neighboring low-resolution grid cells using an interpolation algorithm. The final interpolated parameter set is consistent with the background field data in both spatial resolution and coordinate axis direction, thus meeting the spatial matching requirements of subsequent fusion processing.

[0110] S24. Output the orientation-corrected parameter set, the interpolated parameter set, and the background field data as aligned parameter sets and aligned background field data with the same spatial coordinate grid.

[0111] Optionally, the remote sensing data fusion terminal first performs spatial grid consistency verification on the parameter set after orientation correction (if corrected) and the parameter set after interpolation mapping. The verification includes core grid attributes such as coordinate system, grid origin, pixel size, and number of rows and columns, ensuring complete consistency of the spatial coordinate grids of the two parameter sets. Then, the verified parameter set is spatially matched with the background field data. Based on the principle of geospatial coincidence, the grids of different spatial data are verified to be completely overlapping, confirming that each pixel in the parameter set corresponds one-to-one with the pixel in the background field data in geospatial location, with no grid misalignment or incomplete coverage. If local grid mismatches exist, the parameter set is fine-tuned to ensure complete overlap. Finally, using a standardized geographic data storage format (such as GeoTIFF or NetCDF), the aligned parameter set and the aligned background field data are output and stored. During storage, complete metadata information, including resolution, coordinate system, data source, and processing time, is retained for subsequent data traceability and retrieval. The output aligned data has a completely consistent spatial coordinate grid, eliminating the potential for fusion errors caused by spatial grid differences.

[0112] In an optional embodiment, for a target spatiotemporal unit, a fusion weight is determined based on quality evaluation parameters in the aligned parameter set, and the aligned parameter set and the aligned background field data are weighted and fused based on the fusion weight to generate preliminary fused data containing a second spatial resolution, including the following steps:

[0113] S31. For the target pixels and their corresponding target time points on the second spatial resolution grid, extract the scale transformation parameters and quality evaluation parameters corresponding to the target pixel positions from the aligned parameter set.

[0114] Optionally, the remote sensing data fusion terminal first clarifies the spatial division rules of the second spatial resolution grid, determining the precise geographic coordinate range and spatial index location of each target pixel. A target pixel refers to a high-resolution spatial unit from which fused data is to be generated. Simultaneously, the target time point is determined, which must be within the temporal coverage of the first remote sensing dataset, and preferably a time point with observation data in the second remote sensing dataset, to ensure the reliability of the fused data. Based on the spatial index, the spatial data location is quickly located. According to the geographic coordinates of the target pixel, the corresponding spatial location is quickly retrieved from the aligned parameter set, and the scale transformation parameters and quality evaluation parameters for that location are extracted. A precise positioning algorithm is used during the extraction process to ensure that the extracted parameters completely match the spatial location of the target pixel, avoiding parameter extraction misalignment. If the target pixel location is in the edge region of the parameter set, a boundary processing algorithm is used to ensure the integrity of parameter extraction, avoiding parameter loss due to edge effects. Finally, the scale transformation parameters and quality evaluation parameters of the target pixel at the corresponding target time point are obtained, providing core data for subsequent initial prediction value calculation and fusion weight determination.

[0115] S32. Obtain the first remote sensing data value of the low-resolution pixels covering the area where the target pixel is located at the target time point.

[0116] Optionally, the remote sensing data fusion terminal first clarifies the specific time information of the target time point, ensuring consistency in the time dimension across different datasets based on time matching principles. It then selects the low-resolution data layer corresponding to the target time point from the first remote sensing dataset. Subsequently, based on the geographic coordinates of the target pixel, it determines the range of low-resolution pixels it falls within, i.e., determining which low-resolution grid cell in the first remote sensing dataset the high-resolution target pixel falls into. This determination process is based on the principle of spatial inclusion relationship analysis to determine whether a spatial object is contained within another spatial object. After identifying the corresponding low-resolution pixel, the first remote sensing data value of that pixel at the target time point is extracted. This data value must undergo validity verification to ensure it is not invalid or outlier. If the first remote sensing data value at the target time point is missing, a time interpolation algorithm is used, employing the first remote sensing data values ​​from adjacent valid time points before and after the target time point for interpolation estimation to ensure data integrity. Finally, valid first remote sensing data values ​​of low-resolution pixels covering the area where the target pixel is located are obtained, providing input data for subsequent calculation of high-resolution predicted values ​​using scale transformation parameters.

[0117] S33. Using the scale transformation parameters and the first remote sensing data value, calculate the initial prediction value of the spatiotemporal scale transformation model at the target pixel.

[0118] Optionally, the remote sensing data fusion terminal, based on the previously constructed linear regression model, substitutes the extracted scale transformation parameters and the acquired first remote sensing data values ​​into the model expression. The calculation process follows numerical calculation standards to ensure computational accuracy and avoid prediction deviations caused by calculation errors. The core significance of this initial prediction value is that, through the scale transformation model, it converts the low-resolution first remote sensing data values ​​into prediction results at a high-resolution scale, achieving scale upscaling from low spatial resolution to high spatial resolution. During the calculation process, a strict correspondence between parameters and data values ​​must be established to ensure that the scale transformation parameters of each target pixel accurately match its corresponding first remote sensing data value, avoiding prediction errors caused by parameter misuse. For the initial prediction value calculation of batch target pixels, parallel computing is used to process multiple calculation tasks simultaneously to improve computational efficiency, ensuring efficient completion of prediction value calculation when processing large-scale data. Finally, the initial prediction value of each target pixel at the corresponding target time point is obtained, which is one of the core components of the fused data.

[0119] S34. Obtain the aligned background field data values ​​at the target time point and target pixel position.

[0120] Optionally, the remote sensing data fusion terminal performs data retrieval in the aligned background field data based on the geographic coordinates of the target pixel and the target time point. First, it quickly locates the spatial position of the target pixel in the background field data using a spatial index, ensuring spatial accuracy and avoiding data extraction errors due to spatial location deviations. Then, based on the target time point, it extracts the background field data value corresponding to that spatial position. If the background field data is time-series data, the corresponding value for the target time point is directly extracted; if the background field data is static data, a fixed background value for that spatial position is extracted. During the data extraction process, the extracted data values ​​are validated, including data range and data format verification, to ensure that the extracted background field data values ​​are valid and usable. If the background field data value at the target pixel location is missing, a spatial interpolation method, such as Kriging interpolation, is used to supplement the estimation using the background field data values ​​of neighboring pixels, ensuring data integrity. Finally, the valid background field data value at the target time point and target pixel location is obtained. This value serves as a reference benchmark for the fusion process and is used for weighted fusion with the initial predicted value.

[0121] S35. Based on the values ​​of the quality evaluation parameters, the fusion weight is calculated using a preset weight mapping function.

[0122] Optionally, the remote sensing data fusion terminal first analyzes the numerical distribution characteristics of the quality assessment parameters, including statistical information such as the parameter's value range, mean, median, and distribution density. Based on statistical distribution theory, it analyzes the data distribution patterns and dynamically determines the first and second thresholds. The first threshold is the low reliability critical value for the quality assessment parameters; values ​​below this value indicate extremely low model prediction reliability. The second threshold is the high reliability critical value; values ​​above this value indicate extremely high model prediction reliability. The determination of these two thresholds must ensure a reasonable classification of the parameter's reliability level. The preset weight mapping function adopts a piecewise function form. When the quality assessment parameter is below the first threshold, the function outputs a fusion weight of 0; when the parameter is above the second threshold, the function outputs a fusion weight of 1; when the parameter is between the two thresholds, the function uses a linear interpolation formula to calculate the fusion weight, ensuring that the weight changes continuously and linearly with the parameter value. During the calculation process, the preset function logic must be strictly followed to avoid weight calculation deviations caused by human intervention. For the calculation of fusion weights for a batch of target pixels, a batch processing algorithm is adopted. After uniformly determining the threshold based on the statistical characteristics of the quality evaluation parameters, the fusion weight of each pixel is calculated in batches to ensure the consistency and efficiency of weight calculation; finally, the fusion weight corresponding to each target pixel is obtained.

[0123] S36. Use fusion weights to linearly weight the initial predicted value and the background field data value, calculate the preliminary fusion value of the target pixel at the target time point, and obtain the preliminary fusion data.

[0124] Optionally, the remote sensing data fusion terminal, based on weighted fusion theory, optimizes the combination of multi-source data through weight allocation. It employs a linear weighted formula to fuse the initial predicted values ​​and background field data values. The contribution ratio of the initial predicted values ​​and background field data values ​​in the final fusion result is dynamically adjusted according to the fusion weight: when the fusion weight is close to 1, the contribution of the initial predicted values ​​dominates, reflecting the high reliability of the model's predictions; when the fusion weight is close to 0, the contribution of the background field data values ​​dominates, avoiding fusion errors caused by low model reliability; when the fusion weight is between these two values, a smooth transition is achieved, ensuring the spatial continuity of the fusion result. During the calculation process, the accuracy of numerical calculations must be maintained to avoid deviations caused by decimal operations. For all target spatiotemporal units, calculations are performed according to a unified weighted formula to ensure consistency in the fusion process. After calculation, the preliminary fusion values ​​of all target pixels at the corresponding target time points are organized according to a grid structure with a second spatial resolution to form preliminary fusion data. This data already possesses high spatial resolution characteristics, but further temporal optimization processing is still required.

[0125] In an optional embodiment, the fusion weight is calculated based on the values ​​of the quality evaluation parameters using a preset weight mapping function, including the following steps:

[0126] S41. Based on the numerical distribution characteristics of the quality evaluation parameters, dynamically determine the first threshold and the second threshold.

[0127] The first threshold is less than the second threshold.

[0128] Optionally, the remote sensing data fusion terminal first performs statistical analysis on the quality assessment parameters for all spatial locations, extracting key distribution features of the parameters using statistical methods that describe data distribution characteristics, including indicators such as minimum, maximum, mean, median, quartiles, and standard deviation. Based on these distribution features, and combined with data quality indicators, reliability levels are classified, and a first and second threshold are dynamically determined. The core logic of the determination process is: the first threshold should be set at the upper limit of the concentrated area of ​​low reliability parameter distribution, ensuring that the model prediction reliability corresponding to parameters below this threshold is significantly insufficient; the second threshold should be set at the lower limit of the concentrated area of ​​high reliability parameter distribution, ensuring that the model prediction reliability corresponding to parameters above this threshold is sufficiently reliable. Simultaneously, the principle of the first threshold being less than the second threshold is strictly followed, forming an intermediate reliability region between the two thresholds. The threshold determination process does not rely on fixed empirical values; it is entirely based on adaptive adjustments according to the distribution characteristics of the quality assessment parameters of the currently processed data, ensuring the rationality and relevance of the threshold classification. For example, when the overall distribution of parameters is relatively concentrated and the overall reliability is high, the first threshold can be appropriately increased and the second threshold can be decreased to expand the range of high reliability region; when the distribution of parameters is relatively dispersed, the threshold spacing can be reasonably adjusted to ensure that the number of parameters in each reliability region is evenly distributed; finally, the first threshold and the second threshold that are adapted to the current data characteristics are obtained.

[0129] S42. When the quality evaluation parameter value is less than the first threshold, the fusion weight is 0.

[0130] Optionally, after dynamically determining the first and second thresholds, the remote sensing data fusion terminal compares the quality evaluation parameter value of each target spatiotemporal unit with the first threshold. When the judgment result is that the quality evaluation parameter value is less than the first threshold, the fusion weight of the target spatiotemporal unit is set to 0 according to the preset weight mapping rule. The core theoretical basis for this setting is that a quality evaluation parameter value lower than the first threshold indicates that the goodness of fit of the spatiotemporal scale transformation model at that spatial location is extremely low, and the reliability of the prediction result cannot be guaranteed. If such prediction values ​​are included in the fusion process, the quality of the final fused data will be significantly reduced. Therefore, by setting the fusion weight to 0, the initial fusion result is completely dominated by the background field data value. The background field data has good spatial continuity and stability, which can effectively avoid the fusion error caused by the low reliability of the model. In the weight setting process, a batch judgment and assignment algorithm is used to uniformly assign values ​​to all target spatiotemporal units with quality evaluation parameter values ​​less than the first threshold to ensure processing efficiency and consistency; at the same time, the spatial location information of such target spatiotemporal units is recorded to ensure the traceability of the fusion process.

[0131] S43. When the quality evaluation parameter value is greater than or equal to the first threshold and less than or equal to the second threshold, the fusion weight is calculated using a linear interpolation function; wherein, the expression of the linear interpolation function is:

[0132]

[0133] In the formula, To integrate weights, These are quality evaluation parameter values. The first threshold, This is the second threshold.

[0134] Optionally, the remote sensing data fusion terminal calculates the fusion weights for target spatiotemporal units whose quality assessment parameter values ​​fall between a first threshold and a second threshold using linear interpolation. Quality assessment parameter values ​​within this range indicate a moderate level of model prediction reliability; they should neither be entirely relied upon nor completely discarded. Therefore, a linear interpolation function is used to achieve continuous variation in the fusion weights. When the parameter value is close to the first threshold, the fusion weight is close to 0, and the fusion result is dominated by background field data; when the parameter value is close to the second threshold, the fusion weight is close to 1, and the fusion result is dominated by model predictions; when the parameter value is in the middle of the range, the fusion weight takes the intermediate value, achieving a balanced fusion of the two. During the calculation process, it is necessary to ensure that the numerical precision of the quality assessment parameter values, the first threshold, and the second threshold is consistent to avoid weight calculation errors caused by differences in data precision. Parallel vector processing is used for batch data of target spatiotemporal units to improve computational efficiency and ensure rapid weight calculation. The fusion weights obtained through this linear interpolation function enable a smooth transition between model predictions and background field data, effectively avoiding spatial discontinuities caused by rigid threshold divisions.

[0135] S44. When the quality evaluation parameter value is greater than the second threshold, the fusion weight is 1.

[0136] Optionally, the remote sensing data fusion terminal compares the quality assessment parameter value of the target spatiotemporal unit with a second threshold. When the parameter value is greater than the second threshold, the fusion weight is set to 1 according to the weight mapping rule. The core basis for this setting is that a quality assessment parameter value higher than the second threshold means that the spatiotemporal scale transformation model has an excellent fitting effect at that spatial location, and the prediction results have extremely high reliability and accuracy, accurately reflecting the actual characteristics of the surface. Setting the fusion weight to 1 at this time makes the initial fusion result completely dominated by the model's initial prediction value, giving full play to the prediction advantages of the high-reliability model and ensuring that the fused data can accurately capture high spatial resolution surface details. During the weight assignment process, a fast retrieval assignment algorithm is used to quickly locate all target spatiotemporal units with parameter values ​​greater than the second threshold through spatial indexing, and uniformly assign a fusion weight of 1 to ensure the efficiency and consistency of the processing. At the same time, the distribution characteristics of this type of target spatiotemporal unit are statistically recorded, and the spatial distribution pattern of the high-reliability area of ​​the model is analyzed to provide a reference for subsequent model optimization and data application, further improving the practicality and relevance of the fusion.

[0137] In an optional embodiment, the method further includes the following steps:

[0138] S51. When processing remote sensing data over a large area, the parameter set and background field data are spatially divided into multiple sub-region blocks.

[0139] Optionally, when dealing with large-area remote sensing data processing tasks, to address the issues of excessive memory consumption and low computational efficiency, the remote sensing data fusion terminal divides the large-area spatial data into smaller blocks for parallel processing based on spatial block processing. This involves spatially dividing the parameter set and background field data. First, based on the hardware performance of the data processing terminal (e.g., memory capacity, processor processing power) and data characteristics (e.g., data size, spatial resolution), predefined block rules are established, including parameters such as the spatial size, shape, and overlap of the sub-region blocks. The block process employs a spatial grid partitioning algorithm, uniformly dividing the large-area parameter set and background field data into multiple independent sub-region blocks according to the predefined block rules, ensuring that the data volume of each sub-region block is within the hardware processing capacity. To avoid data discontinuity at the edges of sub-region blocks, a certain overlap area is set between adjacent sub-region blocks. The width of the overlap area is determined based on the data resolution and fusion accuracy requirements. During the block process, the spatial correspondence between the parameter set and background field data within each sub-region block remains unchanged, ensuring that the block-processed data still accurately reflects the spatial characteristics of the original data. At the same time, each sub-region block is assigned a unique identifier, which records its spatial location coordinates in the original data.

[0140] S52. Based on the parameter set sub-data and background field data sub-data corresponding to multiple sub-region blocks, perform spatial resampling mapping with coordinate direction consistency judgment, coordinate flip correction and bilinear interpolation to obtain the parameter subset after alignment of each sub-region block.

[0141] Optionally, the remote sensing data fusion terminal performs spatial alignment operations on the parameter set sub-data and background field data sub-data of each sub-region block, following the complete data processing flow. First, based on the principle of spatial coordinate direction judgment, the consistency of the coordinate axis directions of the parameter set sub-data and background field data sub-data is determined to clarify whether there are directional differences. If inconsistencies exist, a matrix flipping algorithm is used to flip and correct the parameter set sub-data along the corresponding coordinate axes, ensuring complete consistency in coordinate axis directions between sub-data. Subsequently, a bilinear interpolation method is used for spatial resampling mapping. The core principle of this method is based on the parameter set sub-data within the sub-region block, calculating the weighted average of the four adjacent pixels surrounding the target pixel to achieve scale transformation from the first spatial resolution to the second spatial resolution, ensuring that the resampled sub-data has high spatial accuracy and continuity. During processing, the processing parameters (such as interpolation window size and weight calculation rules) of each sub-region block are kept consistent to avoid data inconsistencies between sub-region blocks due to differences in processing parameters. For sub-region blocks containing overlapping areas, the processing of the overlapping parts follows the same rules to ensure consistent data processing results for overlapping areas. After processing, the quality of the parameter subset after alignment of each sub-region block is checked to ensure that the spatial resolution and coordinate axis direction of the parameter subset are completely matched with the corresponding background field data sub-data, laying the foundation for subsequent stitching operations.

[0142] S53. After all sub-region blocks have been processed, the parameter subsets after the sub-region blocks are aligned are spliced ​​together according to the spatial correspondence to generate the aligned parameter set.

[0143] Optionally, after completing the spatial alignment of all sub-region blocks, the remote sensing data fusion terminal performs parameter subset stitching based on spatial location matching to integrate data from different sub-regions according to their spatial location. First, a spatial index system for each sub-region block is established based on its unique identifier and corresponding spatial coordinates, clarifying the specific location of each parameter subset within the original large-scale data. The stitching process employs seamless spatial data stitching to achieve gapless, non-overlapping, and redundant stitching of sub-region data, stitching each parameter subset sequentially according to its spatial location. For overlapping areas between sub-region blocks, a data consistency verification mechanism is used to compare the parameter values ​​of the overlapping parts. If there are minor differences, a weighted average method is used for smoothing to ensure the continuity of the stitched data; if the differences are significant, the processing process for that area is reviewed back, and the processing error is corrected before stitching. After stitching, a global spatial consistency verification is performed on the generated complete parameter set, including verification of core attributes such as grid structure, coordinate system, and resolution, ensuring that the stitched parameter set is spatially complete and continuous, without misalignment, missing, or redundant data. Finally, an aligned parameter set covering the entire large area is generated. This parameter set solves the memory bottleneck problem of large-scale data processing while maintaining the spatial integrity and consistency of the data.

[0144] In an optional embodiment, after generating preliminary fused data for multiple target spatiotemporal units, the preliminary fused data corresponding to the multiple target spatiotemporal units is smoothed and denoised according to the time series to obtain fused remote sensing data, including the following steps:

[0145] S61. Obtain preliminary fusion data of multiple target spatiotemporal units to form a preliminary fusion time series.

[0146] Optionally, the remote sensing data fusion terminal collects preliminary fusion data from all processed target spatiotemporal units. This data covers the fusion results at different spatial locations (second spatial resolution pixels) within the study area at different time points. Subsequently, based on time series construction, the discrete spatiotemporal data is organized chronologically, and the preliminary fusion data is processed along the time dimension. First, the timestamp information of each target spatiotemporal unit is extracted to clarify its corresponding observation time point. Then, the preliminary fusion data at the same spatial location are sorted according to chronological order to form a preliminary fusion time series for a single spatial location. For the entire study area, the time series data from all spatial locations are integrated to construct a global preliminary fusion time series dataset. During the construction process, a time consistency verification mechanism is adopted to ensure that the time series of all spatial locations cover the same time range and that the time intervals are consistent, avoiding subsequent processing errors caused by inconsistencies in the time dimension. Simultaneously, the time series data undergoes a completeness check, and for missing time point data, time interpolation methods are used to supplement it, ensuring the continuity and completeness of the time series data for each spatial location. The final preliminary fusion time series fully preserves the spatiotemporal dynamic changes of surface features within the study area.

[0147] S62. Perform convolution smoothing on the preliminary fused time series to obtain the smoothed time series.

[0148] Optionally, the remote sensing data fusion terminal performs filtering and smoothing on the time series signals based on signal processing, and uses convolutional smoothing to process the initially fused time series. First, a suitable convolution kernel is selected based on the characteristics of the time series (such as time resolution and noise distribution). Commonly used convolution kernels include rectangular convolution kernels and Gaussian convolution kernels. The core process of convolution operation is to perform sliding window convolution with the time series data using the convolution kernel. By calculating the weighted average of the data within the window, the original data value at the center of the window is replaced, thereby eliminating high-frequency noise in the time series. During processing, the size of the convolution kernel must be set appropriately. Too small a kernel size may result in insignificant smoothing, while too large a kernel size may cause loss of detailed information in the time series. For time series containing outliers, convolutional smoothing can effectively suppress the influence of outliers, making the time series more reflective of the true changing trends of surface features. After processing, the smoothed time series is validated. By comparing the time series curves before and after smoothing, it is ensured that the smoothing process eliminates noise interference while fully preserving the overall trend and key changing features of the time series, resulting in a smoothed time series.

[0149] S63. Perform outlier detection and correction on the smoothed time series to obtain the time series with outliers removed.

[0150] Optionally, the remote sensing data fusion terminal uses outlier detection algorithms to process the smoothed time series. Common detection methods include the 3σ criterion and identifying outliers that deviate from the trend by analyzing the time series' changing trends. The core logic of the detection process is to establish a normal range or trend model for the time series and determine values ​​that exceed the normal range or deviate from the trend as outliers. For detected outliers, a rationality analysis needs to be conducted in conjunction with the physical meaning of the data and actual observation conditions to confirm whether they are genuine outliers or outliers caused by data processing errors. If they are outliers caused by data processing errors, outlier correction algorithms are used for processing. Common correction methods include interpolation correction and trend fitting correction. If they are genuine outliers, their original characteristics need to be preserved to avoid over-correction that could lead to data distortion. After correction, the time series is re-detected for outliers to ensure that all unreasonable outliers have been corrected, ultimately resulting in a time series with outliers removed, which has higher temporal continuity and reliability.

[0151] S64. Arrange the time series data after anomaly removal according to the original spatial grid structure and time order to generate fused remote sensing data.

[0152] Optionally, after obtaining the time series data with anomaly removal, the remote sensing data fusion terminal organizes the spatiotemporal data in an ordered manner according to spatial and temporal dimensions, and performs final data processing and arrangement. First, based on the original second spatial resolution grid structure, the anomaly removal time series data corresponding to each spatial location is located according to its spatial index position in the grid. Then, in chronological order, all spatial location data corresponding to each time point are integrated to form a high-resolution spatial data layer for each time point. During the arrangement process, it is ensured that each data value accurately corresponds to its spatial location and time point, without data misalignment or disordered chronological order. Simultaneously, standardized geographic data formats (such as GeoTIFF, NetCDF, etc.) are used to store the data, preserving complete metadata information during storage, including the data's spatial resolution, coordinate system, temporal coverage, data source, processing flow, and quality assessment results, facilitating subsequent data sharing, application, and traceability. The final fused remote sensing data possesses high spatial resolution, long-term series continuity, and high data reliability, meeting the needs of various application scenarios such as geoscience research, resource and environmental monitoring, and ecological assessment, providing high-quality data support for analysis and decision-making in related fields.

[0153] In the aforementioned multi-scale spatiotemporal fusion method for remote sensing data, spatial adaptive alignment processing is used to eliminate parameter spatial matching errors caused by differences in different remote sensing data grid systems, ensuring accurate matching between parameters and the target high-resolution grid and avoiding invalid or erroneous fusion results. A continuous weight fusion strategy based on quality evaluation parameters replaces the traditional hard threshold judgment, achieving a smooth spatial transition of the fusion results, effectively solving the problems of spatial discontinuity and block artifacts in the fusion results, and ensuring spatial consistency. By decoupling the core fusion steps from temporal post-processing and introducing a block processing mechanism, the coupling degree of the computational process is reduced, overcoming the memory bottleneck of large-scale data processing and improving engineering adaptability. Simultaneously, the dynamic threshold adjustment strategy allows the method to adapt to different data statistical characteristics, enhancing regional adaptability and robustness, and possessing good versatility. It is applicable to the fusion of various remote sensing parameters, does not depend on specific sensors or data products, and comprehensively optimizes the quality and practicality of remote sensing data fusion.

[0154] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0155] Based on the same inventive concept, this application also provides a remote sensing data multi-scale spatiotemporal fusion system for implementing the aforementioned remote sensing data multi-scale spatiotemporal fusion method. The solution provided by this system is similar to the implementation scheme described in the above method. Therefore, the specific limitations of one or more embodiments of the remote sensing data multi-scale spatiotemporal fusion system provided below can be found in the limitations of the remote sensing data multi-scale spatiotemporal fusion method described above, and will not be repeated here.

[0156] In one exemplary embodiment, such as Figure 3 As shown, a schematic diagram of the structure of a multi-scale spatiotemporal fusion system 10 for remote sensing data is provided, including:

[0157] The multidimensional data acquisition module 11 is used to acquire a first remote sensing dataset with a first spatial resolution, a second remote sensing dataset with a second spatial resolution, and background field data with a second spatial resolution.

[0158] The parameter modeling module 12 is used to construct a spatiotemporal scale transformation model based on the aggregated data of the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, and to obtain a parameter set containing the first spatial resolution; wherein, the parameter set includes scale transformation parameters and quality evaluation parameters, and the quality evaluation parameters are used to characterize the reliability of the spatiotemporal scale transformation model at the corresponding spatial location;

[0159] The spatial adaptive alignment module 13 is used to perform spatial adaptive alignment of the parameter set and the background field data, so that the spatial coordinate grid of the parameter set is consistent with the spatial coordinate grid of the background field data, and thus obtain the aligned parameter set and the aligned background field data.

[0160] The adaptive weighted fusion module 14 is used to determine the fusion weight based on the quality evaluation parameters in the aligned parameter set for the target spatiotemporal unit, and to perform weighted fusion on the aligned parameter set and the aligned background field data based on the fusion weight to generate preliminary fusion data containing the second spatial resolution.

[0161] The fused remote sensing data generation module 15 is used to smooth and filter the preliminary fused data corresponding to multiple target spatiotemporal units according to the time series after completing the preliminary fused data generation of multiple target spatiotemporal units, so as to obtain fused remote sensing data.

[0162] Furthermore, the parametric modeling module 12 is also used for:

[0163] S11. Extract the time series values ​​of the first remote sensing dataset within the overlapping time period on each grid cell of the first spatial resolution.

[0164] S12. Based on the geographical range of the first spatial resolution grid cell, spatially average all high-resolution pixel values ​​of the second remote sensing dataset that fall within the geographical range to generate a resolution aggregated time series corresponding to the time points of the first remote sensing dataset.

[0165] S13. For each first spatial resolution grid cell, the time series value of the first remote sensing dataset is used as the independent variable, and the corresponding resolution aggregated time series value is used as the dependent variable for fitting to obtain a linear regression model.

[0166] S14. Extract regression coefficients from the linear regression model as scaling parameters, and calculate the coefficient of determination of the linear regression model as a quality evaluation parameter.

[0167] S15. Collect the scale transformation parameters and quality evaluation parameters corresponding to all grid cells of the first spatial resolution to generate a parameter set containing the first spatial resolution.

[0168] Furthermore, the spatial adaptive alignment module 13 is also used for:

[0169] S21. Obtain background field data containing the second spatial resolution, and determine the spatial coordinate axis direction of the grid metadata of the parameter set and the grid metadata of the background field data to obtain the direction determination result.

[0170] S22. When the direction judgment result is that the parameter set is inconsistent with the background field data in at least one coordinate axis direction, perform a matrix flip operation on the parameter set along the corresponding coordinate axis to make the coordinate axis direction of the parameter set consistent with the coordinate axis direction of the background field data, and obtain the parameter set after direction correction.

[0171] S23. When the direction judgment result is that the parameter set is consistent with the background field data in the coordinate axis direction, perform spatial interpolation mapping on the parameter set and the parameter set after direction correction, and resample the parameter set and the parameter set after direction correction from the first spatial resolution to the second spatial resolution to obtain the parameter set after interpolation mapping.

[0172] S24. Output the orientation-corrected parameter set, the interpolated parameter set, and the background field data as aligned parameter sets and aligned background field data with the same spatial coordinate grid.

[0173] Furthermore, the adaptive weighted fusion module 14 is also used for:

[0174] S31. For the target pixels and corresponding target time points on the second spatial resolution grid, extract the scale transformation parameters and quality evaluation parameters corresponding to the target pixel positions from the aligned parameter set.

[0175] S32. Obtain the first remote sensing data value of the low-resolution pixel covering the area where the target pixel is located at the target time point;

[0176] S33. Using the scale transformation parameters and the first remote sensing data value, calculate the initial prediction value of the spatiotemporal scale transformation model at the target pixel;

[0177] S34. Obtain the aligned background field data values ​​at the target time point and target pixel location;

[0178] S35. Based on the values ​​of the quality evaluation parameters, the fusion weight is calculated through a preset weight mapping function;

[0179] S36. Use fusion weights to linearly weight the initial predicted value and the background field data value, calculate the preliminary fusion value of the target pixel at the target time point, and obtain the preliminary fusion data.

[0180] Furthermore, the adaptive weighted fusion module 14 is also used for:

[0181] S41. Based on the numerical distribution characteristics of the quality evaluation parameters, dynamically determine the first threshold and the second threshold; wherein the first threshold is less than the second threshold;

[0182] S42. When the quality evaluation parameter value is less than the first threshold, the fusion weight is 0.

[0183] S43. When the quality evaluation parameter value is greater than or equal to the first threshold and less than or equal to the second threshold, the fusion weight is calculated using a linear interpolation function; wherein, the expression of the linear interpolation function is:

[0184]

[0185] In the formula, To integrate weights, These are quality evaluation parameter values. The first threshold, The second threshold;

[0186] S44. When the quality evaluation parameter value is greater than the second threshold, the fusion weight is 1.

[0187] Furthermore, the spatial adaptive alignment module 13 is also used for:

[0188] S51. When processing remote sensing data over a large area, the parameter set and background field data are spatially divided into multiple sub-region blocks.

[0189] S52. Based on the parameter set subsets and background field data subsets corresponding to multiple sub-region blocks, perform coordinate direction consistency judgment, coordinate flip correction, and bilinear interpolation spatial resampling mapping to obtain the parameter subsets after alignment of each sub-region block.

[0190] S53. After all sub-region blocks have been processed, the parameter subsets after the sub-region blocks are aligned are spliced ​​together according to the spatial correspondence to generate the aligned parameter set.

[0191] Furthermore, the integrated remote sensing data generation module 15 is also used for:

[0192] S61. Obtain preliminary fusion data of multiple target spatiotemporal units to form a preliminary fusion time series;

[0193] S62. Perform convolution smoothing on the preliminary fused time series to obtain the smoothed time series;

[0194] S63. Perform outlier detection and correction on the smoothed time series to obtain the time series with outliers removed.

[0195] S64. Arrange the time series data after anomaly removal according to the original spatial grid structure and time order to generate fused remote sensing data.

[0196] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to implement the steps of the multi-scale spatiotemporal fusion method for remote sensing data as described above.

[0197] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the above method embodiments.

[0198] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts 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 disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0199] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A multiscale spatio-temporal fusion method for remote sensing data, characterized in that, The method includes: S1. Acquire a first remote sensing dataset with a first spatial resolution, a second remote sensing dataset with a second spatial resolution, and background field data with the second spatial resolution; S2. Based on the aggregated data of the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, a spatiotemporal scale conversion model is constructed to obtain a parameter set containing the first spatial resolution; wherein, the parameter set includes scale conversion parameters and quality evaluation parameters, and the quality evaluation parameters are used to characterize the reliability of the spatiotemporal scale conversion model at the corresponding spatial location; S3. Perform spatial adaptive alignment on the parameter set and the background field data to make the spatial coordinate grid of the parameter set consistent with the spatial coordinate grid of the background field data, so as to obtain the aligned parameter set and the aligned background field data. S4. For the target spatiotemporal unit, determine the fusion weight based on the quality evaluation parameters in the aligned parameter set, and perform weighted fusion on the aligned parameter set and the aligned background field data based on the fusion weight to generate preliminary fused data containing the second spatial resolution. S5. After generating the preliminary fusion data of multiple target spatiotemporal units, the preliminary fusion data corresponding to the multiple target spatiotemporal units are smoothed and filtered for noise reduction according to the time series to obtain fused remote sensing data.

2. The method according to claim 1, characterized in that, The spatiotemporal scale transformation model is constructed based on the aggregated data of the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, resulting in a parameter set that includes the first spatial resolution, including: S11. Extract the time series values ​​of the first remote sensing dataset within the overlapping time period on each grid cell of the first spatial resolution; S12. Based on the geographical range of the first spatial resolution grid cell, spatially average all high-resolution pixel values ​​of the second remote sensing dataset falling within the geographical range to generate a resolution aggregated time series corresponding to the time point of the first remote sensing dataset. S13. For each first spatial resolution grid cell, the time series value of the first remote sensing dataset is used as the independent variable, and the corresponding resolution aggregated time series value is used as the dependent variable for fitting to obtain a linear regression model. S14. Extract regression coefficients from the linear regression model as the scaling parameters, and calculate the determination coefficient of the linear regression model as the quality evaluation parameters; S15. Collect the scale transformation parameters and quality evaluation parameters corresponding to all the first spatial resolution grid cells to generate a parameter set containing the first spatial resolution.

3. The method according to claim 2, characterized in that, The step of spatially adaptively aligning the parameter set and the background field data, so that the spatial coordinate grid of the parameter set is consistent with the spatial coordinate grid of the background field data, to obtain the aligned parameter set and the aligned background field data, includes: S21. Obtain background field data containing the second spatial resolution, and determine the spatial coordinate axis direction of the grid metadata of the parameter set and the grid metadata of the background field data to obtain the direction determination result; S22. When the direction judgment result is that the parameter set is inconsistent with the background field data in at least one coordinate axis direction, the parameter set is subjected to a matrix flip operation along the corresponding coordinate axis to make the coordinate axis direction of the parameter set consistent with the coordinate axis direction of the background field data, so as to obtain the parameter set after direction correction. S23. When the direction determination result is that the parameter set is consistent with the background field data in the coordinate axis direction, spatial interpolation mapping is performed on the parameter set and the direction-corrected parameter set, and the parameter set and the direction-corrected parameter set are resampled from the first spatial resolution to the second spatial resolution to obtain the interpolated parameter set. S24. Output the interpolated parameter set and the background field data as the aligned parameter set and the aligned background field data with the same spatial coordinate grid.

4. The method according to claim 3, characterized in that, For the target spatiotemporal unit, a fusion weight is determined based on the quality evaluation parameters in the aligned parameter set, and the aligned parameter set and the aligned background field data are weighted and fused based on the fusion weight to generate preliminary fused data containing the second spatial resolution, including: S31. For the target pixel and the corresponding target time point on the second spatial resolution grid, extract the scale transformation parameters and quality evaluation parameters corresponding to the target pixel position from the aligned parameter set. S32. Obtain the first remote sensing data value of the low-resolution pixel covering the area where the target pixel is located at the target time point; S33. Using the scale conversion parameters and the first remote sensing data value, calculate the initial prediction value of the spatiotemporal scale conversion model at the target pixel; S34. Obtain the aligned background field data value at the target time point and the target pixel position; S35. Based on the values ​​of the quality evaluation parameters, the fusion weight is calculated using a preset weight mapping function; S36. Using the fusion weights, linearly weight the initial predicted value and the background field data value to calculate the preliminary fusion value of the target pixel at the target time point, and obtain the preliminary fusion data.

5. The method according to claim 4, characterized in that, The calculation of the fusion weight based on the values ​​of the quality evaluation parameters using a preset weight mapping function includes: S41. Based on the numerical distribution characteristics of the quality evaluation parameters, dynamically determine a first threshold and a second threshold; wherein the first threshold is less than the second threshold; S42. When the quality evaluation parameter value is less than the first threshold, the fusion weight is 0; S43. When the quality evaluation parameter value is greater than or equal to the first threshold and the quality evaluation parameter value is less than or equal to the second threshold, the fusion weight is calculated using a linear interpolation function; wherein, the expression of the linear interpolation function is: In the formula, To integrate weights, These are quality evaluation parameter values. The first threshold, The second threshold; S44. When the quality evaluation parameter value is greater than the second threshold, the fusion weight is 1.

6. The method according to claim 3, characterized in that, The method further includes: S51. When processing remote sensing data over a large area, the parameter set and the background field data are spatially divided into multiple sub-region blocks. S52. Based on the parameter set subsets and background field data subsets corresponding to the multiple sub-region blocks, perform coordinate direction consistency judgment, coordinate flip correction and bilinear interpolation spatial resampling mapping to obtain the aligned parameter subsets of each sub-region block. S53. After all sub-region blocks have been processed, the parameter subsets after the sub-region blocks are aligned are spliced ​​together according to the spatial correspondence to generate the aligned parameter set.

7. The method according to claim 3, characterized in that, After generating preliminary fused data for multiple target spatiotemporal units, the preliminary fused data corresponding to the multiple target spatiotemporal units is smoothed and denoised according to the time series to obtain fused remote sensing data, including: S61. Obtain preliminary fusion data of multiple target spatiotemporal units to form a preliminary fusion time series; S62. Perform convolution smoothing on the preliminary fused time series to obtain a smoothed time series; S63. Perform outlier detection and correction on the smoothed time series to obtain an outlier-removed time series; S64. The time series of anomaly removal is arranged according to the original spatial grid structure and time order to generate the fused remote sensing data.

8. A multi-scale spatiotemporal fusion system for remote sensing data, characterized in that, The system includes: The multidimensional data acquisition module is used to acquire a first remote sensing dataset with a first spatial resolution, a second remote sensing dataset with a second spatial resolution, and background field data with the second spatial resolution. The parameter modeling module is used to construct a spatiotemporal scale transformation model based on the aggregated data of the first remote sensing dataset within the overlapping time period of the second remote sensing dataset, and to obtain a parameter set containing the first spatial resolution; wherein, the parameter set includes scale transformation parameters and quality evaluation parameters, and the quality evaluation parameters are used to characterize the reliability of the spatiotemporal scale transformation model at the corresponding spatial location; The spatial adaptive alignment module is used to perform spatial adaptive alignment of the parameter set and the background field data, so that the spatial coordinate grid of the parameter set is consistent with the spatial coordinate grid of the background field data, and thus obtain the aligned parameter set and the aligned background field data. An adaptive weighted fusion module is used to determine the fusion weight based on the quality evaluation parameters in the aligned parameter set for the target spatiotemporal unit, and to perform weighted fusion on the aligned parameter set and the aligned background field data based on the fusion weight to generate preliminary fused data containing the second spatial resolution. The fused remote sensing data generation module is used to perform smoothing filtering and noise reduction on the preliminary fused data corresponding to the multiple target spatiotemporal units according to the time series after completing the preliminary fused data generation of multiple target spatiotemporal units, so as to obtain fused remote sensing data.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

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

Images