Method and system for inverting snow depth of source region of Yellow River based on GNSS-IR and airborne hyperspectral information of unmanned aerial vehicle

Abstract

The invention discloses a Yellow River source region snow depth inversion method and system based on GNSS-IR and unmanned aerial vehicle airborne hyperspectral information, and relates to the technical field of earth observation and remote sensing. Comprising the following steps: S1, data acquisition; s2, GNSS-IR data processing is carried out; s3, processing the hyperspectral image data of the unmanned aerial vehicle; and S4, multi-source data fusion. The RTK positioning technology guarantees accurate space-time matching of multi-source data, the generated fused snow depth data set has high space-time resolution, and solid support is provided for follow-up application such as snow melting runoff simulation.

Description

Technical Field

[0001] This invention relates to the field of Earth observation and remote sensing technology, and more specifically to a method and system for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information. Background Technology

[0002] As a climate-sensitive and crucial water resource replenishment area, monitoring snow cover dynamics in the Yellow River source region is of paramount importance. However, the complex terrain and strong spatiotemporal heterogeneity of snow cover in this region pose significant challenges to accurate snow depth inversion. Existing mainstream technologies have obvious limitations: 1. GNSS-IR technology: Limited by the "mode aliasing" problem of traditional empirical mode decomposition (EMD) and the use of single-frequency data from a single GPS system, the inversion accuracy is low (RMSE>0.05m), making it difficult to capture rapid snow cover changes.

[0003] 2. UAV hyperspectral technology: It typically uses all bands and fails to optimize for the spectral characteristics of snow cover in the study area (especially in the sensitive range of 900-1000nm), resulting in high data redundancy and low model efficiency.

[0004] 3. Data fragmentation: The two technologies mentioned above are independent of each other and cannot work together to generate snow depth datasets that simultaneously possess high temporal resolution (the advantage of GNSS-IR) and high spatial resolution (the advantage of UAVs), which limits their value in advanced applications such as snowmelt runoff simulation.

[0005] Therefore, proposing a method and system for snow depth inversion in the Yellow River source area based on GNSS-IR and UAV-borne hyperspectral information to solve the difficulties of existing technologies is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method and system for snow depth inversion in the Yellow River source area based on GNSS-IR and UAV-borne hyperspectral information. The generated fused snow depth dataset has both high spatiotemporal resolution and provides solid support for subsequent applications such as snowmelt runoff simulation.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information includes the following steps: S1. Data Acquisition: GNSS monitoring stations are deployed in the Yellow River source area to collect multi-frequency GNSS-IR data based on the satellite system. UAVs equipped with hyperspectral sensors are used to collect hyperspectral image data in the coverage area of ​​the GNSS monitoring stations. S2. GNSS-IR Data Processing: The CEEMDAN algorithm with added white noise is used to decompose the GNSS observation data. The correlation coefficient between each decomposed signal component and the GNSS observation signal during the snowless period is calculated. Signal components with a correlation coefficient > 0.8 are retained and reconstructed to obtain clean reflection signals. The optimal inversion frequency for each satellite system is determined, and low-precision frequency bands with system deviations > 0.8m in each system are removed. The snow depth inversion results corresponding to the optimal inversion frequency of each satellite system are fused using an inverse precision weighting strategy to obtain time-resolution snow depth data. S3. UAV hyperspectral image data processing: Extract spectral data in the 900-1000nm band from the hyperspectral image data; perform masking processing on the spectral data in the 900-1000nm band based on an NDSI exponent threshold of 0.3 to remove non-snow pixels; use the random forest algorithm to filter band data from the masked spectral data; input the filtered band data into the parameter-optimized XGBoost model to invert and obtain spatially resolved snow depth data; S4. Multi-source data fusion: RTK positioning is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral image data. The resulting temporal resolution snow depth data and spatial resolution snow depth data are spatiotemporally matched to generate a snow depth dataset for the Yellow River source area.

[0008] Optionally, the precision inverse weighting strategy in S2 is as follows: the root mean square error (RMSE) of snow depth inversion corresponding to the optimal inversion frequency of each satellite system is used as the basis for weight calculation. The higher the inversion precision and the smaller the RMSE, the greater the corresponding weight ratio. The root mean square error (RMSE) of the snow depth data obtained after fusion is ≤0.024m.

[0009] Optionally, the mean square error (MSE) of the spatial resolution snow depth data obtained by the XGBoost model inversion in S3 is ≤0.52cm.

[0010] Optionally, the NDSI index in S3 is calculated using the reflectance of the green band and shortwave infrared band in the hyperspectral image. The specific calculation formula is as follows:

[0011] in, For green band reflectivity, This refers to the reflectivity in the shortwave infrared band.

[0012] Optionally, in S4, RTK positioning is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral imagery data. The specific details of spatiotemporal matching of the obtained temporal resolution snow depth data and spatial resolution snow depth data are as follows: Using the timestamps of GNSS monitoring stations as a reference, the spatial resolution snow depth data is time-aligned, and using the coordinates after RTK positioning calibration as a reference, the temporal resolution snow depth data is spatially interpolated to ensure consistency in the spatiotemporal dimensions.

[0013] A snow depth inversion system for the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information, and a snow depth inversion method for the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information applying any of the above, including: a data acquisition module, a GNSS-IR data processing module, a UAV hyperspectral image data processing module, and a multi-source data fusion module; The data acquisition module is connected to the input end of the GNSS-IR data processing module. It is used to deploy GNSS monitoring stations in the Yellow River source area, collect multi-frequency GNSS-IR data based on the satellite system, and use UAVs equipped with hyperspectral sensors to collect hyperspectral image data in the coverage area of ​​the GNSS monitoring stations. The GNSS-IR data processing module, connected to the input of the UAV hyperspectral image data processing module, is used to perform signal decomposition on GNSS observation data using the CEEMDAN algorithm with added white noise. It calculates the correlation coefficient between each decomposed signal component and the GNSS observation signal during the snowless period, retains signal components with a correlation coefficient > 0.8, and reconstructs the pure reflection signal. It determines the optimal inversion frequency for each satellite system, removes low-precision frequency bands with system bias > 0.8m, and fuses the snow depth inversion results corresponding to the optimal inversion frequencies of each satellite system using an inverse precision weighting strategy to obtain time-resolution snow depth data. The UAV hyperspectral image data processing module is connected to the input of the multi-source data fusion module. It is used to extract spectral data in the 900-1000nm band from the hyperspectral image data. Based on the NDSI exponent threshold of 0.3, the spectral data in the 900-1000nm band is masked to remove non-snow pixels. The random forest algorithm is used to filter the band data from the masked spectral data. The filtered band data is then input into the parameter-optimized XGBoost model to invert and obtain spatially resolved snow depth data. The multi-source data fusion module is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral image data using RTK positioning, and to perform spatiotemporal matching on the obtained temporal resolution snow depth data and spatial resolution snow depth data to generate a snow depth dataset for the Yellow River source area.

[0014] As can be seen from the above technical solution, compared with the prior art, the present invention provides a method and system for snow depth inversion in the Yellow River source area based on GNSS-IR and UAV-borne hyperspectral information, which has the following beneficial effects: (1) The CEEMDAN algorithm of this invention, combined with the correlation coefficient screening method, reduces "mode mixing", improves the peak-to-noise ratio by an average of 18% ± 2%, and increases the number of effective orbits by 25%, providing a cleaner signal basis for snow depth inversion; (2) The selection of the optimal frequency for each GNSS system improves the inversion accuracy of a single system by 20% to 25% compared to “mixed use of all frequencies”; after weighted fusion of multiple systems, the RMSE drops to 0.024m, which is 11% higher than the accuracy of a single Galileo system (optimal single system) and more than 35% higher than the accuracy of a traditional single GPS system. (3) After band filtering, the data redundancy of the UAV hyperspectral data is significantly reduced. The test set error (MSE≤0.52cm) of the XGBoost optimized model is much lower than that of the traditional model, realizing high-precision meter-level spatial resolution snow depth mapping. (4) RTK positioning technology ensures accurate spatiotemporal matching of multi-source data, and the generated fused snow depth dataset has both high spatiotemporal resolution, providing solid support for subsequent applications such as snowmelt runoff simulation. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 A flowchart of a method for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information is provided by the present invention. Figure 2 This invention provides a schematic diagram of a snow depth inversion system for the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information. Detailed Implementation

[0017] 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.

[0018] See Figure 1 As shown, this invention discloses a method for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information, including the following steps: S1. Data Acquisition: GNSS monitoring stations are deployed in the Yellow River source area to collect multi-frequency GNSS-IR data based on the satellite system. UAVs equipped with hyperspectral sensors are used to collect hyperspectral image data in the coverage area of ​​the GNSS monitoring stations. S2. GNSS-IR Data Processing: The CEEMDAN algorithm with added white noise is used to decompose the GNSS observation data. The correlation coefficient between each decomposed signal component and the GNSS observation signal during the snowless period is calculated. Signal components with a correlation coefficient > 0.8 are retained and reconstructed to obtain clean reflection signals. The optimal inversion frequency for each satellite system is determined, and low-precision frequency bands with system deviations > 0.8m in each system are removed. The snow depth inversion results corresponding to the optimal inversion frequency of each satellite system are fused using an inverse precision weighting strategy to obtain time-resolution snow depth data. S3. UAV hyperspectral image data processing: Extract spectral data in the 900-1000nm band from the hyperspectral image data; perform masking processing on the spectral data in the 900-1000nm band based on an NDSI exponent threshold of 0.3 to remove non-snow pixels; use the random forest algorithm to filter band data from the masked spectral data; input the filtered band data into the parameter-optimized XGBoost model to invert and obtain spatially resolved snow depth data; S4. Multi-source data fusion: RTK positioning is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral image data. The resulting temporal resolution snow depth data and spatial resolution snow depth data are spatiotemporally matched to generate a snow depth dataset for the Yellow River source area.

[0019] Furthermore, the precision inverse weighting strategy in S2 is as follows: the root mean square error (RMSE) of snow depth inversion corresponding to the optimal inversion frequency of each satellite system is used as the basis for weight calculation. The higher the inversion precision and the smaller the RMSE, the greater the corresponding weight ratio. The root mean square error (RMSE) of the snow depth data obtained after fusion is ≤0.024m.

[0020] Furthermore, the mean square error (MSE) of the spatial resolution snow depth data obtained by the XGBoost model inversion in S3 is ≤0.52cm.

[0021] Furthermore, the NDSI index in S3 is calculated using the reflectance of the green band and shortwave infrared band in the hyperspectral image. The specific calculation formula is as follows:

[0022] in, For green band reflectivity, This refers to the reflectivity in the shortwave infrared band.

[0023] Furthermore, in S4, RTK positioning is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral imagery data. The specific details of spatiotemporal matching of the obtained temporal resolution snow depth data and spatial resolution snow depth data are as follows: Using the timestamps of GNSS monitoring stations as a reference, the spatial resolution snow depth data is time-aligned, and using the coordinates after RTK positioning calibration as a reference, the temporal resolution snow depth data is spatially interpolated to ensure consistency in the spatiotemporal dimensions.

[0024] In one specific embodiment, the following is included: Typical monitoring areas were selected in the Yellow River source region, GNSS monitoring stations were deployed, and multi-frequency observation data from four major systems, GPS, GLONASS, Galileo, and BDS, were collected simultaneously. Drones equipped with hyperspectral sensors were used to take aerial photos in the monitoring area to obtain hyperspectral image data. The CEEMDAN algorithm was used to decompose the GNSS observation signal, and the correlation coefficient between each component and the snowless period signal was calculated. Components with a correlation coefficient > 0.8 were retained for reconstruction. The optimal inversion frequency of each satellite system was determined, and the data from multiple systems were fused by a precision inverse weighting strategy to obtain high temporal resolution snow depth data. Data in the 900-1000nm band of hyperspectral images were extracted, and non-snowy pixels were removed using an NDSI exponential threshold (0.3) mask. The 10 most important bands were selected by random forest algorithm and input into the optimized XGBoost model to obtain spatially resolved snow depth data. By using RTK positioning technology to calibrate the coordinates of GNSS stations and UAV imagery, and performing spatiotemporal matching of the two types of snow depth data, a snow depth dataset of the Yellow River source area with both 15-minute / time resolution and spatial resolution is generated, enabling snow depth inversion in the Yellow River source area based on GNSS-IR and UAV-borne hyperspectral information.

[0025] A snow depth inversion system for the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information, such as Figure 2 As shown, a method for snow depth inversion in the Yellow River source area based on GNSS-IR and UAV-borne hyperspectral information, applying any of the above, includes: a data acquisition module, a GNSS-IR data processing module, a UAV hyperspectral image data processing module, and a multi-source data fusion module; The data acquisition module is connected to the input end of the GNSS-IR data processing module. It is used to deploy GNSS monitoring stations in the Yellow River source area, collect multi-frequency GNSS-IR data based on the satellite system, and use UAVs equipped with hyperspectral sensors to collect hyperspectral image data in the coverage area of ​​the GNSS monitoring stations. The GNSS-IR data processing module, connected to the input of the UAV hyperspectral image data processing module, is used to perform signal decomposition on GNSS observation data using the CEEMDAN algorithm with added white noise. It calculates the correlation coefficient between each decomposed signal component and the GNSS observation signal during the snowless period, retains signal components with a correlation coefficient > 0.8, and reconstructs the pure reflection signal. It determines the optimal inversion frequency for each satellite system, removes low-precision frequency bands with system bias > 0.8m, and fuses the snow depth inversion results corresponding to the optimal inversion frequencies of each satellite system using an inverse precision weighting strategy to obtain time-resolution snow depth data. The UAV hyperspectral image data processing module is connected to the input of the multi-source data fusion module. It is used to extract spectral data in the 900-1000nm band from the hyperspectral image data. Based on the NDSI exponent threshold of 0.3, the spectral data in the 900-1000nm band is masked to remove non-snow pixels. The random forest algorithm is used to filter the band data from the masked spectral data. The filtered band data is then input into the parameter-optimized XGBoost model to invert and obtain spatially resolved snow depth data. The multi-source data fusion module is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral image data using RTK positioning, and to perform spatiotemporal matching on the obtained temporal resolution snow depth data and spatial resolution snow depth data to generate a snow depth dataset for the Yellow River source area.

[0026] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0027] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information, characterized in that, Includes the following steps: S1. Data Acquisition: GNSS monitoring stations are deployed in the Yellow River source area to collect multi-frequency GNSS-IR data based on the satellite system. UAVs equipped with hyperspectral sensors are used to collect hyperspectral image data in the coverage area of ​​the GNSS monitoring stations. S2. GNSS-IR Data Processing: The CEEMDAN algorithm with added white noise is used to decompose the GNSS observation data. The correlation coefficient between each decomposed signal component and the GNSS observation signal during the snowless period is calculated. Signal components with a correlation coefficient > 0.8 are retained and reconstructed to obtain clean reflection signals. The optimal inversion frequency for each satellite system is determined, and low-precision frequency bands with system deviations > 0.8m in each system are removed. The snow depth inversion results corresponding to the optimal inversion frequency of each satellite system are fused using an inverse precision weighting strategy to obtain time-resolution snow depth data. S3. UAV hyperspectral image data processing: Extract spectral data in the 900-1000nm band from the hyperspectral image data; perform masking processing on the spectral data in the 900-1000nm band based on an NDSI exponent threshold of 0.3 to remove non-snow pixels; use the random forest algorithm to filter band data from the masked spectral data; input the filtered band data into the parameter-optimized XGBoost model to invert and obtain spatially resolved snow depth data; S4. Multi-source data fusion: RTK positioning is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral image data. The resulting temporal resolution snow depth data and spatial resolution snow depth data are spatiotemporally matched to generate a snow depth dataset for the Yellow River source area.

2. The method for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information as described in claim 1, characterized in that, The precision inverse weighting strategy in S2 is as follows: the root mean square error (RMSE) of snow depth inversion corresponding to the optimal inversion frequency of each satellite system is used as the basis for weight calculation. The higher the inversion precision and the smaller the RMSE, the larger the corresponding weight ratio. The root mean square error (RMSE) of the snow depth data obtained after fusion is ≤0.024m.

3. The method for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information as described in claim 1, characterized in that, The mean square error (MSE) of the spatial resolution snow depth data obtained by inversion using the XGBoost model in S3 is ≤0.52cm.

4. The method for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information as described in claim 1, characterized in that, The NDSI index in S3 is calculated using the reflectance of the green band and shortwave infrared band in hyperspectral images. The specific calculation formula is as follows: in, For green band reflectivity, This refers to the reflectivity in the shortwave infrared band.

5. The method for snow depth inversion in the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information according to claim 1, characterized in that, In S4, RTK positioning is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral imagery data. The specific details of spatiotemporal matching of the obtained temporal resolution snow depth data and spatial resolution snow depth data are as follows: Using the timestamps of GNSS monitoring stations as a reference, the spatial resolution snow depth data is time-aligned, and using the coordinates after RTK positioning calibration as a reference, the temporal resolution snow depth data is spatially interpolated to ensure consistency in the spatiotemporal dimensions.

6. A snow depth inversion system for the Yellow River source region based on GNSS-IR and UAV-borne hyperspectral information, characterized in that, The method for snow depth inversion in the Yellow River source area based on GNSS-IR and UAV-borne hyperspectral information according to any one of claims 1-5 includes: a data acquisition module, a GNSS-IR data processing module, a UAV hyperspectral image data processing module, and a multi-source data fusion module; The data acquisition module is connected to the input end of the GNSS-IR data processing module. It is used to deploy GNSS monitoring stations in the Yellow River source area, collect multi-frequency GNSS-IR data based on the satellite system, and use UAVs equipped with hyperspectral sensors to collect hyperspectral image data in the coverage area of ​​the GNSS monitoring stations. The GNSS-IR data processing module, connected to the input of the UAV hyperspectral image data processing module, is used to perform signal decomposition on GNSS observation data using the CEEMDAN algorithm with added white noise. It calculates the correlation coefficient between each decomposed signal component and the GNSS observation signal during the snowless period, retains signal components with a correlation coefficient > 0.8, and reconstructs the pure reflection signal. It determines the optimal inversion frequency for each satellite system, removes low-precision frequency bands with system bias > 0.8m, and fuses the snow depth inversion results corresponding to the optimal inversion frequencies of each satellite system using an inverse precision weighting strategy to obtain time-resolution snow depth data. The UAV hyperspectral image data processing module is connected to the input of the multi-source data fusion module. It is used to extract spectral data in the 900-1000nm band from the hyperspectral image data. Based on the NDSI exponent threshold of 0.3, the spectral data in the 900-1000nm band is masked to remove non-snow pixels. The random forest algorithm is used to filter the band data from the masked spectral data. The filtered band data is then input into the parameter-optimized XGBoost model to invert and obtain spatially resolved snow depth data. The multi-source data fusion module is used to perform coordinate calibration on GNSS-IR data and UAV hyperspectral image data using RTK positioning, and to perform spatiotemporal matching on the obtained temporal resolution snow depth data and spatial resolution snow depth data to generate a snow depth dataset for the Yellow River source area.

Images