A method and device for generating high-precision geochemical data based on remote sensing images

Abstract

The present application provides a method and device for generating high-precision geochemical data based on remote sensing images. The method for generating high-precision geochemical data can fuse small-scale geochemical data and hyperspectral remote sensing image data in the entire region through a trained geochemical data generation model, thereby eliminating the need to collect large-scale geochemical data in the entire region, and can also generate large-scale geochemical data that can cover the entire region. Furthermore, on the basis of reducing data collection costs and improving data collection efficiency, geological survey personnel can simultaneously obtain geochemical data of multiple different scales corresponding to the entire region, so as to meet the geological survey personnel's higher-precision work task requirements.

Description

Technical Field

[0001] The present application relates to the field of geological research technology, and in particular to a method and device for generating high-precision geochemical data based on remote sensing images. Background Art

[0002] The collection of geochemical data through regional geochemical exploration is a major method for delineating prospective mineralization areas. Its purpose is to further carry out regional mineral surveys and detailed investigations. The scale of geochemical data reflects the level of detail of geological surveys and the accuracy of the data. The higher the data accuracy of large-scale geochemical data, the more detailed the element enrichment information it contains. Therefore, when determining the prospective mineralization areas within the study area, it is extremely important to obtain large-scale geochemical data for the entire study area.

[0003] At present, according to different work stages or task nature, traditional large-scale geochemical data are mainly obtained by setting up more dense sampling points in the study area, replacing the sampling medium, and testing and analyzing the collected samples. However, on the one hand, since large-scale geochemical data requires more dense sampling points to be set up in the field, the data collection cost of large-scale geochemical data is high and the data collection efficiency is low; on the other hand, due to the limitations of the work stage or task nature, after sampling the study area according to the sampling density specified in a work stage or a task nature, since the sampling density has been limited, only geochemical data of a single scale matching the above-specified sampling density can be obtained, which often cannot meet the needs of higher-precision work tasks. Summary of the invention

[0004] In view of this, the present application provides a method and device for generating high-precision geochemical data based on remote sensing images. Through a trained geochemical data generation model, small-scale geochemical data and hyperspectral remote sensing image data in the entire area can be integrated, thereby eliminating the need to collect large-scale geochemical data in the entire area, and generating large-scale geochemical data that can cover the entire area. This can reduce data collection costs and improve data collection efficiency, and also enable geological survey personnel to simultaneously obtain geochemical data of multiple different scales corresponding to the entire area, so as to meet the higher-precision work task requirements of geological survey personnel.

[0005] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings.

[0006] In a first aspect, an embodiment of the present application provides a method for generating high-precision geochemical data based on remote sensing images, the method comprising:

[0007] Obtaining first geochemical data, hyperspectral remote sensing image data and second geochemical data of a local area in the study area; wherein the scale of the first geochemical data is smaller than the scale of the second geochemical data;

[0008] Inputting the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the original model, fusing the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area through the original model to generate the second geochemical data for prediction, and outputting the prediction result for the second geochemical data;

[0009] According to the prediction loss between the prediction result and the second geochemical data, the model parameters of the original model are adjusted to obtain the original model including the adjusted model parameters as a geochemical data generation model;

[0010] The first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area are input into the geochemical data generation model, and the fusion result of the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area is output as the second geochemical data of the entire study area.

[0011] In a second aspect, an embodiment of the present application provides a high-precision geochemical data generation device based on remote sensing images, the high-precision geochemical data generation device comprising:

[0012] A data acquisition module is used to obtain first geochemical data, hyperspectral remote sensing image data and second geochemical data of a local area in the study area; wherein the scale of the first geochemical data is smaller than the scale of the second geochemical data;

[0013] A prediction module, used for inputting the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the original model, fusing the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area through the original model to generate the second geochemical data for prediction, and outputting the prediction result for the second geochemical data;

[0014] A parameter adjustment module, used for adjusting the model parameters of the original model according to the prediction loss between the prediction result and the second geochemical data, and obtaining the original model including the adjusted model parameters as a geochemical data generation model;

[0015] The data generation module is used to input the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area into the geochemical data generation model, and output the fusion result of the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area as the second geochemical data of the entire study area.

[0016] In a third aspect, an embodiment of the present application provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for generating high-precision geochemical data based on remote sensing images when executing the computer program.

[0017] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of the above-mentioned method for generating high-precision geochemical data based on remote sensing images are executed.

[0018] The technical solution provided by the embodiments of the present application may have the following beneficial effects:

[0019] The embodiment of the present application provides a method and device for generating high-precision geochemical data based on remote sensing images. Through a trained geochemical data generation model, it is possible to fuse small-scale geochemical data and hyperspectral remote sensing image data in the entire area, thereby eliminating the need to collect large-scale geochemical data in the entire area. Large-scale geochemical data covering the entire area can also be generated. Furthermore, on the basis of reducing data collection costs and improving data collection efficiency, geological survey personnel can simultaneously obtain geochemical data of multiple different scales corresponding to the entire area, so as to meet the higher-precision work task requirements of geological survey personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0021] Figure 1 A schematic diagram of a process for generating high-precision geochemical data based on remote sensing images provided in an embodiment of the present application is shown;

[0022] Figure 2A schematic diagram of a flow chart of a model training method of an original model provided in an embodiment of the present application is shown;

[0023] Figure 3 A schematic diagram of the structure of a high-precision geochemical data generation device based on remote sensing images provided in an embodiment of the present application is shown;

[0024] Figure 4 A schematic diagram of the structure of an electronic device 400 provided in an embodiment of the present application. DETAILED DESCRIPTION

[0025] To make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. It should be understood that the drawings in the present application only serve the purpose of explanation and description and are not used to limit the scope of protection of the present application. In addition, it should be understood that the schematic drawings are not drawn in real proportion. The flowchart used in this application shows the operations implemented according to some embodiments of the present application. It should be understood that the operations of the flowchart can be implemented out of sequence, and the steps without logical context can be reversed in order or implemented simultaneously. In addition, those skilled in the art can add one or more other operations to the flowchart under the guidance of the content of the present application, or remove one or more operations from the flowchart.

[0026] In addition, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various configurations. Therefore, the following detailed description of the embodiments of the present application provided in the drawings is not intended to limit the scope of the application claimed for protection, but merely represents the selected embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present application.

[0027] It should be noted that the term "comprising" will be used in the embodiments of the present application to indicate the existence of the features declared thereafter, but does not exclude the addition of other features.

[0028] In one of the embodiments of the present application, a high-precision geochemical data generation method based on remote sensing images can be applied to a terminal device or a server; wherein the terminal device can be a local terminal device, and when the high-precision geochemical data generation method based on remote sensing images is run on the server, the high-precision geochemical data generation method based on remote sensing images can be implemented and executed based on a cloud interaction system, wherein the cloud interaction system includes a server and a client device (ie, a terminal device); the embodiment of the present application does not impose any limitation on the specific device type of the application device of the high-precision geochemical data generation method based on remote sensing images.

[0029] To facilitate understanding of the embodiments of the present application, a method and device for generating high-precision geochemical data based on remote sensing images provided in the embodiments of the present application are introduced in detail below.

[0030] Reference Figure 1 As shown, Figure 1 A schematic flow chart of a method for generating high-precision geochemical data based on remote sensing images provided in an embodiment of the present application is shown, wherein the method for generating high-precision geochemical data includes steps S101-S104; specifically:

[0031] S101, obtaining first geochemical data, hyperspectral remote sensing image data and second geochemical data of a local area in a study area.

[0032] Here, all the data obtained in step S101 are training sample data used for subsequent model training; wherein, the scale of the first geochemical data in step S101 is smaller than the scale of the second geochemical data, that is, the first geochemical data and the second geochemical data belong to geochemical data of the same area (i.e., a local area within the study area), and due to the different scales, the data accuracy of the first geochemical data will be lower than the data accuracy of the second geochemical data (i.e., the larger the scale, the higher the data accuracy).

[0033] Specifically, compared to the geochemical data obtained by artificially setting up data sampling points in the field for data sampling and analysis, satellite remote sensing images have the characteristics of wide coverage and easy collection. Therefore, in the embodiment of the present application, the above-mentioned hyperspectral remote sensing image data of the local area in the study area can be easily obtained from the satellite remote sensing data of the study area. By training the model, it is learned how to fuse the low-precision small-scale geochemical data (i.e., the above-mentioned first geochemical data) and the easy-to-collect hyperspectral remote sensing image data to generate high-precision large-scale geochemical data (i.e., the above-mentioned second geochemical data), so as to reduce the data collection cost and improve the data collection efficiency.

[0034] It should be noted that, considering that the larger the scale, the higher the density of data sampling points that need to be deployed, the corresponding data collection cost is also higher, which leads to the higher difficulty of data collection corresponding to the larger the scale. Therefore, in order to improve data collection efficiency and reduce data collection costs, in the embodiment of the present application, there is no need to collect large-scale geochemical data for the entire study area. It is only necessary to collect large-scale geochemical data for local areas within the study area to obtain large-scale geochemical data of the local area (that is, the second geochemical data of the local area mentioned above) as high-precision geochemical data in the model training data.

[0035] Specifically, considering that the data collection difficulty of low-precision (i.e., small-scale) geochemical data and hyperspectral remote sensing image data is relatively low (at least lower than the data collection difficulty of the above-mentioned large-scale geochemical data), and in order to subsequently generate high-precision geochemical data that can cover the entire study area (i.e., the second geochemical data of the entire study area), it is also necessary to obtain low-precision geochemical data of the entire study area (i.e., the first geochemical data of the entire study area) and hyperspectral remote sensing image data of the entire study area. Therefore, in an embodiment of the present application, as an optional embodiment, the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area can be first obtained by data collection, and then the first geochemical data of the above-mentioned local area and the hyperspectral remote sensing image data of the above-mentioned local area that are consistent with the position coordinates (also equivalent to the pixel point coordinates of the image data) of the high-precision geochemical data of the above-mentioned local area (i.e., the second geochemical data of the above-mentioned local area) can be obtained by data clipping.

[0036] It should be noted that the specific area size and specific area location represented by the above-mentioned local area in the study area can be determined according to actual geological survey needs, and the embodiments of the present application do not impose any limitation on this.

[0037] S102, inputting the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the original model, fusing the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area through the original model to generate the second geochemical data for prediction, and outputting the prediction result for the second geochemical data.

[0038] Here, the above original model can specifically be a convolutional neural network model that introduces geographically weighted regression.

[0039] Specifically, in the traditional convolutional neural network model, the weight is generally determined according to the feature probability distribution, while in the above-mentioned original model provided in the embodiment of the present application, GWR (Geographically Weighted Regression) is introduced to adjust the weight coefficient of image fusion (that is, the weight coefficients corresponding to the first geochemical data of the above-mentioned local area and the hyperspectral remote sensing image data of the above-mentioned local area when they are fused), so that the fused data after the first geochemical data of the above-mentioned local area and the hyperspectral remote sensing image data of the above-mentioned local area are fused based on the above-mentioned weights as the prediction result of the second geochemical data of the above-mentioned local area.

[0040] S103, adjusting the model parameters of the original model according to the prediction loss between the prediction result and the second geochemical data, and obtaining the original model including the adjusted model parameters as a geochemical data generation model.

[0041] Specifically, in combination with the relevant instructions on the model structure of the original model in the above step S102, it can be known that the above prediction result represents the fused data obtained by the original model after fusing the first geochemical data of the above local area and the hyperspectral remote sensing image data of the above local area based on their respective corresponding weight coefficients, and the above second geochemical data is the real high-precision geochemical data corresponding to the above local area obtained by data sampling. Therefore, the weight coefficients (i.e., the model parameters of the original model) assigned by the original model to the first geochemical data of the above local area and the hyperspectral remote sensing image data of the above local area when performing data fusion can be adjusted through the prediction loss between the above prediction result and the above second geochemical data, until the original model converges (i.e., the above prediction loss reaches the minimum), and the original model that has reached convergence is used as the trained geochemical data generation model (i.e., the original model including the adjusted model parameters).

[0042] S104, inputting the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area into the geochemical data generation model, and outputting a fusion result of the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area as the second geochemical data of the entire study area.

[0043] Here, in order to obtain high-precision geochemical data that can cover the entire study area (i.e., the second geochemical data of the entire study area), in an embodiment of the present application, the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area can also be input into the trained above-mentioned geochemical data generation model, and the first geochemical data of the entire area and the hyperspectral remote sensing image data of the entire area are fused through the geochemical data generation model to obtain the fusion result as the second geochemical data of the entire study area (which is also equivalent to the data prediction result of the second geochemical data of the entire study area). Therefore, there is no need to collect large-scale geochemical data for the entire area, and large-scale geochemical data covering the entire area (i.e., the second geochemical data of the entire area) can be generated. Therefore, on the basis of reducing data collection costs and improving data collection efficiency, geological survey personnel can also obtain geochemical data of multiple different scales corresponding to the entire area at the same time (i.e., the first geochemical data of the entire area and the second geochemical data of the entire area), so as to meet the higher precision work task requirements of geological survey personnel.

[0044] The following is an exemplary description of each step in the method for generating high-precision geochemical data based on remote sensing images provided in the embodiment of the present application:

[0045] Regarding the specific implementation process of the above step S101, in the embodiment of the present application, as an optional embodiment, the first geochemical data of the above local area can be obtained according to the method shown in the following steps a1-a3, specifically:

[0046] Step a1: sampling the first sampling medium in the study area according to the sampling point density corresponding to the scale of the first geochemical data to obtain first sampling data, and analyzing the element content of the first sampling data to obtain element concentration distribution data for the entire study area.

[0047] Here, referring to the relevant content of the above step S101, it can be known that since the data collection difficulty of low-precision geochemical data (i.e., the first geochemical data) is relatively low and the low-precision geochemical data of the entire study area (i.e., the first geochemical data of the entire area) is required in the subsequent process, in the embodiment of the present application, the first geochemical data of the entire study area can be first collected through data collection, and then the first geochemical data of the above-mentioned local area can be obtained by data clipping.

[0048] Specifically, when collecting data, the specific sampling point density corresponding to different scales (including but not limited to: sampling unit area, specific number of sampling points arranged in the sampling unit, specific number of sampling points arranged per square kilometer, etc.), as well as the specific sampling media that can be selected according to the different work stages or task nature of geological surveys, can be referred to as shown in Table 1 below (wherein the content of Table 1 refers to the "Geological Prospecting Manual (Trial)" issued by the State Geological Administration in 1978):

[0049]

[0050] Table 1

[0051] For example, in the present embodiment, the first sampling medium may be stream sediments, and the scale of the first geochemical data may be 1:250,000. That is, when executing step a1, the study area may be divided into a plurality of sampling units with a side length of 2.5 km (i.e., a sampling unit area of ​​6.25 km). 2 ), 1 or 2 sampling points were arranged in each sampling unit (refer to the data scale shown in Table 1, the sampling unit area is 6.25km 2 Between 25km 2 and 4km 2 ), sampling the stream sediments in each sampling unit to obtain the above-mentioned first sampling data.

[0052] It should be noted that the specific sampling medium represented by the first sampling medium and the specific scale corresponding to the first geochemical data can be designed according to actual geological survey requirements, and the embodiments of the present application do not impose any limitation on this.

[0053] Here, after obtaining the above-mentioned first sampling data, the first sampling data can be analyzed for element content by various sampling and analysis methods such as X-ray fluorescence spectroscopy, plasma powder spectroscopy, emission spectroscopy, polarography, etc. The embodiment of the present application does not impose any limitation on the specific sampling and analysis method used in the element content analysis in step a1.

[0054] It should be noted that different types of sampling and analysis methods are aimed at different types of elements to be analyzed. In practical applications, the appropriate sampling and analysis method can be selected based on the specific element types that need to be focused on in the geological survey.

[0055] It should be noted that the above-mentioned element concentration distribution data can cover the concentration distribution data of various major, minor, trace elements and oxide elements including Au, Cu, As, Sb, Hg, etc. in the entire study area. The embodiment of this application does not impose any limitation on the specific element types and element quantities contained in the above-mentioned element concentration distribution data.

[0056] Step a2: convert the data format of the element concentration distribution data of the entire area into a raster data format that matches the hyperspectral remote sensing image data, and adjust the coordinate system of the element concentration distribution data of the entire area after the format conversion to a projection coordinate system that matches the hyperspectral remote sensing image data to obtain the element concentration distribution data of the entire area to be cropped.

[0057] Here, since the data format of the original geochemical data (i.e., the element concentration distribution data of the above-mentioned entire area) is in point data format, and the data format of the remote sensing image data is in raster data format, in order to adapt to the hyperspectral remote sensing image data and facilitate the subsequent model to perform image fusion on it, it is necessary to convert the data format of the element concentration distribution data of the above-mentioned entire area into a raster data format that matches the hyperspectral remote sensing image data; for example, the pixel size of the element concentration distribution data of the above-mentioned entire area with a scale of 1:250,000 can be set to 2.5km×2.5km, so as to complete the rasterization processing of the element concentration distribution data of the above-mentioned entire area.

[0058] It should be noted that, in order to reduce data errors, the element concentration distribution data of the entire area converted into a raster data format may be interpolated, wherein the interpolation interval may be the same as the pixel size.

[0059] Specifically, since the coordinate system of the original geochemical data obtained by data sampling (i.e., the element concentration distribution data of the above-mentioned entire area) is the geographic coordinate system, and the coordinate system of the hyperspectral remote sensing image data is the projection coordinate system, after obtaining the element concentration distribution data of the above-mentioned entire area after format conversion, it is also necessary to convert the geographic coordinate system and the projection coordinate system to make it consistent with the coordinate system of the hyperspectral remote sensing image data corresponding to the study area.

[0060] Step a3: clipping the element concentration distribution data of the entire area to be clipped according to the position coordinates corresponding to the local area in the projection coordinate system to obtain the first geochemical data of the local area.

[0061] Here, referring to the relevant content in the above step S101, it can be seen that since the data collection of high-precision geochemical data (i.e., the second geochemical data) is relatively difficult and the purpose of model training is to generate high-precision geochemical data for the entire study area, only the high-precision geochemical data of the local area in the study area (i.e., the second geochemical data of the local area) is used when making the model training data.

[0062] Based on this, in order to make the low-precision geochemical data in the model training data consistent with the position coordinates (i.e., pixel coordinates) of the high-precision geochemical data of the above-mentioned local area, the first geochemical data of the above-mentioned local area can also be obtained by cropping from the element concentration distribution data of the entire area.

[0063] Regarding the specific implementation process of the above step S101, in the embodiment of the present application, as an optional embodiment, the second geochemical data of the above local area can be obtained according to the method shown in the following steps b1-b2, specifically:

[0064] Step b1, sampling the second sampling medium in the local area according to the sampling point density corresponding to the scale of the second geochemical data to obtain second sampling data, and obtaining element concentration distribution data of the local area by performing element content analysis on the second sampling data.

[0065] Here, it is only necessary to ensure that the scale of the second geochemical data is greater than that of the first geochemical data. For example, if the scale of the first geochemical data is 1:250,000, the scale of the second geochemical data can be set to 1:50,000 or 1:10,000. The specific scale corresponding to the second geochemical data is not limited in any way in the embodiments of the present application.

[0066] Here, the second sampling medium may be the same as or different from the first sampling medium. For example, if the first sampling medium is water sediment, the second sampling medium may be soil different from it, etc. This embodiment of the application does not impose any limitation on this.

[0067] Specifically, the specific execution method of data sampling and sampling analysis in step b1 can refer to the content of the above step a1. It is only necessary to reduce the layout area of ​​the sampling unit from the entire study area to a local area within the study area, and adjust the specific sampling point density and the specific sampling medium according to the scale of the second geochemical data. The repetitive parts will not be repeated here.

[0068] For example, in the embodiment of the present application, the second sampling medium may be soil, and the scale of the second geochemical data may be 1:50,000. That is, when executing step b1, the local area in the study area may be divided into a plurality of sampling units with a side length of 500 m (i.e., a sampling unit area of ​​0.25 km2). 2 ), 1 or 2 sampling points were arranged in each sampling unit (refer to the data scale shown in Table 1, the sampling unit area is 0.25km 2 Corresponding to a scale of 1:50,000), the soil in each sampling unit is sampled to obtain the above-mentioned second sampling data.

[0069] Step b2: convert the data format of the element concentration distribution data of the local area into a raster data format that matches the hyperspectral remote sensing image data, and adjust the coordinate system of the element concentration distribution data of the local area after the format conversion to a projection coordinate system that matches the hyperspectral remote sensing image data to obtain the second geochemical data of the local area.

[0070] Specifically, the specific execution method of step b2 can refer to the content of the above step a2, except that the specific pixel size set during the rasterization processing needs to be adjusted according to different scales. For example, the pixel size of the element concentration distribution data of the above local area with a scale of 1:50,000 can be set to 500m×500m, so as to complete the rasterization processing of the element concentration distribution data of the above local area. The rest of the repetitions will not be repeated here.

[0071] Regarding the specific implementation process of the above step S101, in the embodiment of the present application, as an optional embodiment, the hyperspectral remote sensing image data of the above local area can be obtained according to the method shown in the following steps c1-c2, specifically:

[0072] Step c1: selecting target hyperspectral remote sensing image data corresponding to multiple bands within the target band range from the hyperspectral remote sensing image data of the entire study area.

[0073] Here, the above-mentioned target band range represents the band range with obvious absorption and reflection characteristics for altered minerals (which is conducive to more concentrated mineralization characteristics); among them, for hyperspectral remote sensing images, the near-infrared band range of 499-997nm and the short-wave infrared band range of 2014-2451nm have obvious absorption of altered minerals. Therefore, as an optional embodiment, the near-infrared band range of 499-997nm and the short-wave infrared band range of 2014-2451nm can be set as the above-mentioned target band range.

[0074] Specifically, as an optional embodiment, target hyperspectral remote sensing image data corresponding to multiple bands within the target band range may be directly screened from the hyperspectral remote sensing image data of the entire region.

[0075] Specifically, as another optional embodiment, in order to reduce the burden of data processing, it is also possible to first screen the hyperspectral remote sensing image data of the entire area, in the near infrared band 13-71 (499-996nm) and the short wave infrared band 147-150 (2182-2233nm), and then perform principal component analysis on the screened hyperspectral remote sensing image data of the multiple bands, and screen out the top 50 bands that can cover 99% of the information, and then calculate the correlation coefficients of the above screened bands with the interpolation elements such as Cu and Au at 1:250,000 (low-precision geochemical data), and comprehensively select the hyperspectral remote sensing image data of the top 10 bands with higher correlation coefficients as the above target hyperspectral remote sensing image data for subsequent model training.

[0076] Step c2: clipping the target hyperspectral remote sensing image data according to the position coordinates corresponding to the local area in the projection coordinate system to obtain the hyperspectral remote sensing image data of the local area.

[0077] Specifically, the specific execution method of step c2 can refer to the content of the above step a3, and the repeated parts will not be repeated here.

[0078] With respect to the specific model training process of the original model in the above steps S102-S103, in the embodiment of the present application, as an optional embodiment, the specific model structure of the above original model can be split into: an input layer, multiple convolutional layers, a connection layer, a geographically weighted layer, and a fusion layer; wherein, in combination with the specific model structure after the splitting, in an optional implementation manner, Figure 2 A schematic diagram of a flow chart of a model training method of an original model provided in an embodiment of the present application is shown, such as Figure 2 As shown, when executing step S102, the method includes steps S201-S205; specifically:

[0079] S201, inputting the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the input layer, adjusting the resolution of the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area to be the same as the second geochemical data of the local area through the input layer, and inputting the adjusted data into the multiple convolutional layers.

[0080] Here, referring to the contents described in the above steps a1-a3, b1-b2, and c1-c2, the first geochemical data, hyperspectral remote sensing image data, and second geochemical data of the above local area have been adjusted to qualified input data (i.e., input data that meets the model training conditions) with coordinate system one (based on the projection coordinate system corresponding to the hyperspectral remote sensing image data), unified data format (based on the raster data format corresponding to the hyperspectral remote sensing image data), and unified pixel point positions (all corresponding to the local areas within the study area). Therefore, in step S201, the above three different types of data can be directly input into the input layer of the original model.

[0081] Specifically, the input layer will downsample and upsample the above three different types of input data respectively, and unify the resolution of the input data to the resolution of the high-precision geochemical data (i.e., the above-mentioned second geochemical data), that is, the resolution of the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area is adjusted to be the same as the second geochemical data of the local area. After completing the above adjustment steps, the input layer can input the adjusted data (i.e., the adjusted first geochemical data of the local area, the hyperspectral remote sensing image data of the local area, and the second geochemical data of the local area) into subsequent multiple convolutional layers.

[0082] It should be noted that, as another optional embodiment, the clipping steps in the above step a3 and the above step c2 can also be performed in the input layer (that is, the pixel positions in the input layer are uniformly adjusted to be the same as the high-precision second geochemical data of the above local area), and the repetitions will not be repeated here.

[0083] S202, performing multi-scale feature extraction on the received first geochemical data, hyperspectral remote sensing image data and second geochemical data of the local area through the multiple convolutional layers, and inputting the obtained multi-scale feature maps corresponding to the first geochemical data, hyperspectral remote sensing image data and second geochemical data of the local area into the connection layer.

[0084] Here, each convolution layer corresponds to a feature extraction scale, and different convolution layers correspond to different feature extraction scales. The embodiment of the present application does not impose any limitation on the specific number of the above-mentioned multiple convolution layers.

[0085] Specifically, taking the case where the multiple convolutional layers are three convolutional layers as an example, multi-scale feature extraction is performed on the received first geochemical data of the local area through the three convolutional layers, so that multi-scale feature maps at three different scales corresponding to the first geochemical data of the local area can be obtained.

[0086] Specifically, taking the case where the multiple convolutional layers are three convolutional layers as an example, multi-scale feature extraction is performed on the received hyperspectral remote sensing image data of the above-mentioned local area through the three convolutional layers, and multi-scale feature maps at three different scales corresponding to the hyperspectral remote sensing image data of the above-mentioned local area can be obtained.

[0087] Specifically, taking the case where the multiple convolutional layers are three convolutional layers as an example, multi-scale feature extraction is performed on the received second geochemical data of the local area through the three convolutional layers, so that multi-scale feature maps at three different scales corresponding to the second geochemical data of the local area can be obtained.

[0088] S203, respectively fusing the multi-scale feature maps corresponding to the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area through the connection layer to obtain a first feature map corresponding to the first geochemical data of the local area, a second feature map corresponding to the hyperspectral remote sensing image data of the local area and a third feature map corresponding to the second geochemical data of the local area, and inputting the first feature map, the second feature map and the third feature map into the geographical weighted layer.

[0089] Specifically, still taking the example that the above-mentioned multiple convolutional layers are 3 convolutional layers, in the connection layer, the multi-scale feature maps at 3 different scales corresponding to the first geochemical data of the above-mentioned local area can be spliced ​​to obtain the splicing result as the first feature map corresponding to the first geochemical data of the above-mentioned local area.

[0090] Specifically, still taking the example that the above-mentioned multiple convolutional layers are 3 convolutional layers, in the connection layer, the multi-scale feature maps at 3 different scales corresponding to the hyperspectral remote sensing image data of the above-mentioned local area can be spliced ​​to obtain the splicing result as the second feature map corresponding to the hyperspectral remote sensing image data of the above-mentioned local area.

[0091] Specifically, still taking the example that the above-mentioned multiple convolutional layers are 3 convolutional layers, in the connection layer, the multi-scale feature maps at 3 different scales corresponding to the second geochemical data of the above-mentioned local area can be spliced ​​to obtain the splicing result as the third feature map corresponding to the second geochemical data of the above-mentioned local area.

[0092] S204, in the geographic weighted layer, using the first feature map and the second feature map as independent variables, and the third feature map as the dependent variable, for each pixel point, according to the independent variable value corresponding to the pixel point in each of the independent variables and the geographic weighting coefficient, calculate the prediction result of the dependent variable value corresponding to the pixel point.

[0093] In the embodiment of the present application, since both the hyperspectral remote sensing image data and the geochemical data have spatial heterogeneity, that is, the degree of influence on the variables is different at different locations, the introduction of geographically weighted regression into the fusion model of multi-source data can further improve the spatial details of the data; wherein, the GWR model is an extension of the linear regression model, and the spatial position of the data is introduced into the regression parameters. The regression model is shown in the following formula 1:

[0094]

[0095] Among them, (u i ,v i ) is the coordinate of the i-th sampling point, such as longitude and latitude;

[0096] β k (u i ,v i ) represents the kth regression coefficient of the ith sampling point, which is a function representing the spatial position, is independently distributed and obeys a normal distribution with a mean of 0 and a variance of σ;

[0097] p represents the number of independent variables, and in the embodiment of the present application, it can be the number of bands corresponding to the hyperspectral remote sensing image data.

[0098] Since matrix operations are required to fuse hyperspectral remote sensing image data and geochemical data, the above formula 1 can be rewritten into a matrix form, as shown in the following formula 2:

[0099] y=(X×β′)I+ε

[0100] Formula 2;

[0101] Among them, X and β′ are both n×(p+1)-dimensional matrices, I is a 1×(p+1)-dimensional unit matrix, X consists of n groups of sample point data, and β consists of n groups of regression coefficients, as shown in the following formula 3:

[0102]

[0103] Among them, for the determination of the regression coefficient in formula 1, according to the first law of geography, the WLS (Weighted Least Squares) method can be used to estimate the regression coefficient, and the regression coefficient Q of the i-th sampling point can be determined by the following formula 4:

[0104]

[0105] Among them, W ij Represents the function of the distance between the i-th sampling point and other sample points j. The weight matrix W is composed of W ij Composition, as shown in Formula 5 below:

[0106]

[0107] According to the above formula 3 and formula 5, the estimated value of the regression coefficient at the i-th sampling point can be obtained: As shown in Formula 6 below:

[0108]

[0109] Based on this, we can further get the estimated value at the i-th sampling point As shown in Formula 7 below:

[0110]

[0111] Among them, according to the above method, regression analysis is performed point by point, and a matrix containing the parameter estimation values ​​of each point can be obtained Therefore, the estimated value of each sampling point As shown in the following formula 8:

[0112]

[0113] In the above-mentioned geographically weighted layer of the embodiment of the present application, the dependent variable is the above-mentioned third characteristic map corresponding to the high-precision geochemical data (i.e., the second geochemical data), and the independent variable is the above-mentioned second characteristic map corresponding to the hyperspectral remote sensing image data and the above-mentioned first characteristic map corresponding to the low-precision geochemical data (i.e., the first geochemical data); wherein, according to the calculation method shown in the above-mentioned formula 1 to formula 8, through geographically weighted regression and sliding window operation, each independent variable value corresponding to each pixel point (i.e., the value of the pixel point in the first characteristic map is the independent variable value of one independent variable, and the value of the pixel point in the second characteristic map is the independent variable value of another independent variable) can be calculated. The corresponding geographically weighted focal value W ij (i.e. geographical weighting factor).

[0114] In summary, the regression model constructed in the above geographical weighted layer is shown in the following formula 9:

[0115]

[0116] Among them, (x i ,y i ) is the position coordinate of pixel point i in the study area, β0(x i ,y i ) is the intercept of the original model, ε i is the residual at position i;

[0117] β1(x i ,y i ) and β2(xi ,y i ) are respectively the hyperspectral remote sensing image data X rs and the first geochemical data X gd The regression coefficient of β1(x i ,y i ) and β2(x i ,y i ) are respectively the geographical weighted focal value W corresponding to the pixel i in the second feature map ij (i.e., geographic weighted coefficient) and the geographic weighted focal value W′ corresponding to pixel i in the first feature map ij (i.e., geographic weighted coefficient) is determined (after weight normalization, W′ ij It can also be expressed as 1-W ij );

[0118] The prediction result of the dependent variable value corresponding to pixel i is .

[0119] S205, in the fusion layer, for each of the pixel points, when it is determined that the similarity between the independent variable value of the pixel point in the first feature map and the independent variable value in the second feature map is greater than a preset fusion parameter, the geographical weighting coefficients corresponding to the pixel point in the first feature map and the second feature map are obtained from the geographical weighting layer, and based on the obtained geographical weighting coefficients, the independent variable value of the pixel point in the first feature map and the independent variable value in the second feature map are fused to obtain a prediction result of the dependent variable value corresponding to the pixel point.

[0120] Here, in the fusion layer, by comparing the similarity (product) and overall intensity (energy) of the local area of ​​the image, the numerical value representing the similarity between the first feature map M1 (corresponding to the first geochemical data) and the second feature map M2 (corresponding to the hyperspectral remote sensing image data) can be calculated in a standardized manner. The specific calculation method is as follows:

[0121] First, the significance of the first feature map M1 and the second feature map M2 can be calculated respectively using the following formulas 10 and 11:

[0122]

[0123] Among them, conv2() is a two-dimensional convolution operation, and es2() calculates the local energy features of a single feature image. Used to ensure that the convolution kernel maintains symmetry and consistency when processing the edge of the matrix. ones(um) represents the use of a um×um convolution kernel with a value of 1 for calculation;

[0124] M1·M1 represents the dot product of the eigenvalues ​​corresponding to the same pixel point in the two first feature maps M1, and M2·M2 represents the dot product of the eigenvalues ​​corresponding to the same pixel point in the two second feature maps M2.

[0125] It should be noted that, in the embodiments of the present application, it was determined through experiments that um can be preferably set to 3.

[0126] Through the above formulas 10 and 11, the significance of the first feature map M1 and the second feature map M2 can be obtained, which are defined as S1 and S2 respectively. Similarly, the significance between the first feature map M1 and the second feature map M2 can also be calculated, which is recorded as MA (for example, M2·M2 in the above formula 11 can be replaced by M1·M2).

[0127] Finally, the similarity matrix MAT of the first feature map M1 and the second feature map M2 can be calculated by the following formula 12:

[0128]

[0129] Among them, ε is a very small constant used to prevent division by zero errors; in the final similarity matrix, each value generally falls between [0, 1], 1 indicates a complete match, and 0 indicates a complete mismatch.

[0130] In summary, in the fusion layer, each pixel in the first feature map M1 and the second feature map M2 can be traversed. For each pixel, if the similarity of the pixel at the corresponding position in the first feature map M1 and the second feature map M2 (that is, the similarity between the independent variable value of the pixel in the first feature map and the independent variable value in the second feature map) is greater than the preset fusion parameter p, then the two geographical weighting coefficients Wj and (1-Wj) corresponding to the pixel in the first feature map M1 and the second feature map M2 can be obtained from the above-mentioned geographical weighting layer. ij ), and according to the obtained geographic weighted coefficient, the independent variable value X of the pixel point in the first feature map M1 is taken M1 and the independent variable value X in the second feature map M2 M2 After fusion, the prediction result of the dependent variable value corresponding to the pixel point is obtained: W ij ×X M1 +(1-W ij )×X M2 .

[0131] Specifically, if the similarity of the pixel point at the corresponding position in the first feature map M1 and the second feature map M2 (i.e., the similarity between the independent variable value of the pixel point in the first feature map and the independent variable value in the second feature map) is less than or equal to the preset fusion parameter p, then the independent variable value X of the pixel point in the first feature map M1 can be selected. M1 and the independent variable value X in the second feature map M2 M2 The independent variable value with higher significance is determined as the prediction result of the dependent variable value corresponding to the pixel point.

[0132] Combined with the contents of the above steps S204-S205, it can be seen that, whether in the geographical weighted layer or in the fusion layer, the prediction result of the dependent variable value corresponding to each pixel point is consistent with the above two geographical weighted coefficients W corresponding to the pixel point in the first feature map and the second feature map respectively. ij and (1-W ij ), therefore, when training the original model, for each pixel point, the prediction result can be obtained according to the value of the dependent variable corresponding to the pixel point. The prediction loss between the dependent variable value corresponding to the pixel point (including the prediction results of the two dependent variable values ​​corresponding to the pixel point in the geographical weighted layer and the fusion layer respectively) and the dependent variable value corresponding to the pixel point (that is, the dependent variable value y corresponding to the pixel point in the third feature map) adjusts the geographical weighting coefficient corresponding to the pixel point in the first feature map M1 and the second feature map M1 respectively until the original model converges.

[0133] In the embodiment of the present application, as an optional embodiment, the second geochemical data of the whole area generated in the above steps S101-S104 can also be used as low-precision geochemical data according to the method described in the following step d1, and the third geochemical data of higher precision (with a scale larger than the scale of the second geochemical data) of the whole area of ​​the study area can be generated by the trained geochemical data generation model. Specifically:

[0134] Step d1, adjusting the resolution of the second geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area to the target resolution, and then inputting them into the geochemical data generation model, and outputting the fusion result of the second geochemical data of the entire area and the hyperspectral remote sensing image data of the entire area as the third geochemical data of the entire study area.

[0135] Here, the scale of the second geochemical data is smaller than that of the third geochemical data; the above-mentioned target resolution represents the resolution corresponding to the third geochemical data, that is, the specific implementation of step d1 is basically the same as the aforementioned step S104, except that the high-precision geochemical data of the above-mentioned local area is changed from the second geochemical data to the third geochemical data. Therefore, referring to the content of the above-mentioned step S201, it is necessary to use the resolution corresponding to the third geochemical data (i.e., the target resolution) as the basis in the input layer of the model to unify the resolution of all model input data, and the repetitions will not be repeated here.

[0136] In the embodiment of the present application, as another optional embodiment, considering that when geochemical data of different scales are fused with the same hyperspectral remote sensing image data, if the same geographic weighting coefficient is still used (i.e., directly using the previously trained geochemical data generation model as in the above step d1), the data accuracy of the higher-precision geochemical data obtained after fusion (such as the above-mentioned third geochemical data) may be reduced. Therefore, unlike the above-mentioned step d1, the above-mentioned third geochemical data with higher precision can also be generated according to the method described in the following steps e1-e4, specifically:

[0137] Step e1, obtaining second geochemical data, hyperspectral remote sensing image data and third geochemical data of a local area in the study area; wherein the scale of the second geochemical data is smaller than the scale of the third geochemical data.

[0138] Step e2: input the second geochemical data, the hyperspectral remote sensing image data and the third geochemical data of the local area into the geochemical data generation model; and use the geochemical data generation model to fuse the second geochemical data of the local area and the hyperspectral remote sensing image data of the local area to generate the third geochemical data for prediction, and output a prediction result for the third geochemical data.

[0139] Step e3: adjusting the model parameters of the geochemical data generation model according to the prediction result of the third geochemical data and the prediction loss between the third geochemical data, until the geochemical data generation model reaches convergence again.

[0140] Step e4, input the second geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire area into the geochemical data generation model after convergence again, and output the fusion result of the second geochemical data of the entire area and the hyperspectral remote sensing image data of the entire area as the third geochemical data of the entire study area.

[0141] It should be noted that the specific implementation of the above steps e1-e4 is basically the same as the above steps S101-S104, and the repeated parts are not repeated here.

[0142] In addition, before executing the above step e2, a certain proportion (such as 40%) of the data can be extracted from the low-precision geochemical data (i.e., the second geochemical data of the local area) as a validation set, and the remaining data (such as the remaining 60%) are divided into data sets according to different division ratios (i.e., the data proportions of the training set and the test set are different) (e.g., 60% of the remaining data are used as the training set, and 40% of the remaining data are used as the training set), so as to obtain a plurality of training set and test set combinations corresponding to different division ratios, and use different combinations of training sets and test sets to respectively train the above geochemical data generation model, so as to obtain a plurality of trained geochemical data generation models with the same structure but different training sets and training results used during training; then, the same validation set is used to respectively verify the model performance of each of the above geochemical data generation models, and the model evaluation results shown in Table 2 are obtained, so as to comprehensively select the model with the best evaluation result as the geochemical data generation model used in the actual training in the subsequent step e2, specifically:

[0143]

[0144] Table 2

[0145] As an exemplary explanation, as shown in Table 2, when the proportions of the training set and the test set are 65% and 35% respectively, the geochemical data generation model has the best performance. At this time, the geochemical data generation model trained under this combination of training set and test set can be selected as the geochemical data generation model used in the actual training in the subsequent step e2.

[0146] It should be noted that, after subsequent experimental statistics, the high-precision geochemical data of the entire study area generated by the method shown in the above steps S101-S104 provided in the embodiment of the present application and the high-precision geochemical data of the entire area obtained by data collection in the traditional method (such as the second geochemical data of the entire area obtained by data collection) have a high degree of consistency in multiple specific indicators representing the element concentration distribution of each chemical element in the study area (equivalent to the difference between the high-precision geochemical data of the entire study area generated by the method shown in the above steps S101-S104 and the high-precision geochemical data of the entire area obtained by data collection in the traditional method is small), which can also prove that generating high-precision geochemical data according to the generation method shown in the above steps S101-S104 provided in the embodiment of the present application can ensure data accuracy.

[0147] Specifically, the mean, standard deviation, and average gradient of the element concentration distribution of a single chemical element (such as copper, gold, antimony, etc.) are used as the above-mentioned specific indicators representing the element concentration distribution of each chemical element in the study area. Through experimental comparison, the difference between the high-precision geochemical data of the entire study area generated by the method shown in the above steps S101-S104 provided in the embodiment of the present application and the high-precision geochemical data of the entire area obtained by data collection in the traditional method can be obtained as shown in the following Tables 3-5:

[0148]

[0149] Table 3

[0150]

[0151] Table 4

[0152]

[0153] Table 5

[0154] Among them, the above Table 3 corresponds to the element concentration distribution of the Cu element in the study area; the above Table 4 corresponds to the element concentration distribution of the Au element in the study area; and the above Table 5 corresponds to the element concentration distribution of the Sb element in the study area.

[0155] Specifically, the mean, standard deviation, and average gradient of the element concentration distribution of the element combination (such as copper and gold) are used as the above-mentioned specific indicators representing the element concentration distribution of each chemical element in the study area. Through experimental comparison, the difference between the high-precision geochemical data of the entire study area generated by the method shown in the above steps S101-S104 provided in the embodiment of the present application and the high-precision geochemical data of the entire area obtained by data collection in the traditional method can be obtained as shown in the following Tables 6-7:

[0156]

[0157] Table 6

[0158]

[0159] Table 7

[0160] Among them, the above Table 6 corresponds to the element concentration distribution of the Cu-Au element combination in the study area; the above Table 7 corresponds to the element concentration distribution of the As-Sb element combination in the study area.

[0161] Based on the above-mentioned high-precision geochemical data generation method based on remote sensing images provided in the embodiment of the present application, through the trained geochemical data generation model, the small-scale geochemical data and hyperspectral remote sensing image data in the entire area can be integrated, so there is no need to collect large-scale geochemical data in the entire area, and large-scale geochemical data covering the entire area can be generated. Furthermore, on the basis of reducing data collection costs and improving data collection efficiency, geological survey personnel can also obtain geochemical data of multiple different scales corresponding to the entire area at the same time, so as to meet the higher-precision work task requirements of geological survey personnel.

[0162] Based on the same inventive concept, the present application also provides a high-precision geochemical data generation device corresponding to the above-mentioned high-precision geochemical data generation method based on remote sensing images. Since the principle of solving the problem by the high-precision geochemical data generation device in the embodiment of the present application is similar to the above-mentioned high-precision geochemical data generation method based on remote sensing images in the embodiment of the present application, the implementation of the high-precision geochemical data generation device can refer to the implementation of the above-mentioned generation method, and the repeated parts will not be repeated.

[0163] Reference Figure 3 As shown, Figure 3 The schematic diagram of the structure of a high-precision geochemical data generation device based on remote sensing images provided in an embodiment of the present application is shown, wherein the high-precision geochemical data generation device includes:

[0164] The data acquisition module 301 is used to obtain first geochemical data, hyperspectral remote sensing image data and second geochemical data of a local area in the study area; wherein the scale of the first geochemical data is smaller than the scale of the second geochemical data;

[0165] A prediction module 302 is used to input the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the original model, fuse the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area through the original model to generate the second geochemical data for prediction, and output a prediction result for the second geochemical data;

[0166] A parameter adjustment module 303, used for adjusting the model parameters of the original model according to the prediction loss between the prediction result and the second geochemical data, and obtaining the original model including the adjusted model parameters as a geochemical data generation model;

[0167] The data generation module 304 is used to input the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area into the geochemical data generation model, and output the fusion result of the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area as the second geochemical data of the entire study area.

[0168] In an optional implementation, the data acquisition module 301 is used to obtain the first geochemical data of the local area by the following method:

[0169] According to the sampling point density corresponding to the scale of the first geochemical data, the first sampling medium in the study area is sampled to obtain the first sampling data, and the element concentration distribution data of the whole study area is obtained by performing element content analysis on the first sampling data;

[0170] Converting the data format of the element concentration distribution data of the entire area into a raster data format that matches the hyperspectral remote sensing image data, and adjusting the coordinate system of the element concentration distribution data of the entire area after the format conversion to a projection coordinate system that matches the hyperspectral remote sensing image data, to obtain the element concentration distribution data of the entire area to be clipped;

[0171] The element concentration distribution data of the entire area to be clipped is clipped according to the position coordinates corresponding to the local area in the projection coordinate system to obtain the first geochemical data of the local area.

[0172] In an optional implementation, the data acquisition module 301 is used to obtain the second geochemical data of the local area by the following method:

[0173] According to the density of sampling points corresponding to the scale of the second geochemical data, sampling the second sampling medium in the local area to obtain second sampling data, and performing element content analysis on the second sampling data to obtain element concentration distribution data of the local area;

[0174] The data format of the element concentration distribution data of the local area is converted into a raster data format that matches the hyperspectral remote sensing image data, and the coordinate system of the element concentration distribution data of the local area after the format conversion is adjusted to a projection coordinate system that matches the hyperspectral remote sensing image data to obtain the second geochemical data of the local area.

[0175] In an optional implementation, the data acquisition module 301 is used to obtain the hyperspectral remote sensing image data of the local area by the following method:

[0176] Filter target hyperspectral remote sensing image data corresponding to multiple bands within a target band range from the hyperspectral remote sensing image data of the entire study area; wherein the target band range represents a band range with obvious absorption and reflection characteristics for altered minerals;

[0177] The target hyperspectral remote sensing image data is clipped according to the position coordinates corresponding to the local area in the projection coordinate system to obtain the hyperspectral remote sensing image data of the local area.

[0178] In an optional implementation, the model structure of the original model includes: an input layer, multiple convolutional layers, a connection layer, a geographically weighted layer, and a fusion layer.

[0179] In an optional implementation, when the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area are input into the original model, and the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area are fused by the original model to generate the second geochemical data for prediction, the prediction module 302 is used to:

[0180] Inputting the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the input layer, adjusting the resolution of the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area to be the same as the second geochemical data of the local area through the input layer, and inputting the adjusted data into the multiple convolutional layers;

[0181] Performing multi-scale feature extraction on the received first geochemical data, hyperspectral remote sensing image data and second geochemical data of the local area through the multiple convolutional layers, respectively, and inputting the obtained multi-scale feature maps corresponding to the first geochemical data, hyperspectral remote sensing image data and second geochemical data of the local area respectively into the connection layer;

[0182] The multi-scale feature maps corresponding to the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area are respectively fused through the connection layer to obtain a first feature map corresponding to the first geochemical data of the local area, a second feature map corresponding to the hyperspectral remote sensing image data of the local area and a third feature map corresponding to the second geochemical data of the local area, and the first feature map, the second feature map and the third feature map are input into the geographical weighted layer;

[0183] In the geographic weighted layer, the first feature map and the second feature map are used as independent variables, and the third feature map is used as a dependent variable. For each pixel point, the prediction result of the dependent variable value corresponding to the pixel point is calculated according to the independent variable value corresponding to the pixel point in each of the independent variables and the geographic weighting coefficient;

[0184] In the fusion layer, for each pixel point, when it is determined that the similarity between the independent variable value of the pixel point in the first feature map and the independent variable value in the second feature map is greater than a preset fusion parameter, the geographical weighting coefficient corresponding to the pixel point in the first feature map and the second feature map is obtained from the geographical weighting layer, and based on the obtained geographical weighting coefficient, the independent variable value of the pixel point in the first feature map and the independent variable value in the second feature map are fused to obtain the prediction result of the dependent variable value corresponding to the pixel point.

[0185] In an optional implementation, when adjusting the model parameters of the original model according to the prediction loss between the prediction result and the second geochemical data, the parameter adjustment module 303 is used to:

[0186] For each of the pixel points, according to the prediction loss between the prediction result of the dependent variable value corresponding to the pixel point and the dependent variable value corresponding to the pixel point, adjust the geographic weighting coefficient corresponding to the pixel point in the first feature map and the second feature map respectively until the original model converges.

[0187] In an optional embodiment, the high-precision geochemical data generating device further includes: a first updating module, wherein the first updating module is used to:

[0188] The resolutions of the second geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area are adjusted to the target resolution and then input into the geochemical data generation model, and the fusion results of the second geochemical data of the entire area and the hyperspectral remote sensing image data of the entire area are output as the third geochemical data of the entire study area; wherein the scale of the second geochemical data is smaller than the scale of the third geochemical data; and the target resolution represents the resolution corresponding to the third geochemical data.

[0189] In an optional embodiment, the high-precision geochemical data generating device further includes: a second updating module, wherein the second updating module is used to:

[0190] Acquire second geochemical data, hyperspectral remote sensing image data and third geochemical data of a local area in the study area; wherein the scale of the second geochemical data is smaller than the scale of the third geochemical data;

[0191] Inputting the second geochemical data, the hyperspectral remote sensing image data and the third geochemical data of the local area into the geochemical data generation model, fusing the second geochemical data of the local area and the hyperspectral remote sensing image data of the local area through the geochemical data generation model to generate the third geochemical data for prediction, and outputting the prediction result for the third geochemical data;

[0192] adjusting a model parameter of the geochemical data generation model according to a prediction result of the third geochemical data and a prediction loss between the third geochemical data until the geochemical data generation model reaches convergence again;

[0193] The second geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire area are input into the geochemical data generation model after convergence again, and the fusion result of the second geochemical data of the entire area and the hyperspectral remote sensing image data of the entire area is output as the third geochemical data of the entire study area.

[0194] like Figure 4 As shown, an embodiment of the present application provides an electronic device 400 for executing the high-precision geochemical data generation method based on remote sensing images in the present application, the device includes a memory 401, a processor 402, and a computer program stored in the memory 401 and executable on the processor 402, wherein the memory 401 and the processor 402 are communicatively connected via a bus, and the processor 402 implements the steps of the high-precision geochemical data generation method based on remote sensing images when executing the computer program.

[0195] Specifically, the above-mentioned memory 401 and processor 402 can be general-purpose memory and processor, which are not specifically limited here. When the processor 402 runs the computer program stored in the memory 401, it can execute the above-mentioned high-precision geochemical data generation method based on remote sensing images.

[0196] Corresponding to the high-precision geochemical data generation method based on remote sensing images in the present application, an embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned high-precision geochemical data generation method based on remote sensing images are executed.

[0197] Specifically, the storage medium can be a general storage medium, such as a mobile disk, a hard disk, etc. When the computer program on the storage medium is run, the above-mentioned high-precision geochemical data generation method based on remote sensing images can be executed.

[0198] In the embodiments provided in the present application, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0199] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0200] In addition, each functional unit in the embodiments provided in the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0201] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0202] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0203] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

[0204] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present application, which are used to illustrate the technical solutions of the present application, rather than to limit them. The protection scope of the present application is not limited thereto. Although the present application is described in detail with reference to the above-mentioned embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-mentioned embodiments within the technical scope disclosed in the present application, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application. They should all be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for generating high-precision geochemical data based on remote sensing images, characterized in that: The high-precision geochemical data generation method comprises: Obtaining first geochemical data, hyperspectral remote sensing image data and second geochemical data of a local area in the study area; wherein the scale of the first geochemical data is smaller than the scale of the second geochemical data; Inputting the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the original model, fusing the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area through the original model to generate the second geochemical data for prediction, and outputting the prediction result for the second geochemical data; According to the prediction loss between the prediction result and the second geochemical data, the model parameters of the original model are adjusted to obtain the original model including the adjusted model parameters as a geochemical data generation model; The first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area are input into the geochemical data generation model, and the fusion result of the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area is output as the second geochemical data of the entire study area.

2. The method for generating high-precision geochemical data according to claim 1, characterized in that: The first geochemical data of the local area is obtained by the following method: According to the sampling point density corresponding to the scale of the first geochemical data, the first sampling medium in the study area is sampled to obtain the first sampling data, and the element concentration distribution data of the whole study area is obtained by performing element content analysis on the first sampling data; Converting the data format of the element concentration distribution data of the entire area into a raster data format that matches the hyperspectral remote sensing image data, and adjusting the coordinate system of the element concentration distribution data of the entire area after the format conversion to a projection coordinate system that matches the hyperspectral remote sensing image data, to obtain the element concentration distribution data of the entire area to be clipped; The element concentration distribution data of the entire area to be clipped is clipped according to the position coordinates corresponding to the local area in the projection coordinate system to obtain the first geochemical data of the local area.

3. The method for generating high-precision geochemical data according to claim 1, characterized in that: The second geochemical data of the local area is obtained by the following method: According to the sampling point density corresponding to the scale of the second geochemical data, sampling the second sampling medium in the local area to obtain second sampling data, and obtaining element concentration distribution data of the local area by performing element content analysis on the second sampling data; The data format of the element concentration distribution data of the local area is converted into a raster data format that matches the hyperspectral remote sensing image data, and the coordinate system of the element concentration distribution data of the local area after the format conversion is adjusted to a projection coordinate system that matches the hyperspectral remote sensing image data to obtain the second geochemical data of the local area.

4. The method for generating high-precision geochemical data according to claim 1, characterized in that: The hyperspectral remote sensing image data of the local area is obtained by the following method: Filter target hyperspectral remote sensing image data corresponding to multiple bands within a target band range from the hyperspectral remote sensing image data of the entire study area; wherein the target band range represents a band range with obvious absorption and reflection characteristics for altered minerals; The target hyperspectral remote sensing image data is clipped according to the position coordinates corresponding to the local area in the projection coordinate system to obtain the hyperspectral remote sensing image data of the local area.

5. The method for generating high-precision geochemical data according to claim 1, characterized in that: The model structure of the original model includes: an input layer, multiple convolutional layers, a connection layer, a geographical weighted layer and a fusion layer.

6. The method for generating high-precision geochemical data according to claim 5, characterized in that: The first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area are input into the original model, and the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area are fused by the original model to generate the second geochemical data for prediction, including: Inputting the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the input layer, adjusting the resolution of the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area to be the same as the second geochemical data of the local area through the input layer, and inputting the adjusted data into the multiple convolutional layers; Performing multi-scale feature extraction on the received first geochemical data, hyperspectral remote sensing image data and second geochemical data of the local area through the multiple convolutional layers, respectively, and inputting the obtained multi-scale feature maps corresponding to the first geochemical data, hyperspectral remote sensing image data and second geochemical data of the local area respectively into the connection layer; The multi-scale feature maps corresponding to the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area are respectively fused through the connection layer to obtain a first feature map corresponding to the first geochemical data of the local area, a second feature map corresponding to the hyperspectral remote sensing image data of the local area and a third feature map corresponding to the second geochemical data of the local area, and the first feature map, the second feature map and the third feature map are input into the geographical weighted layer; In the geographic weighted layer, the first feature map and the second feature map are used as independent variables, and the third feature map is used as a dependent variable. For each pixel point, the prediction result of the dependent variable value corresponding to the pixel point is calculated according to the independent variable value corresponding to the pixel point in each of the independent variables and the geographic weighting coefficient; In the fusion layer, for each pixel point, when it is determined that the similarity between the independent variable value of the pixel point in the first feature map and the independent variable value in the second feature map is greater than a preset fusion parameter, the geographical weighting coefficient corresponding to the pixel point in the first feature map and the second feature map is obtained from the geographical weighting layer, and based on the obtained geographical weighting coefficient, the independent variable value of the pixel point in the first feature map and the independent variable value in the second feature map are fused to obtain the prediction result of the dependent variable value corresponding to the pixel point.

7. The method for generating high-precision geochemical data according to claim 6, characterized in that: The adjusting the model parameters of the original model according to the prediction loss between the prediction result and the second geochemical data comprises: For each of the pixel points, according to the prediction loss between the prediction result of the dependent variable value corresponding to the pixel point and the dependent variable value corresponding to the pixel point, adjust the geographic weighting coefficient corresponding to the pixel point in the first feature map and the second feature map respectively until the original model converges.

8. The method for generating high-precision geochemical data according to claim 1, characterized in that: The high-precision geochemical data generation method further comprises: The resolutions of the second geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area are adjusted to the target resolution and then input into the geochemical data generation model, and the fusion results of the second geochemical data of the entire area and the hyperspectral remote sensing image data of the entire area are output as the third geochemical data of the entire study area; wherein the scale of the second geochemical data is smaller than the scale of the third geochemical data; and the target resolution represents the resolution corresponding to the third geochemical data.

9. The method for generating high-precision geochemical data according to claim 1, characterized in that: The high-precision geochemical data generation method further comprises: Acquire second geochemical data, hyperspectral remote sensing image data and third geochemical data of a local area in the study area; wherein the scale of the second geochemical data is smaller than the scale of the third geochemical data; Inputting the second geochemical data, the hyperspectral remote sensing image data and the third geochemical data of the local area into the geochemical data generation model, fusing the second geochemical data of the local area and the hyperspectral remote sensing image data of the local area through the geochemical data generation model to generate the third geochemical data for prediction, and outputting the prediction result for the third geochemical data; adjusting a model parameter of the geochemical data generation model according to a prediction result of the third geochemical data and a prediction loss between the third geochemical data until the geochemical data generation model reaches convergence again; The second geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire area are input into the geochemical data generation model after convergence again, and the fusion result of the second geochemical data of the entire area and the hyperspectral remote sensing image data of the entire area is output as the third geochemical data of the entire study area.

10. A high-precision geochemical data generation device based on remote sensing images, characterized in that: The high-precision geochemical data generating device comprises: A data acquisition module is used to obtain first geochemical data, hyperspectral remote sensing image data and second geochemical data of a local area in the study area; wherein the scale of the first geochemical data is smaller than the scale of the second geochemical data; A prediction module, used for inputting the first geochemical data, the hyperspectral remote sensing image data and the second geochemical data of the local area into the original model, fusing the first geochemical data of the local area and the hyperspectral remote sensing image data of the local area through the original model to generate the second geochemical data for prediction, and outputting the prediction result for the second geochemical data; A parameter adjustment module, used for adjusting the model parameters of the original model according to the prediction loss between the prediction result and the second geochemical data, and obtaining the original model including the adjusted model parameters as a geochemical data generation model; The data generation module is used to input the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area into the geochemical data generation model, and output the fusion result of the first geochemical data of the entire study area and the hyperspectral remote sensing image data of the entire study area as the second geochemical data of the entire study area.

Images