Ice and snow product inversion method and equipment based on geological resource hyperspectral satellite

Abstract

The invention provides an ice and snow product inversion method and device based on a geological resource hyperspectral satellite, and relates to the technical field of remote sensing data processing.The inversion method comprises the steps that remote sensing image data and auxiliary data of a research area are obtained through the geological resource hyperspectral satellite, the remote sensing image data are preprocessed, and the preprocessed remote sensing image data are obtained; obtaining real reflectivity and spectral data of the earth surface; using a multilevel decision tree accumulated snow recognition algorithm to generate an accumulated snow range data product; obtaining a normalized snow index according to the spectral data; inputting the real earth surface reflectivity and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data; and inputting the auxiliary data, the normalized snow index and the remote sensing image data into a preset deep belief network to obtain accumulated snow area proportion data. According to the invention, based on the high-frequency observation capability of a geological resource hyperspectral satellite, high-precision ice and snow product inversion is realized under a complex terrain condition through the combination of a satellite remote sensing technology and a machine learning algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing data processing, and more particularly, to a method and device for retrieving ice and snow products based on a geological resource hyperspectral satellite. Background Art

[0002] Ice and snow, as important fresh water resources on Earth, play a crucial role in global climate regulation, hydrological cycle and ecosystem maintenance. The snow-covered surface has significant solar radiation reflection ability, which can regulate the energy exchange between the surface and the atmosphere, thus affecting regional and global climate. For example, the Qinghai-Tibet Plateau is known as the "water tower" of Asia, and its snow cover dynamics not only directly affect the water resource utilization in downstream areas, but also have a profound impact on the global climate system. Therefore, it is particularly important to construct an accurate, efficient and snow cover monitoring method adaptable to complex terrain conditions.

[0003] Traditional snow cover monitoring methods mainly rely on ground station observations and manual measurements. Although they have high accuracy in a local range, there are problems such as limited coverage, low time update frequency and discontinuous data. With the development of satellite remote sensing technology, using remote sensing technology to monitor ice and snow parameters has become the mainstream, which can achieve large-scale and high-frequency acquisition of ice and snow parameters, providing strong support for global climate change monitoring and water resource management. For example, optical remote sensing technology can be used to obtain the snow cover extent, while passive microwave brightness temperature data can be used to obtain the snow depth and snow water equivalent. These information are crucial for understanding the changes of ice and snow and their impact on water resources.

[0004] However, the observation capabilities and inversion algorithms of existing high-resolution meteorological satellites still have limitations, resulting in relatively low resolutions of existing snow products, which are difficult to meet the requirements of high-precision and real-time monitoring. Especially in areas with complex terrain and large local variations, the problem of insufficient accuracy is still prominent. Summary of the Invention

[0005] The problem solved by the present invention is how to improve the relatively low resolution of existing snow products in areas with complex terrain and large local variations, which are difficult to meet the requirements of high-precision and real-time monitoring.

[0006] To solve the above problems, the present invention provides a method and device for retrieving ice and snow products based on a geological resource hyperspectral satellite.

[0007] In a first aspect, the present invention provides a method for retrieving ice and snow products based on a geological resource hyperspectral satellite, which is applied to the geological resource hyperspectral satellite and includes:

[0008] Obtain remote sensing image data of the study area through the geological resource hyperspectral satellite, and preprocess the remote sensing image data to obtain remote sensing image data, where the remote sensing image data includes surface true reflectance and spectral data;

[0009] Obtain the normalized snow index according to the spectral data;

[0010] Obtain the auxiliary data of the study area, and input the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data, where the preset snow water equivalent inversion model is constructed based on the XGBoost machine learning module;

[0011] Input the auxiliary data, the normalized snow index, and the remote sensing image data into a preset deep belief network to obtain snow cover area ratio data;

[0012] Among them, the ice and snow product inversion result includes the normalized snow index, the snow water equivalent data, and the snow cover area ratio data.

[0013] Optionally, the ice and snow product inversion result further includes a snow cover extent data product, and the ice and snow product inversion method based on the geological resource hyperspectral satellite further includes:

[0014] Input the spectral data, the auxiliary data, and the normalized snow index into a preset snow cover identification model to obtain a temporary snow cover extent data product; the preset snow cover identification model is constructed based on a multi-level decision tree;

[0015] Obtain a microwave snow depth data set, and based on a spatio-temporal interpolation algorithm, fuse the microwave snow depth data set with the temporary snow cover extent data product to obtain the snow cover extent data product.

[0016] Optionally, the spectral data includes multiple band data, and the auxiliary data includes a normalized vegetation index, brightness temperature data, and DEM data; the step of inputting the spectral data, the auxiliary data, and the normalized snow index into a preset snow cover identification model to obtain a temporary snow cover extent data product includes:

[0017] Determine the first preset band data in the spectral data, where the first preset band data includes the seventh band data, the fourteenth band data, and the eighteenth band data;

[0018] Input the seventh band data, the fourteenth band data, the eighteenth band data, the normalized vegetation index, the brightness temperature data, the DEM data, and the normalized snow index into the preset snow cover identification model to obtain the temporary snow cover extent data product. [[ID=�2]]

[0019] Optionally, the spectral data includes a plurality of band data; obtaining the normalized snow index according to the spectral data includes:

[0020] Determine the second preset band data in the spectral data, where the second preset band data includes the fifth band data and the nineteenth band data;

[0021] Obtain the normalized snow index according to the fifth band data and the nineteenth band data.

[0022] Optionally, obtaining the normalized snow index according to the fifth band data and the nineteenth band data includes:

[0023] Obtain the normalized snow index according to the fifth band data and the nineteenth band data through Equation 1;

[0024] Equation 1 includes:

[0025]

[0026] where, NDSI is the normalized snow index, R DD-1-b5 is the fifth band data, R DD-1-b19 is the nineteenth band data.

[0027] Optionally, the spectral data includes a plurality of band data, the auxiliary data includes DEM data and normalized vegetation index, and the remote sensing image data further includes geographic coordinates; inputting the auxiliary data, the normalized snow index, and the remote sensing image data into a preset deep belief network to obtain snow cover area ratio data includes:

[0028] Determine the third preset band data in the spectral data, where the third preset band data includes the fourteenth band data and the eighteenth band data;

[0029] Input the DEM data, the normalized vegetation index, the normalized snow index, the geographic coordinates, the fourteenth band data, and the eighteenth band data into the preset deep belief network to obtain the snow cover area ratio data.

[0030] Optionally, the auxiliary data includes snow pressure data and snow density data; inputting the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data includes:

[0031] Input the surface true reflectance, the snow pressure data, and the snow density data into the preset snow water equivalent inversion model to obtain the snow water equivalent data.

[0032] Optionally, preprocessing the remote sensing image data to obtain remote sensing image data, including:

[0033] Performing radiometric calibration on the remote sensing image data to obtain processed remote sensing image data; and converting the processed remote sensing image data according to a preset calibration coefficient to obtain temporary image data;

[0034] Performing atmospheric correction on the temporary image data to obtain the remote sensing image data.

[0035] In a second aspect, the present invention provides an ice and snow product inversion device based on a geological resource hyperspectral satellite, which is applied to the geological resource hyperspectral satellite, including:

[0036] An acquisition unit, configured to acquire remote sensing image data of a research area through the geological resource hyperspectral satellite, and preprocess the remote sensing image data to obtain remote sensing image data, where the remote sensing image data includes surface true reflectance and spectral data;

[0037] A processing unit, configured to obtain a normalized snow index according to the spectral data;

[0038] The acquisition unit is further configured to acquire auxiliary data of the research area;

[0039] The processing unit is further configured to input the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data, where the preset snow water equivalent inversion model is constructed based on an XGBoost machine learning module;

[0040] The processing unit is further configured to input the auxiliary data, the normalized snow index, and the remote sensing image data into a preset deep belief network to obtain snow cover area ratio data; wherein, the ice and snow product inversion result includes the normalized snow index, the snow water equivalent data, and the snow cover area ratio data.

[0041] In a third aspect, the present invention provides an ice and snow product inversion system based on a geological resource hyperspectral satellite, including a memory and a processor; the memory is used to store a computer program; the processor is used to implement the ice and snow product inversion method based on the geological resource hyperspectral satellite as described in the first aspect when executing the computer program.

[0042] The beneficial effects of the snow and ice product inversion method, device, and system based on a geological resource hyperspectral satellite of the present invention are as follows: Remote sensing image data of the research area is obtained through a geological resource hyperspectral satellite and preprocessed to obtain remote sensing image data containing the true surface reflectance and spectral data. This process ensures the accuracy and usability of the data. Among them, the geological resource hyperspectral satellite has a higher spatial resolution and rich spectral bands, capable of providing a large amount of spectral information for optical inversion of snow and ice parameters.

[0043] The Normalized Difference Snow Index (NDSI) is calculated based on the preprocessed spectral data. This index is an important indicator reflecting the snow cover situation, which helps improve the accuracy of subsequent snow cover monitoring. At the same time, the true surface reflectance and auxiliary data of the research area are input into a preset snow water equivalent inversion model constructed based on the XGBoost machine learning module to obtain snow water equivalent data.

[0044] During the calculation of the snow cover area ratio, the auxiliary data, Normalized Difference Snow Index, and remote sensing image data are input into a preset Deep Belief Network (DBN) to accurately calculate the snow cover area ratio. This data is crucial for evaluating the distribution and change trend of snow and ice resources. This study innovatively introduces deep learning algorithms, and through the DBN model, fits the complex non-linear relationship between the snow cover area ratio and multiple auxiliary factors, thereby improving the inversion accuracy and regional adaptability.

[0045] Finally, the Normalized Difference Snow Index, snow water equivalent data, and snow cover area ratio data are combined to obtain the snow and ice product inversion result, providing a scientific basis for water resource management and climate change research.

[0046] In summary, the present invention realizes high-precision snow and ice product inversion under complex terrain conditions through the combination of satellite remote sensing technology and machine learning algorithms. At the same time, based on the high-frequency observation ability of the geological resource hyperspectral satellite, this method can monitor snow cover changes in real time and provide information required for water resource management in a timely manner. This method comprehensively utilizes spectral data and auxiliary data, realizes the efficient integration of information, and improves the comprehensiveness and accuracy of snow cover monitoring. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is one of the flow diagrams of a snow and ice product inversion method based on a geological resource hyperspectral satellite according to an embodiment of the present invention;

[0048] Figure 2 It is a schematic diagram of the snow zoning situation on the Qinghai-Tibet Plateau according to an embodiment of the present invention;

[0049] Figure 3 It is another flow diagram of a snow and ice product inversion method based on a geological resource hyperspectral satellite according to an embodiment of the present invention;

[0050] Figure 4 It is a structural diagram of a preset deep belief network according to an embodiment of the present invention;

[0051] Figure 5 It is a schematic diagram of an ice and snow product inversion device based on a geological resource hyperspectral satellite according to an embodiment of the present invention. Specific embodiments

[0052] To make the above objects, features and advantages of the present invention more obvious and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Although some embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be construed as limited to the embodiments described herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present invention. It should be understood that the drawings and embodiments of the present invention are only for exemplary purposes and are not used to limit the protection scope of the present invention.

[0053] The term "including" and its variations used herein are open-ended, that is, "including but not limited to"; the term "based on" is "at least partially based on"; the term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments"; the term "optionally" means "optional embodiment". The relevant definitions of other terms will be given in the following description. It should be noted that the concepts such as "first" and "second" mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence relationship of the functions performed by these devices, modules or units.

[0054] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only for illustrative purposes and are not used to limit the scope of these messages or information.

[0055] The embodiments of the present invention provide an ice and snow product inversion method and device based on a geological resource hyperspectral satellite.

[0056] As Figure 1 shown, an ice and snow product inversion method based on a geological resource hyperspectral satellite provided by an embodiment of the present invention, the ice and snow product inversion method based on a geological resource hyperspectral satellite includes;

[0057] Step S100, obtaining remote sensing image data of a research area through the geological resource hyperspectral satellite, and preprocessing the remote sensing image data to obtain remote sensing image data, wherein the remote sensing image data includes surface true reflectance and spectral data;

[0058] Specifically, using the sensors of the geological resource hyperspectral satellite, remote sensing images covering the study area are acquired. These images can contain information in different spectral bands, providing multi-dimensional data of surface features.

[0059] Among them, the preprocessing process of remote sensing image data can include noise reduction processing. For example, the acquired remote sensing images are subjected to noise reduction processing to eliminate artifacts and noise caused by atmospheric effects, sensor noise, or other interference sources. Then image correction is carried out. For example, geometric correction and radiometric correction can be performed to ensure the accurate geographical location in the image and eliminate the radiometric differences caused by imaging under different times and conditions. This may include using ground control points for geometric correction and radiometric response correction according to the sensor characteristics. Spectral data extraction, such as extracting the required spectral data from the corrected images to generate remote sensing image data containing information in multiple bands. These bands will be used for subsequent monitoring of snow and ice parameters.

[0060] Through preprocessing techniques such as noise reduction and radiometric correction, the accuracy of remote sensing images is ensured, making the obtained surface reflectance more real and reliable. This process significantly improves the quality of the basic data for subsequent analysis and parameter inversion. Moreover, the remote sensing image data generated by the geological resource hyperspectral satellite contains rich spectral information and diverse bands, which can support various environmental and ecological monitoring needs and provide the necessary data basis for subsequent snow and ice parameter inversion.

[0061] Step S200, obtaining the normalized difference snow index according to the spectral data;

[0062] Specifically, the normalized difference snow index (NDSI) is an index that can effectively reflect the snow cover situation. Currently, it is a common method for optical remote sensing to extract snow cover and is widely used in the differentiation between snow cover and clouds, and surface background; NDSI is considered an ideal cloud-snow identifier. Through NDSI, the snow-covered and non-snow-covered areas can be effectively distinguished, thus providing a reliable basis for subsequent snow and ice parameter inversion.

[0063] Step S300, obtaining the auxiliary data of the study area, and inputting the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data. The preset snow water equivalent inversion model is constructed based on the XGBoost machine learning module;

[0064] Specifically, the auxiliary data refers to the relevant information that can affect the snow water equivalent, usually including the following categories: topographic feature data, such as elevation, slope, and aspect; meteorological data, such as temperature, precipitation, and humidity; other environmental parameters, such as vegetation coverage rate and land use type. These data can be obtained through ground observation stations, meteorological data, or other remote sensing technologies to ensure their timeliness and accuracy.

[0065] When actually performing snow water equivalent inversion, an XGBoost algorithm is used to construct a snow water equivalent inversion model. XGBoost is an enhanced learning algorithm based on decision trees, with the characteristics of high efficiency and accuracy, and is suitable for processing large-scale data sets.

[0066] After inputting the obtained surface true reflectance and auxiliary data into the preset XGBoost snow water equivalent inversion model, the model outputs the corresponding snow water equivalent data, reflecting the snow moisture content in the study area. Among them, the XGBoost algorithm constructs a multi-layer decision tree, which is a flexible gradient boosting decision tree (GBDT) model and can effectively handle complex non-linear relationships, thereby significantly improving the accuracy of snow water equivalent inversion and reducing model bias.

[0067] Introducing auxiliary data enables the inversion model to comprehensively consider various factors affecting the snow water equivalent, improving the overall performance of the model. By integrating different types of data, the representativeness of the model is enhanced. In addition, the well-trained XGBoost model has strong adaptability and can adapt to snow water equivalent inversion under different geographical and climatic conditions, which helps to broaden its application scenarios.

[0068] The above process not only improves the accuracy and efficiency of snow water equivalent inversion by integrating surface true reflectance and multi-dimensional auxiliary data, but also provides strong support for scientific research and applications in related fields.

[0069] Step S400: Input the auxiliary data, the normalized snow index, and the remote sensing image data into a preset deep belief network to obtain the snow cover area ratio data; among them, the ice and snow product inversion results include the normalized snow index, the snow water equivalent data, and the snow cover area ratio data.

[0070] Specifically, the auxiliary data may include geographical information related to the study area (such as elevation, slope, terrain type) and meteorological data (such as temperature, humidity, precipitation), which provide rich environmental context information for the model. The normalized snow index (NDSI) is calculated from remote sensing images, which reflects the reflection difference between snow cover and the background (such as water bodies, soil), and can effectively indicate the coverage of the snow surface. The remote sensing image data contains information such as the surface true reflectance obtained through preprocessing, multi-spectral data, and geographical coordinates, providing important data on surface characteristics and snow cover.

[0071] Input the auxiliary data, normalized snow index, and remote sensing image data into the trained deep belief network to obtain the corresponding snow cover area ratio data. As an important indicator describing the snow cover degree, the monitoring accuracy of the snow cover area ratio (FSC) is significantly affected by multiple factors, such as terrain, geographical location, surface temperature, and vegetation index. In complex terrain regions such as the Qinghai-Tibet Plateau (as shown in Figure 2 the schematic diagram of the snow partition situation in the Qinghai-Tibet Plateau), the traditional FSC inversion methods usually have problems of low accuracy and poor adaptability, especially in the transition and edge zones between snow and non-snow areas, which makes accurate estimation difficult. Therefore, this method introduces a deep belief network (DBN) model to fit the complex non-linear relationship between the snow cover area ratio and multiple auxiliary factors, thereby improving the inversion accuracy and regional adaptability.

[0072] Among them, the deep belief network is a deep learning framework based on the generative model, which contains multiple layers of neural networks and can automatically learn the high-level features of the input data. The DBN usually consists of multiple restricted Boltzmann machines (RBMs). Each RBM layer extracts features from the previous layer through unsupervised learning and passes the results to the next layer. Finally, it is adjusted through supervised learning and the network output is optimized through backpropagation.

[0073] The deep belief network can automatically extract the complex features in the data. Through multi-level feature learning, the prediction accuracy of the snow cover area ratio can be greatly improved. In addition, the DBN can better cope with the changes under different geographical and climatic conditions, thereby enhancing the adaptability and robustness of the model. The DBN also has the ability to process high-dimensional data (such as the multi-spectral bands in remote sensing images), extracts key information using its deep network structure, and avoids the complexity brought by manual feature selection.

[0074] In this embodiment, the remote sensing image data of the research area is obtained by the geological resource hyperspectral satellite and preprocessed to obtain the remote sensing image data containing the surface true reflectance and spectral data. This process ensures the accuracy and availability of the data. Among them, the geological resource hyperspectral satellite has higher spatial resolution and rich spectral bands, and can provide a large amount of spectral information for optical inversion of ice and snow parameters.

[0075] Calculate the normalized snow index (NDSI) according to the preprocessed spectral data. This index is an important indicator reflecting the snow cover situation and helps to improve the accuracy of subsequent snow monitoring. At the same time, input the surface true reflectance and the auxiliary data of the research area into the preset snow water equivalent inversion model constructed based on the XGBoost machine learning module to obtain the snow water equivalent data.

[0076] During the calculation of the snow cover area ratio, auxiliary data, the normalized snow index, and remote sensing image data are input into a preset deep belief network (DBN) to accurately calculate the snow cover area ratio. This data is crucial for evaluating the distribution and changing trends of ice and snow resources. This study innovatively introduced a deep learning algorithm to fit the complex non-linear relationship between the snow cover area ratio and multiple auxiliary factors through the DBN model, thereby improving the inversion accuracy and regional adaptability.

[0077] Finally, by combining the normalized snow index, snow water equivalent data, and snow cover area ratio data, the ice and snow product inversion result is obtained, providing a scientific basis for water resource management and climate change research.

[0078] In summary, through the combination of satellite remote sensing technology and machine learning algorithms, this embodiment achieves high-precision ice and snow product inversion under complex terrain conditions. At the same time, based on the high-frequency observation ability of the geological resource hyperspectral satellite, this method can monitor snow cover changes in real time and provide timely information required for water resource management. This method comprehensively utilizes spectral data and auxiliary data to achieve efficient integration of information, enhancing the comprehensiveness and accuracy of snow cover monitoring.

[0079] In addition, the ice and snow products generated in this embodiment are characterized by a large range, high resolution, and high precision, meeting the monitoring requirements of complex terrains, diverse climate conditions, and different spatial scales. By using machine learning and deep learning algorithms, this method can effectively fit the complex non-linear relationship between hyperspectral data and various ice and snow parameters, giving full play to the advantages of geological resource hyperspectral satellite data. The adoption of advanced remote sensing technology and machine learning algorithms not only promotes the development of ice and snow monitoring technology but also opens up new directions for the research and application exploration of related technologies in the future.

[0080] Optionally, the ice and snow product inversion result further includes a snow cover extent data product, and the method for inverting ice and snow products based on a geological resource hyperspectral satellite further includes:

[0081] Inputting the spectral data, the auxiliary data, and the normalized snow index into a preset snow cover identification model to obtain a temporary snow cover extent data product; the preset snow cover identification model is constructed based on a multi-level decision tree;

[0082] Obtaining a microwave snow depth data set, and based on a spatio-temporal interpolation algorithm, fusing the microwave snow depth data set with the temporary snow cover extent data product to obtain the snow cover extent data product.

[0083] Optionally, the spectral data includes multiple band data, and the auxiliary data includes normalized difference vegetation index, brightness temperature data, and DEM data; inputting the spectral data, the auxiliary data, and the normalized snow index into a preset snow cover identification model to obtain a temporary snow cover extent data product includes:

[0084] Determine the first preset band data in the spectral data, where the first preset band data includes the seventh band data, the fourteenth band data, and the eighteenth band data;

[0085] Input the seventh band data, the fourteenth band data, the eighteenth band data, the normalized difference vegetation index, the brightness temperature data, the DEM data, and the normalized snow index into the preset snow cover identification model to obtain the temporary snow cover extent data product.

[0086] Specifically, as Figure 3 shown, the spectral data includes multiple bands, and the first preset band data is the seventh band, the fourteenth band, and the eighteenth band. These bands are closely related to the characteristics of ice and snow and help to effectively identify snow-covered areas.

[0087] Normalized difference vegetation index (NDVI): Used to evaluate vegetation coverage and help distinguish snow and ice areas from other surface features.

[0088] Brightness temperature data: The surface temperature obtained based on remote sensing images provides further understanding of the surface thermal state.

[0089] Digital elevation model (DEM) data: Provides topographic information and helps identify places where snow accumulates (such as ridges and depressions).

[0090] Determining the first preset band data in the spectral data is a key step in retrieving the snow cover extent. Selecting the data of the seventh band (620 - 650 nm), the fourteenth band (840 - 860 nm), and the eighteenth band (1580 - 1620 nm) is mainly because these bands are most sensitive to the reflection characteristics of ice and snow and can effectively separate snow cover from other surface features.

[0091] In this process, the decision tree realizes efficient identification of snow-covered areas by splitting and classifying the input features. The model can judge which areas are snow-covered and which are not.

[0092] Next, obtain the snow depth data set through microwave remote sensing technology. Microwave sensors are very sensitive to the depth and water content of snow cover, so they can provide accurate snow depth information. These data show good stability under different meteorological conditions.

[0093] Finally, to address the issue of partial gaps in remote sensing data caused by cloud cover or sensor anomalies, a microwave snow depth dataset is used to fill in the missing areas. Specifically, a spatio-temporal interpolation algorithm is employed to fuse the microwave snow depth dataset with the temporary snow cover extent data product. Microwave data has the advantages of penetrating clouds and operating under all-weather conditions. Through spatio-temporal interpolation techniques and data fusion algorithms, the gaps in the snow cover extent data (temporary snow cover extent data product) of the study area (such as the Tibetan Plateau) are filled, thus generating a snow cover extent data product of the Tibetan Plateau without gaps. Among them, the microwave snow depth dataset includes snow depth data, which can be obtained through, for example, the Science Data Center of the Tibetan Plateau.

[0094] This process involves synthesizing the microwave snow depth data obtained at a certain time and space position with the previously extracted temporary snow cover extent data according to the changes in time and space using an interpolation algorithm, so that the final snow cover extent data product has higher accuracy and consistency.

[0095] In some embodiments, the normalized difference vegetation index (NDVI) with a resolution of 500 meters is obtained through the MODIS satellite to reduce the interference of vegetation cover on snow recognition. At the same time, the brightness temperature data with a resolution of 5 kilometers is obtained through the BT11 band of the AVHRR satellite to assist in identifying the distribution of clouds and snow. In addition, the elevation data (DEM, with a resolution of 30 meters) provided by SRTM is used for terrain correction and regional characteristic analysis, thereby optimizing the spatial distribution accuracy of snow recognition.

[0096] The spectral data contains 26 bands, has a spatial resolution of 14 meters, and the spectral resolution ranges from 410 nanometers to 2480 nanometers. When obtaining the snow cover extent data product, Band7 (620 - 650 nanometers, the data of the seventh band), Band14 (840 - 860 nanometers, the data of the fourteenth band), and Band18 (1580 - 1620 nanometers, the data of the eighteenth band) are mainly used. These bands are selected because of their high sensitivity to snow. Among them, Band7 is located in the red light band and can effectively distinguish snow from other ground objects, especially outstanding under dry snow conditions; Band14 is located in the near-infrared band and is extremely sensitive to the reflection of snow, which can help accurately identify the snow cover range; Band18 belongs to the short-wave infrared band and is sensitive to the water content change in snow, and can effectively identify the state of snow. These bands provide rich spectral information for snow retrieval.

[0097] Compared with other satellites (such as MODIS, Landsat, etc.), the band combination of the geological resource hyperspectral satellite can better meet the sensitivity requirements for snow cover. In particular, the introduction of the short-wave infrared (SWIR) band makes the distinction between snow cover and other ground features such as water bodies and wetlands more precise. At the same time, the NDSI generated through inversion further improves the discrimination ability between snow cover and other land cover types.

[0098] Through the comprehensive processing of the above data by the snow cover recognition model, a temporary snow cover extent data product is obtained. For the unique terrain, climate conditions and snow cover distribution characteristics in complex areas such as the Qinghai-Tibet Plateau, the model utilizes the flexibility of the multi-level decision tree. Through the training of a large number of samples, the optimal threshold of the multi-level decision tree snow cover recognition algorithm is obtained, and a snow cover recognition algorithm (snow cover recognition model) suitable for the corresponding area is established.

[0099] Optionally, the spectral data includes multiple band data; obtaining the normalized snow index according to the spectral data includes:

[0100] Determine the second preset band data in the spectral data, where the second preset band data includes the fifth band data and the nineteenth band data;

[0101] Obtain the normalized snow index according to the fifth band data and the nineteenth band data.

[0102] Optionally, obtaining the normalized snow index according to the fifth band data and the nineteenth band data includes:

[0103] Obtain the normalized snow index according to the fifth band data and the nineteenth band data through Equation 1;

[0104] Equation 1 includes:

[0105]

[0106] where, NDSI is the normalized snow index, R DD-1-b5 is the fifth band data, R DD-1-b19 is the nineteenth band data.

[0107] Specifically, the normalized snow index (NDSI) is a common method for extracting snow cover by optical remote sensing at present and is widely used in the distinction between snow cover and clouds and surface backgrounds. NDSI is considered an ideal cloud-snow discriminator, and its specific formula is as follows:

[0108]

[0109] where, R VIS and R SWIRThey are the reflectances of the visible and shortwave infrared channels respectively. This index was initially proposed by Hall et al. When applied to EOS / MODIS, it corresponds to Channel 4 (0.545 - 0.565 μm) and Channel 6 (1.628 - 1.652 μm). In the present invention, based on the spectral response characteristics of the hyperspectral satellite images of geological resources and the analysis of the channel image quality, it is determined that Channel 5 (0.54 - 0.56 μm) and Channel 19 (1.63 - 1.67 μm) of the hyperspectral satellite for geological resources are respectively used as R VIS and R SWIR .

[0110] Compared with Channel 4 of MODIS, Channel 5 of the hyperspectral satellite for geological resources has a finer wavelength range and higher spatial resolution, which gives the hyperspectral satellite for geological resources obvious advantages in snow and ice inversion. The band of Channel 5 of the hyperspectral satellite for geological resources (the corresponding band is the fifth band data) can more accurately distinguish snow cover from other ground objects, while the shortwave infrared band of Channel 19 (the nineteenth band data) is more sensitive to the water content change in the snow cover and can provide a more accurate identification of the snow cover state. In addition, the optimization of the hyperspectral satellite image quality for geological resources also makes the identification of the snow cover boundary clearer, thus improving the accuracy of snow and ice inversion.

[0111] Optionally, the spectral data includes multiple band data, the auxiliary data includes DEM data and normalized vegetation index, and the remote sensing image data also includes geographical coordinates; the step of inputting the auxiliary data, the normalized snow index, and the remote sensing image data into a preset deep belief network to obtain the snow cover area ratio data includes:

[0112] Determining the third preset band data in the spectral data, where the third preset band data includes the fourteenth band data and the eighteenth band data;

[0113] Inputting the DEM data, the normalized vegetation index, the normalized snow index, the geographical coordinates, the fourteenth band data, and the eighteenth band data into the preset deep belief network to obtain the snow cover area ratio data.

[0114] Specifically, the fractional snow cover (FSC) is an important indicator for describing the snow cover extent, and its monitoring accuracy is significantly affected by multiple factors such as terrain, geographical location, surface temperature, and vegetation index. In complex terrain regions such as the Qinghai-Tibet Plateau, traditional FSC inversion methods often face problems of low accuracy and poor adaptability, especially in the transition and marginal zones between snow-covered and non-snow-covered areas, where it is difficult to provide accurate estimations. Therefore, in this embodiment, a deep learning method is introduced, and a deep belief network (DBN) model is used to fit the complex non-linear relationship between the fractional snow cover and multiple auxiliary factors, thereby improving the inversion accuracy and regional adaptability. As Figure 4 shown, it is the structure diagram of the preset deep belief network. Among them, RBM is the restricted Boltzmann machine layer and backpropagation (BP), and Input Layer, Hidden Layer, and Output Layer represent the input layer, hidden layer, and output layer respectively. Other letters in the middle (such as W0, W1, and W2) represent weight parameters.

[0115] DBN is a deep learning method based on a multi-layer neural network, usually composed of multiple restricted Boltzmann machine (RBM) layers and a backpropagation (BP) layer stacked together. Each RBM consists of a visible input layer and a hidden layer, and there are fully undirected connections between layers. The training of DBN includes two steps: unsupervised pre-training and supervised fine-tuning. During the pre-training process, the RBM maximizes the probability distribution of data reconstruction layer by layer to extract the latent features of the data; subsequently, through the BP algorithm, the entire network is supervised and fine-tuned to optimize the network parameters to better fit the relationship between the input and output.

[0116] On this basis, through the correlation analysis of relevant factors, the digital elevation model (DEM) of cloud-free pixels, the normalized difference snow index (NDSI), the normalized difference vegetation index (NDVI), longitude, latitude, and multiple bands (Band14 and Band18) of the hyperspectral satellite of geological resources are selected as the inputs of the model. At the same time, the fractional snow cover data (Landsat-FSC) provided by Landsat-8 OLI is used as the "true value of FSC (fractional snow cover)", that is, the dependent variable, for constructing the model. Based on the trained DBN model, high-resolution FSC distribution maps of complex regions such as the Qinghai-Tibet Plateau are generated using full-coverage hyperspectral satellite images and relevant auxiliary data.

[0117] Optionally, the auxiliary data includes snow pressure data and snow density data; the step of inputting the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data includes:

[0118] Input the true surface reflectivity, the snow pressure data, and the snow density data into the preset snow water equivalent inversion model to obtain the snow water equivalent data.

[0119] Specifically, the snow water equivalent is an important factor reflecting the change of surface snow cover, and is a key parameter in surface hydrological models and climate models. It is widely used in hydrology, climate, meteorology, flood forecasting and other fields, all of which require snow water equivalent information in the region.

[0120] In some embodiments, measured snow water equivalent data of meteorological stations in complex regions such as the Qinghai-Tibet Plateau are collected, as well as snow pressure and snow layer density auxiliary data of ERA-5. The inversion of the snow water equivalent requires obtaining the reflectivity information of meteorological stations on the remote sensing image of the current day according to the latitude and longitude coordinates, and at the same time obtaining the snow pressure and snow density data on the ERA-5 image. By pairing the remote sensing reflectivity data, snow pressure, snow density data with the measured snow water equivalent data, "auxiliary data - snow water equivalent" data pairs can be obtained (the data used includes: preprocessed reflectivity data, snow pressure and snow density data in the ERA-5 dataset, and the snow water equivalent as the target value, and these data can also be directly downloaded).

[0121] Since there is a complex non-linear relationship between reflectivity, snow pressure and snow layer density and snow water equivalent, the present invention is based on the XGBoost machine learning model for fitting, using reflectivity, snow pressure and snow layer density as independent variables, and the measured snow water equivalent of the station as the dependent variable. Applying the trained model to the inversion of the regional snow water equivalent, so as to obtain a complete and detailed regional snow water equivalent product, providing data support for the subsequent monitoring and management of regional water resources.

[0122] Optionally, the preprocessing of the remote sensing image data to obtain remote sensing image data includes:

[0123] Radiometrically calibrate the remote sensing image data to obtain processed remote sensing image data; and convert the processed remote sensing image data according to a preset calibration coefficient to obtain temporary image data;

[0124] Perform atmospheric correction on the temporary image data to obtain the remote sensing image data.

[0125] Specifically, remote sensing images are obtained by a geological resource hyperspectral satellite. The satellite is equipped with multi-spectral sensors and can obtain the optical signals reflected by the surface in multiple bands. These images contain digital quantization values (DN values) in different bands and reflect the spectral information of the surface and its features.

[0126] Radiometric calibration is the process of converting the digital quantification values (DN values) of the original image into physical quantities. The DN value is the digital representation of the optical signal received by the sensor, and radiometric calibration is required to convert these digital values into specific physical quantities, such as reflectance. The specific methods usually include using standardized reflectance values and calibration coefficients of the sensor to ensure the comparability of image data at different times and under different conditions.

[0127] Radiometric correction aims to eliminate the errors of the sensor itself and the atmospheric effects, making the image data closer to the true surface reflectance of the ground objects. This process may involve, for example, sensor error correction: adjusting the systematic deviation in the image according to the characteristics and parameters of the sensor. Atmospheric effect correction: Gases and particulate matters in the atmosphere will affect the signal intensity received by the sensor, generating scattering and absorption effects. Therefore, it is necessary to use an atmospheric correction model (such as MODTRAN or 6S model) to estimate and remove these effects, so as to improve the accuracy of the data.

[0128] Geometric correction is also included in the correction process. Geometric correction is the process of correcting the geometric distortion of the image to ensure that each pixel corresponds accurately to the ground position. For example, geographic coordinate registration: ensuring that each pixel in the image corresponds to the actual geographic location, which usually requires using geographic reference data for registration. Projection system conversion: converting the image data into a standard projection system to make it suitable for further analysis.

[0129] For the atmospheric correction process of the temporary image data, an atmospheric model (such as MODTRAN) and ground observation data can be used to simulate the influence of the atmosphere on the optical signal, and apply the calculated atmospheric influence to the image data to obtain the true surface reflectance.

[0130] Through these processing steps, the finally obtained image has accurate true surface reflectance and spectral data (data of each band), making it suitable for subsequent analysis and applications, such as snow water equivalent inversion, surface feature monitoring, etc. This series of correction processes is the basis for ensuring the reliability and accuracy of remote sensing data.

[0131] In some embodiments, the coefficient of determination (R2), root mean squared error (RMSE), and mean absolute error (MAE) are used to evaluate the result accuracy, robustness, and transferability of the model through ten-fold cross-validation.

[0132] Let y i represent the actual measured value at the site, represent the predicted value, represent the mean value, and n represent the number of samples.

[0133] 1) Root Mean Square Error (RMSE) is the square root of the ratio of the sum of the squares of the differences between the predicted values and the true values to the number of observations n. The smaller it is, the higher the prediction ability.

[0134]

[0135] 2) Coefficient of determination R 2 , which is used to represent the degree of correlation between the predicted values and the measured values. The closer it is to 1, the better the correlation degree.

[0136]

[0137] 3) Mean Absolute Error (MAE) is the average of the absolute values of the differences between the predicted values and the true values. The smaller it is, the better the prediction ability.

[0138]

[0139] As Figure 5 shown, an ice and snow product inversion device based on a geological resource hyperspectral satellite provided by an embodiment of the present invention includes:

[0140] An acquisition unit 510, configured to acquire remote sensing image data of a research area through the geological resource hyperspectral satellite, and preprocess the remote sensing image data to obtain remote sensing image data, where the remote sensing image data includes surface true reflectance and spectral data;

[0141] A processing unit 520, configured to obtain a normalized snow index according to the spectral data;

[0142] The acquisition unit 510 is further configured to acquire auxiliary data of the research area;

[0143] The processing unit 520 is further configured to input the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data, where the preset snow water equivalent inversion model is constructed based on an XGBoost machine learning module;

[0144] The processing unit 520 is further configured to input the auxiliary data, the normalized snow index, and the remote sensing image data into a preset deep belief network to obtain snow cover area ratio data; wherein, the ice and snow product inversion result includes the normalized snow index, the snow water equivalent data, and the snow cover area ratio data.

[0145] An ice and snow product inversion system based on a geological resource hyperspectral satellite provided by an embodiment of the present invention includes a memory and a processor; the memory is configured to store a computer program; the processor is configured to, when executing the computer program, implement the ice and snow product inversion method based on a geological resource hyperspectral satellite as described above.

[0146] A computer-readable storage medium provided by an embodiment of the present invention, on which a computer program is stored. When the computer program is executed by a processor, the above-described method for retrieving ice and snow products based on a hyperspectral satellite for geological resources is implemented.

[0147] Now, a system for retrieving ice and snow products based on a hyperspectral satellite for geological resources, which can be a server or a client of the present invention, will be described. It is an example of a hardware device that can be applied to various aspects of the present invention. The system for retrieving ice and snow products based on a hyperspectral satellite for geological resources is intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The system for retrieving ice and snow products based on a hyperspectral satellite for geological resources can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0148] Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will all fall within the protection scope of the present invention.

Claims

1. A method for retrieving ice and snow products based on a hyperspectral satellite of geological resources, characterized in that, Applied to the geological resource hyperspectral satellite, including: Obtain remote sensing image data of the research area through the geological resource hyperspectral satellite, and preprocess the remote sensing image data to obtain remote sensing image data, where the remote sensing image data includes surface true reflectance and spectral data; Obtain the Normalized Difference Snow Index according to the spectral data; Obtain the auxiliary data of the research area, and input the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data, and the preset snow water equivalent inversion model is constructed based on the XGBoost machine learning module; Input the auxiliary data, the Normalized Difference Snow Index, and the remote sensing image data into a preset deep belief network to obtain snow cover area ratio data; Among them, the ice and snow product inversion result includes the Normalized Difference Snow Index, the snow water equivalent data, and the snow cover area ratio data.

2. The ice and snow product inversion method based on the hyperspectral satellite of geological resources according to claim 1, wherein, The ice and snow product inversion result further includes a snow cover extent data product, and the method for inverting ice and snow products based on a geological resource hyperspectral satellite further includes: Input the spectral data, the auxiliary data, and the Normalized Difference Snow Index into a preset snow cover identification model to obtain a temporary snow cover extent data product; the preset snow cover identification model is constructed based on a multi-level decision tree; Obtain a microwave snow depth data set, and fuse the microwave snow depth data set with the temporary snow cover extent data product based on a spatio-temporal interpolation algorithm to obtain the snow cover extent data product.

3. The ice and snow product inversion method based on the hyperspectral satellite for geological resources according to claim 2, wherein The spectral data includes multiple band data, and the auxiliary data includes a Normalized Difference Vegetation Index, brightness temperature data, and DEM data; the step of inputting the spectral data, the auxiliary data, and the Normalized Difference Snow Index into a preset snow cover identification model to obtain a temporary snow cover extent data product includes: Determine the first preset band data in the spectral data, where the first preset band data includes the seventh band data, the fourteenth band data, and the eighteenth band data; Input the seventh band data, the fourteenth band data, the eighteenth band data, the Normalized Difference Vegetation Index, the brightness temperature data, the DEM data, and the Normalized Difference Snow Index into the preset snow cover identification model to obtain the temporary snow cover extent data product.

4. The method for retrieving ice and snow products based on a hyperspectral satellite of geological resources according to claim 1, wherein The spectral data includes multiple band data; the step of obtaining the Normalized Difference Snow Index according to the spectral data includes: Determine the second preset band data in the spectral data, where the second preset band data includes the fifth band data and the nineteenth band data; Obtain the Normalized Difference Snow Index according to the fifth band data and the nineteenth band data.

5. The ice and snow product inversion method based on the hyperspectral satellite for geological resources according to claim 4, wherein The step of obtaining the Normalized Difference Snow Index according to the fifth band data and the nineteenth band data includes: Obtain the Normalized Difference Snow Index according to the fifth band data and the nineteenth band data through Equation (1); Equation (1) includes: Among them, NDSI is the normalized snow index, R DD-1-b5 is the fifth band data, and R DD-1-b19 is the nineteenth band data.

6. The ice and snow product inversion method based on the hyperspectral satellite for geological resources according to claim 1, characterized in that, The spectral data includes multiple band data, the auxiliary data includes DEM data and normalized difference vegetation index, and the remote sensing image data further includes geographic coordinates; the step of inputting the auxiliary data, the normalized snow index, and the remote sensing image data into a preset deep belief network to obtain snow cover area ratio data includes: Determine the third preset band data in the spectral data, where the third preset band data includes the fourteenth band data and the eighteenth band data; Input the DEM data, the normalized difference vegetation index, the normalized snow index, the geographic coordinates, the fourteenth band data, and the eighteenth band data into the preset deep belief network to obtain the snow cover area ratio data.

7. The ice and snow product inversion method based on a hyperspectral satellite for geological resources according to claim 1, wherein The auxiliary data includes snow pressure data and snow density data; the step of inputting the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data includes: Input the surface true reflectance, the snow pressure data, and the snow density data into the preset snow water equivalent inversion model to obtain the snow water equivalent data.

8. The ice and snow product inversion method based on the hyperspectral satellite for geological resources according to claim 1, characterized in that The step of preprocessing the remote sensing image data to obtain remote sensing image data includes: Perform radiometric calibration on the remote sensing image data to obtain processed remote sensing image data; and perform conversion on the processed remote sensing image data according to a preset calibration coefficient to obtain temporary image data; Perform atmospheric correction on the temporary image data to obtain the remote sensing image data.

9. An ice and snow product inversion device based on a geological resource hyperspectral satellite, characterized in that, Applied to the geological resource hyperspectral satellite, it includes: An acquisition unit for acquiring remote sensing image data of a study area through the geological resource hyperspectral satellite and preprocessing the remote sensing image data to obtain remote sensing image data, where the remote sensing image data includes surface true reflectance and spectral data; A processing unit for obtaining a normalized snow index according to the spectral data; The acquisition unit is further configured to acquire auxiliary data of the study area; The processing unit is further configured to input the surface true reflectance and the auxiliary data into a preset snow water equivalent inversion model to obtain snow water equivalent data, and the preset snow water equivalent inversion model is constructed based on an XGBoost machine learning module; The processing unit is further configured to input the auxiliary data, the normalized snow index, and the remote sensing image data into a preset deep belief network to obtain snow cover area ratio data; wherein, the ice and snow product inversion result includes the normalized snow index, the snow water equivalent data, and the snow cover area ratio data.

10. An ice and snow product inversion system based on a hyperspectral satellite for geological resources, characterized in that, It includes a memory and a processor; the memory is used for storing a computer program; the processor is configured to, when executing the computer program, implement the ice and snow product inversion method based on the geological resource hyperspectral satellite according to any one of claims 1 to 8.

Images