Data processing method and system based on remote sensing information cloud platform

Abstract

The invention discloses a data processing method and system based on a remote sensing information cloud platform, and the method comprises the steps: S1, setting a collection rule, collecting the original data flow of a multi-source satellite, and obtaining the primary image data with unified space-time reference through automatic calibration and geometric correction; s2, calling a workflow based on the primary image data, and fusing the workflow with multi-source satellite data to generate a structured basic data set; s3, based on the structured basic data set, performing preloading optimization processing by using a time sequence perception engine and an LSTM model, and outputting time sequence thematic data; and S4, the optimal remote sensing information cloud platform data processing is realized based on the time sequence thematic data processing time-space knowledge graph, association between entities is established, including time sequence association and space association, and the established time-space knowledge graph can intuitively reflect the change rule of ground feature indexes in the time-space dimension and support intelligent transition from data query to decision suggestion.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing data processing, and particularly relates to a data processing method and system based on a remote sensing information cloud platform. BACKGROUND

[0002] With the development of remote sensing technology, more and more satellite sensors are used for earth observation, generating massive multi-source remote sensing data, including optical satellite data, synthetic aperture radar data, etc., which have the characteristics of multi-source and heterogeneity; the data of different sensors have significant differences in spatial resolution, spectral characteristics, etc., resulting in great challenges in data fusion and collaborative analysis.

[0003] The existing remote sensing information cloud platform has different characteristics of data of different sensors in the process of obtaining data, including spatial resolution differences, for example, optical satellite data may have high spatial resolution, which can clearly distinguish surface details, while SAR data may have unique penetration ability in some cases, but the spatial resolution is different, which makes it difficult to analyze multi-source data comprehensively, and the amount of long-time sequence remote sensing data is huge, and the traditional data processing method is low in efficiency when processing these data, therefore, a data processing method and system based on a remote sensing information cloud platform are proposed. SUMMARY

[0004] In order to overcome the above-mentioned defects of the prior art, in order to achieve the above-mentioned purpose, the present application proposes the following technical scheme: The data processing method based on the remote sensing information cloud platform comprises: S1: setting acquisition rules to acquire original data streams of multi-source satellites, and obtaining primary image data with unified time and space reference through automatic calibration and geometric correction; S2: calling a workflow based on the primary image data to fuse multi-source satellite data to generate a structured basic data set; S3: based on the structured basic data set, using a time sequence perception engine and an LSTM model for preloading optimization processing, and outputting time sequence thematic data; S4: based on the time sequence thematic data, processing a time and space knowledge graph to achieve optimal remote sensing information cloud platform data processing.

[0005] The multi-source satellites include optical satellites and SAR satellites; The acquisition rules include that the optical satellite ensures at least 1 cloud-free image per month with a cloud cover of less than 20%, and the cloud pixel is determined by a threshold combination of blue band reflectance greater than 0.3 and near-infrared band reflectance less than 0.2, and the SAR satellite collects 2 images per month, with a time span covering the target period, and the data format is Level-1, and the optical data is saved in GeoTIFF format.

[0006] The automated calibration and geometric correction process is as follows: Automated calibration converts raw digital quantization values ​​into radiance or reflectance. For optical satellites, radiance is calculated, and then the radiance is converted into surface reflectance through reflectance conversion. For SAR satellites, the backscattering coefficient formula is used: Where DN is the digital quantization value. Here, K is the backscattering coefficient, and K is the calibration constant. Geometric correction automatically extracts control points (GCPs) from the raw data stream using the SIFT algorithm. Each image requires at least 20 evenly distributed GCPs with a matching error ≤ 1 pixel. A quadratic polynomial transform is used, and the GCP coordinates are calculated by fitting them using the least squares method. For mountainous terrain, DEM data is introduced for terrain correction. The corrected reflectance formula is as follows: ,in, To correct the reflectivity, α represents the original reflectivity, β represents the slope, and β represents the difference between the slope aspect and the solar azimuth angle.

[0007] The process of invoking the workflow to fuse multi-source satellite data is as follows: The process of calling the workflow to fuse multi-source satellite data is as follows: Gram-Schmidt transform is employed to incorporate the backscattering coefficients of the primary image data as additional bands into the optical image, preserving the spectral information of the optical data and the texture information of the primary image data. The first 20 bands, retaining 90% of the information, are then fused to a 20m resolution using a weighted average method. The weighting formula is as follows: ,in, Let be the signal-to-noise ratio of the i-th band. , The mean value of the band. The standard deviation of the bands is used to ensure that high signal-to-noise ratio bands have a higher weight in the fusion process. Indicates the calculation of weights At that time, the total number of bands involved in the summation, where j is the index number and ; Then, the original image bands and basic indices are uniformly normalized to the [0,1] interval. The dataset format adopts a spatiotemporal cube structure, i.e.: Where t is the image acquisition time, x and y are spatial coordinates, and k is the band or index number, the data is organized in a structured manner in the spatiotemporal dimension to obtain a structured basic dataset.

[0008] The process of preloading optimization using a time-aware engine is as follows: A combination of time windows and spatial grids is used, with the time window... Set to 30 days, the block index formula is: Time block index ,in, The initial time, It refers to the window size; Spatial Block Index , ,in, For spatial coordinates, , The starting coordinates of the spatial region. For spatial grids; Data blocks within the target spatiotemporal range are quickly located using time block indexes and spatial block indexes, and the preloading priority is dynamically adjusted based on data popularity and user demand. The priority formula is as follows: ,in, Indicates priority. Based on current search popularity, The search popularity of historical data; The time-series-aware engine quickly locates data blocks within the target spatiotemporal range using time block indexes and spatial block indexes. Then, based on the data popularity and user demand calculated by the preloading priority formula, the time-series-aware engine dynamically adjusts the preloading order of the data blocks to complete the preloading optimization.

[0009] The process of preloading optimization using the LSTM model is as follows: While preloading priority based on the time-aware engine, missing values ​​in time-series data are imputed, and differentiated optimization methods are used for different missing durations: For short-term missing data of 30 days or less, linear interpolation is used directly. For data missing for more than 30 days, combined with historical data from the same period, intelligent dynamic imputation is performed using a pre-trained LSTM model. The formula is as follows: ,in, This represents the data value at time t. This is historical data from the same period. These are the predicted values ​​from the LSTM model, where α is the weight, ranging from 0.3 to 0.5. The LSTM model takes a structured dataset of the first three months as input and outputs the predicted values ​​for the missing period.

[0010] The time-series thematic data includes monthly average data, seasonal variation data, and trend analysis data.

[0011] The process of constructing a spatiotemporal knowledge graph is as follows: First, we define entities, attributes, and the relationships between entities and attributes. Entities include image entities, index entities, regional entities, and time-series product entities. We then obtain entity relationships based on time-series thematic data, including inclusion relationships, time-series associations, and spatial associations. Attributes are extracted from time-series thematic data based on entity relationships. Relationship extraction is achieved through a rule engine. Derivative relationships are automatically established based on data lineage. Inclusion relationships are calculated through spatial indexing. Time-series associations are linked to index entities through index IDs in product metadata. The final spatiotemporal knowledge graph is constructed using the extracted entity attributes and associations.

[0012] The data processing system based on the remote sensing information cloud platform includes: Data acquisition module: Set acquisition rules to acquire raw data streams from multiple satellite sources, and obtain primary image data with unified spatiotemporal reference through automated calibration and geometric correction; Data fusion module: Based on primary image data, it calls the workflow to fuse multi-source satellite data to generate a structured basic dataset; The time series optimization module uses a time series-aware engine and LSTM model to perform pre-loading optimization based on a structured basic dataset, and outputs time series thematic data. Data processing module: Based on time-series thematic data processing and spatiotemporal knowledge graph, it achieves optimal remote sensing information cloud platform data processing.

[0013] The present invention has the following beneficial effects: In this invention, firstly, the fusion of different types of remote sensing data, such as optical and SAR, is achieved through methods such as SIFT+RANSAC algorithm and Gram-Schmidt transform. This solves the problem of spatial resolution differences and spectral characteristic complementarity of data from different sensors, enabling the fused data to provide more comprehensive and accurate ground feature information and overcome the conflict of spatial resolution differences. Secondly, preloading optimization is achieved using a time-series-aware engine. By setting time windows and spatial grids, and employing block indexing formulas and preloading priority formulas, the processing efficiency of time-series data is improved. This effectively addresses data missingness issues; for example, linear interpolation is used for short-term missing data, while LSTM models are used for intelligent dynamic imputation of long-term missing data, ensuring the integrity and reliability of time-series data. Finally, by extracting entity attributes and relationships, attributes such as monthly averages, seasonal changes, and trend analysis are extracted from the time-series thematic data results. Relationships between entities are established, including time-series relationships and spatial relationships. The constructed spatiotemporal knowledge graph can intuitively reflect the changing patterns of ground features in the spatiotemporal dimension, supporting the intelligent leap from data query to decision-making suggestions. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of the data processing method based on a remote sensing information cloud platform proposed in this invention. Figure 2 This is a system block diagram of the data processing system based on the remote sensing information cloud platform proposed in this invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1: like Figure 1 As shown, the data processing method based on a remote sensing information cloud platform proposed in this invention includes: S1: Set acquisition rules to acquire raw data streams from multiple satellite sources, and obtain primary image data with unified spatiotemporal references through automated calibration and geometric correction; The process of acquiring raw data streams from multiple satellite sources is as follows: The multi-source satellite raw data stream includes optical data, synthetic aperture radar data, hyperspectral multi-type satellite data and auxiliary data. The optical satellite is selected, the SAR satellite is Hyperion, set to 242 bands, 30m resolution, and the auxiliary data includes SRTMDEM, set to 30m resolution and precipitation and temperature data from meteorological stations. Set data acquisition rules that specify spatiotemporal coverage requirements for the target area, such as 112°E-118°E, 32°N-38°N, with the following conditions: Optical satellites must guarantee at least one cloud-free image per month, with cloud cover <20%. Cloud pixels are determined by a combination of thresholds: blue light band reflectance >0.3 and near-infrared band reflectance <0.2. SAR satellites, unaffected by clouds, acquire two images per month to ensure temporal continuity, covering the target period. Data format is uniformly Level-1: optical data is saved in GeoTIFF format, hyperspectral data in HDF5 format, and primary image data in SLC format. Data acquisition is automated via the remote sensing information cloud platform API. Interface parameters include region boundary, time range, and sensor type. The acquisition request logic is represented as "data acquisition request = region boundary ∩ time range ∩ sensor type," ensuring that only target sensor data within the target area and specified time period is acquired. The process of automated calibration and geometric correction is as follows: Automated calibration converts raw digital quantization values ​​into radiance or reflectance. For optical satellites, radiance is calculated, and then reflectance is converted into surface reflectance. This process can eliminate the influence of solar distance and incident angle on the amount of radiation in the raw data stream of multi-source satellites. For SAR satellites, the backscattering coefficient formula is used: Where DN is the digital quantization value. K is the backscattering coefficient, and K is the calibration constant, which includes range attenuation and incident angle correction terms. It is provided by SAR metadata to ensure the radiometric consistency of primary image data at different times in the multi-source satellite raw data stream. Geometric correction aims to unify all imagery to the same coordinate system, such as WGS84-UTM projection, to eliminate geometric deviations caused by factors such as sensor pose and terrain. The process is as follows: First, control points (GCPs) are automatically extracted from the raw data stream using the SIFT algorithm. Each image requires at least 20 evenly distributed GCPs, with a matching error ≤ 1 pixel. For example, the error for a 20m resolution image should be ≤ 20m. A quadratic polynomial transform is used, and the GCP coordinates are calculated by fitting using the least squares method. For mountainous terrain, DEM data is introduced for terrain correction. The corrected reflectance formula is as follows: ,in, To correct the reflectivity, The original reflectance is given by α, which is the slope calculated from the DEM, and β is the difference between the slope aspect and the solar azimuth angle to eliminate reflectance deviation caused by terrain undulation. The primary image product output includes radiometrically calibrated and geometrically corrected image data in GeoTIFF format, ultimately yielding primary image data. Specifically, regarding radiometric consistency, the standard deviation of reflectance for the same type of ground features in images of the same area on different dates must be ≤5% to ensure the stability of radiometric values ​​over time. In terms of geometric accuracy, the registration error between the corrected image and the reference image must be ≤30m or 5m; Specifically, primary image data, primarily referring to radar image data generated after automated calibration processing by Sentinel-1 (SAR satellite), has its core metric being the backscattering coefficient. It is used to characterize the scattering characteristics of ground objects to radar waves, such as the significant differences in the backscattering coefficients of buildings, water bodies, and vegetation.

[0017] S2: Based on the primary image data, the workflow is called to fuse multi-source satellite data to generate a structured basic dataset; The structured base dataset is generated from primary image products. Its core function is to achieve spatiotemporal matching of data through multi-source data fusion and to calculate basic indices to impart semantic information to the data. Multi-source satellite data fusion needs to address the spatial resolution differences and complementary spectral characteristics of data from different sensors. The first step is data registration. The process of invoking the workflow to fuse multi-source satellite data is as follows: The SIFT+RANSAC algorithm is used to achieve automatic registration of optical and primary image data. The registration error must be ≤1 target resolution pixel. If the resolution is uniformly 20m, the error is ≤20m. The process of calling the workflow to fuse multi-source satellite data is as follows: Gram-Schmidt transform is used to incorporate the backscattering coefficients of the primary image data as "additional bands" into the optical image, preserving the spectral information of the optical data and the texture information of the primary image data, which is suitable for data supplementation in cloudy areas. Specifically, hyperspectral and optical fusion first requires selecting the top 20 bands that retain 90% of the information, and then fusing them to a 20m resolution using a weighted average method. The weighting formula is as follows: ,in, Let be the signal-to-noise ratio of the i-th band. , The mean value of the band. The standard deviation of the bands is used to ensure that high signal-to-noise ratio bands have a higher weight in the fusion process. Indicates the calculation of weights At that time, the total number of bands involved in the summation, where j is the index number and ; Structured datasets require special processing. The original image bands (reflectivity, backscattering coefficient) and basic indices are uniformly normalized to the [0,1] interval. The dataset format adopts a spatiotemporal cube structure, with dimensions defined as (time t, longitude x, latitude y, index k), i.e.: Where t is the image acquisition time, x and y are spatial coordinates, and k is the band or index number, such as k=1 corresponding to NDVI and k=2 corresponding to NDWI, to realize the structured organization of data in the spatiotemporal dimensions and finally obtain a structured basic dataset.

[0018] S3: Based on the structured basic dataset, the time series-aware engine and LSTM model are used for preloading and optimization to output time series thematic data; The time-series thematic data is generated based on structured basic datasets. It needs to be processed by a time-series-aware engine and preloaded with an LSTM model to extract time-series features and generate thematic data. It is designed for the time-series characteristics of structured datasets (time series have missing data and concentrated access hotspots). The core is to improve data access efficiency through block strategy and dynamic priority sorting. The process of preloading optimization using a time-aware engine is as follows: A combination of time windows and spatial grids is used, with the time window... Set to 30 days (monthly) to ensure coverage of temporal changes within a natural monthly cycle; Spatial grid Set to 1km × 1km, balancing spatial resolution and data block size, the block index formula is: Time block index ,in, The start time, It refers to the window size and the time block index. This is used to divide time into different block indexes, making it easier to manage and process time-series data; Spatial Block Index , ,in, For spatial coordinates, , The starting coordinates of the spatial region. For spatial grids, spatial block indexes and Used to divide space into different block indexes, making it easier to manage and process spatial data; Data blocks within the target spatiotemporal range are quickly located using time block indexes and spatial block indexes, and the preloading priority is dynamically adjusted based on data popularity and user demand. The priority formula is as follows: ,in, Indicates priority. Based on current search popularity, The query popularity of historical data is represented by normalized values, ranging from [0,1]. The time-series-aware engine can quickly locate data blocks within the target spatiotemporal range using time block indexes and spatial block indexes. Then, the time-series-aware engine dynamically adjusts the preloading order of data blocks based on the data heat and user demand calculated by the preloading priority formula. Data blocks with higher priority will be loaded into memory first so that they can be accessed quickly when needed. The process of preloading optimization using the LSTM model is as follows: While preloading priority based on the time-aware engine, missing values ​​in time-series data are imputed, and differentiated optimization methods are used for different missing durations: Short-term missing data is defined as 30 days or less. Since the data trend is relatively stable, linear interpolation is used, with the following formula: ,in, For the missing time period, For the time before the missing information, For the missing time, The value is a known value prior to the missing time. The value is a known value after the missing time, applicable to short-term missing optical data caused by cloud cover; Long-term missing data is defined as data longer than 30 days. Based on historical data from the same period, intelligent dynamic imputation is performed using a pre-trained LSTM model. The formula is as follows: ,in, This refers to historical data from the same period, such as data from the same month in 2022. (This represents the predicted value from the LSTM model, with α being the weight, ranging from 0.3 to 0.5, adjusted based on the correlation between historical and current data). The LSTM model takes a structured dataset from the three months prior to the missing data as input and outputs the predicted values ​​for the missing period. Combining this with historical data from the same period can reduce prediction bias. By preloading and optimizing the structured basic dataset using a time-aware engine and an LSTM model, valuable time-series information can be extracted from the structured basic dataset, and the extracted time-series information forms time-series thematic data. Specifically, time-series thematic data includes: Monthly average data: This is obtained by averaging the data from each month and can reflect the monthly variation pattern of land cover indicators. Seasonal change data: obtained by extracting key phenological periods from the normalized vegetation index (NDVI) time series curve, such as the greening period (the time point when NDVI first exceeds 0.3 and lasts for 7 days) and the yellowing period (the time point when NDVI first falls below 0.2 and lasts for 7 days), for crop growth cycle monitoring; Trend analysis data: By calculating the slope, the annual change trend of the indicator is assessed. It can be used to monitor the long-term changes of ground feature indicators, such as the NDVI trend of vegetation cover areas.

[0019] S4: Optimal remote sensing information cloud platform data processing based on spatiotemporal knowledge graphs for time-series thematic data processing; The process of acquiring a spatiotemporal knowledge graph is as follows: Spatiotemporal knowledge graphs are the core carrier for making remote sensing data queryable, and they need to be implemented by defining entities, attributes, and relationships. The entities include image entities, index entities, regional entities, and time-series product entities; Furthermore, image entity attributes include image ID, sensor type, acquisition time, and spatial range; index entity attributes include index ID, type, calculation time, spatial resolution, and numerical range; region entity attributes include region ID, administrative level, boundary, and area; and time-series product entity attributes include product ID, type, time granularity, and spatial granularity. Entity relationships are obtained from time-series thematic data, including inclusion relationships, time-series associations, and spatial associations. The process of obtaining the inclusion relationship is as follows: Define the association between the image and index entities and the region entities, and determine it through spatial intersection. If the intersection of the image spatial range and the region boundary is not empty ( If R is the region boundary and I is the image spatial range, then it is determined that the image covers the region. The process of obtaining time-series associations is as follows: Define the association between time-series product entities and index entities. For example, "Monthly NDVI Product" is generated from NDVI time-series data and associated through the "Index ID" field in the product metadata. The spatial association acquisition process is as follows: Define the association between entities in the region, and determine it using the Jaccard index, the formula of which is: When J > 0.5, it is determined to be an "inclusion" relationship. Entities representing different regions For judgment functions; Attributes are extracted from time-series thematic data based on entity relationships. The process is as follows: From the monthly average data of the time series thematic data, the average value of each month's land cover index is extracted as an attribute; from the seasonal change product data of the time series thematic data, the key phenological period attribute of vegetation is extracted; and from the trend analysis product data of the time series thematic data, the long-term change trend attribute of land cover index is extracted. Furthermore, attributes extracted from time-series thematic data, such as the greening period (the time point when NDVI first exceeds 0.3 and lasts for 7 days) and the yellowing period (the time point when NDVI first falls below 0.2 and lasts for 7 days), can help understand the changes in vegetation growth status in different seasons. These attributes are of great significance for agricultural production management and ecological environment monitoring. In addition, by calculating the slope to assess the annual variation trend of the indicator, it can be used to monitor the NDVI trend of vegetation-covered areas. These trend attributes can reflect whether the vegetation is recovering or degrading, which is of great value for assessing the quality of the ecological environment and formulating ecological protection policies. These attributes can help understand the overall situation of vegetation cover in the region and the differences in vegetation between different plots. The final spatiotemporal knowledge graph is constructed using the extracted entities, attributes, and relationships. The spatiotemporal knowledge graph is stored using the graph database Neo4j, with nodes storing entities, edges storing relationships, and attributes stored in key-value pairs, such as the "cloud amount" attribute value of an image node being 0.15. The spatiotemporal query interface is designed based on the SPARQL language. After new data is accessed, index calculation is automatically triggered to generate time series thematic data for map updates, realizing data processing and knowledge representation of remote sensing data. Specifically, when new data (new images in the remote sensing information cloud platform) is accessed, the data processing and map update process will be automatically triggered. After the new images are accessed, index calculation will be performed automatically to generate time-series thematic data and update the spatiotemporal knowledge graph. This update ensures that the spatiotemporal knowledge graph can reflect the latest data situation in a timely manner. When time-series thematic data is generated, the spatiotemporal knowledge graph will be automatically updated. For example, new indices or new time-series products calculated based on the time-series data will be added to the spatiotemporal knowledge graph, and the relevant entities and relationships will be updated. Specifically, the index is obtained by statistically analyzing all index values ​​within the target area, as follows: By statistically analyzing all index values ​​within a target area, the range of index values ​​can be obtained. For example, within a specific agricultural area, the minimum and maximum values ​​of the NDVI index can be statistically analyzed to obtain the range of NDVI values ​​for that area. In this way, the dynamic processing and knowledge-based representation of remote sensing data are realized, enabling the remote sensing information cloud platform to continuously enrich and improve its data resources, and provide users with efficient and accurate data query and analysis services in the form of knowledge graphs.

[0020] Example 2: like Figure 2 As shown, the data processing system based on the remote sensing information cloud platform includes: Data acquisition module: Set acquisition rules to acquire raw data streams from multiple satellite sources, and obtain primary image data with unified spatiotemporal reference through automated calibration and geometric correction; Data fusion module: Based on primary image data, it calls the workflow to fuse multi-source satellite data to generate a structured basic dataset; The time series optimization module uses a time series-aware engine and LSTM model to perform pre-loading optimization based on a structured basic dataset, and outputs time series thematic data. Data processing module: Based on time-series thematic data processing and spatiotemporal knowledge graph, it achieves optimal remote sensing information cloud platform data processing.

[0021] In the application, several formulas are calculated by removing dimensions and taking their numerical values. The formulas are established by collecting a large amount of data and simulating the most recent real situation. Some coefficients or weights in the formulas are set by those skilled in the art according to the actual situation, so they will not be elaborated here.

[0022] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0023] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data processing method based on a remote sensing information cloud platform, characterized in that, include: S1: Set acquisition rules to acquire raw data streams from multiple satellite sources, and obtain primary image data with unified spatiotemporal references through automated calibration and geometric correction; S2: Based on the primary image data, the workflow is called to fuse multi-source satellite data to generate a structured basic dataset; S3: Based on the structured basic dataset, the time series-aware engine and LSTM model are used for preloading and optimization to output time series thematic data; S4: Optimal remote sensing information cloud platform data processing based on spatiotemporal knowledge graphs for time-series thematic data processing.

2. The data processing method based on a remote sensing information cloud platform according to claim 1, characterized in that, The multi-source satellites include optical satellites and SAR satellites; The acquisition rules include that optical satellites guarantee at least one cloudless image per month with cloud cover of less than 20%, cloud pixels are determined by a combination of thresholds that the blue light band reflectance is greater than 0.3 and the near-infrared band reflectance is less than 0.2, SAR satellites acquire two images per month, the time span covers the target period, the data format is Level-1, and the optical data is saved in GeoTIFF format.

3. The data processing method based on a remote sensing information cloud platform according to claim 2, characterized in that, The automated calibration and geometric correction process is as follows: Automated calibration converts raw digital quantization values ​​into radiance or reflectance. For optical satellites, radiance is calculated, and then the radiance is converted into surface reflectance through reflectance conversion. For SAR satellites, the backscattering coefficient formula is used: Where DN is the digital quantization value. Here, K is the backscattering coefficient, and K is the calibration constant. Geometric correction automatically extracts control points (GCPs) from the raw data stream using the SIFT algorithm. Each image requires at least 20 evenly distributed GCPs with a matching error ≤ 1 pixel. A quadratic polynomial transform is used, and the GCP coordinates are calculated by fitting them using the least squares method. For mountainous terrain, DEM data is introduced for terrain correction. The corrected reflectance formula is as follows: ,in, To correct the reflectivity, α represents the original reflectivity, β represents the slope, and β represents the difference between the slope aspect and the solar azimuth angle.

4. The data processing method based on a remote sensing information cloud platform according to claim 1, characterized in that, The process of invoking the workflow to fuse multi-source satellite data is as follows: Gram-Schmidt transform is employed to incorporate the backscattering coefficients of the primary image data as additional bands into the optical image, preserving the spectral information of the optical data and the texture information of the primary image data. The first 20 bands, retaining 90% of the information, are then fused to a 20m resolution using a weighted average method. The weighting formula is as follows: ,in, Let be the signal-to-noise ratio of the i-th band. , The mean value of the band. The standard deviation of the band. Indicates the calculation of weights At that time, the total number of bands involved in the summation, where j is the index number and ; Then, the original image bands and basic indices are normalized to the [0,1] interval. The dataset format adopts a spatiotemporal cube structure, i.e.: Where t is the image acquisition time, x and y are spatial coordinates, and k is the band or index number, the data is organized in a structured manner in the spatiotemporal dimension to obtain a structured basic dataset.

5. The data processing method based on a remote sensing information cloud platform according to claim 4, characterized in that, The process of preloading optimization using a time-aware engine is as follows: A combination of time windows and spatial grids is used, with the time window... Set to 30 days, the block index formula is: Time block index ,in, The initial time, It refers to the window size; Spatial Block Index , ,in, For spatial coordinates, , The starting coordinates of the spatial region. For spatial grids; Data blocks within the target spatiotemporal range are located using time block indexes and spatial block indexes, and preloading priorities are dynamically adjusted based on data popularity and user demand. The priority formula is as follows: ,in, Indicates priority. Based on current search popularity, The search popularity of historical data; The time-series-aware engine quickly locates data blocks within the target spatiotemporal range using time block indexes and spatial block indexes. Then, based on the data popularity and user demand calculated by the preloading priority formula, the time-series-aware engine dynamically adjusts the preloading order of the data blocks to complete the preloading optimization.

6. The data processing method based on a remote sensing information cloud platform according to claim 5, characterized in that, The process of preloading optimization using the LSTM model is as follows: While preloading priority based on the time-aware engine, missing values ​​in time-series data are imputed, and differentiated optimization methods are used for different missing durations: For short-term missing data of 30 days or less, linear interpolation is used directly. For data missing for more than 30 days, combined with historical data from the same period, intelligent dynamic imputation is performed using a pre-trained LSTM model. The formula is as follows: , This represents the data value at time t, where, This is historical data from the same period. The values ​​are predicted by the LSTM model, where α is the weight, ranging from 0.3 to 0.

5. The LSTM model takes a structured dataset of the first three months as input and outputs the predicted values ​​for the missing period.

7. The data processing method based on a remote sensing information cloud platform according to claim 6, characterized in that, The time-series thematic data includes monthly average data, seasonal variation data, and trend analysis data.

8. The data processing method based on a remote sensing information cloud platform according to claim 7, characterized in that, The process of constructing a spatiotemporal knowledge graph is as follows: First, we define entities, attributes, and the relationships between entities and attributes. Entities include image entities, index entities, regional entities, and time-series product entities. We then obtain entity relationships based on time-series thematic data, including inclusion relationships, time-series associations, and spatial associations. Attributes are extracted from time-series thematic data based on entity relationships. Relationship extraction is achieved through a rule engine. Derivative relationships are automatically established based on data lineage. Inclusion relationships are calculated through spatial indexing. Time-series associations are linked to index entities through index IDs in product metadata. The final spatiotemporal knowledge graph is constructed using the extracted entity attributes and associations.

9. A data processing system based on a remote sensing information cloud platform, implemented according to any one of claims 1-8, characterized in that, include: Data acquisition module: Set acquisition rules to acquire raw data streams from multiple satellite sources, and obtain primary image data with unified spatiotemporal reference through automated calibration and geometric correction; Data fusion module: Based on primary image data, it calls the workflow to fuse multi-source satellite data to generate a structured basic dataset; The time series optimization module uses a time series-aware engine and LSTM model to perform pre-loading optimization based on a structured basic dataset, and outputs time series thematic data. Data processing module: Based on time-series thematic data processing and spatiotemporal knowledge graph, it achieves optimal remote sensing information cloud platform data processing.

Images