A Quantitative Inversion Method and System for Heavy Metal Content by Combining GF-5 Remote Sensing Images and Ground Soil Hyperspectra

Abstract

The present invention is a method and system for quantitatively retrieving heavy metal contents by combining GF-5 remote sensing images and ground soil hyperspectra, including: obtaining a bare soil image based on the normalized vegetation index value of the GF-5 remote sensing image, and performing DS correction and first derivative spectral transformation on the bare soil image; obtaining ground hyperspectral reflectance data, soil heavy metal contents, and remote sensing image hyperspectral reflectance data of sampling points; performing DS correction and first derivative spectral transformation on the remote sensing image hyperspectral reflectance data; screening the remote sensing image hyperspectral reflectance data after the first derivative spectral transformation through the Boruta algorithm, and combining with the soil heavy metal contents to construct a heavy metal inversion model; predicting the bare soil image after the first derivative spectral transformation based on the heavy metal inversion model to obtain a spatial distribution map of heavy metal content pollution. The present invention can quickly and accurately monitor the soil heavy metal pollution situation in a large area, providing a basis for ecological restoration.

Description

Technical Field

[0001] The present invention belongs to the technical field of hyperspectral remote sensing quantitative inversion of heavy metal content in soil, and particularly relates to a method and system for quantitatively inverting heavy metal content by combining GF-5 remote sensing images and ground soil hyperspectra. Background Art

[0002] Heavy metal pollution in soil has become an important research topic in the field of environmental science. The accumulation of heavy metal content will not only have a negative impact on the soil ecosystem, but also pose a serious threat to human health through the food chain. Therefore, how to quickly and accurately monitor the heavy metal content in soil is of great significance for environmental remediation. Traditional methods for detecting heavy metals in soil usually involve field soil collection and laboratory chemical analysis. This method has high detection accuracy, but is time-consuming, costly, and difficult to achieve large-scale monitoring.

[0003] Meanwhile, with the development of remote sensing technology, satellite remote sensing has gradually become the main means of environmental monitoring due to its advantages such as high efficiency and low cost. Among them, hyperspectral remote sensing has been widely used in the monitoring and research of soil heavy metal pollution due to its advantages of many bands and strong spectral continuity, making it possible to quantitatively invert the heavy metal content in soil. Especially the domestic GF-5 hyperspectral remote sensing, whose spectral range covers 400 - 2500 nm, is consistent with the ground hyperspectral band range, and is deeply favored by remote sensing personnel in the field of inverting soil heavy metal content.

[0004] Although the effect of inverting soil heavy metal content with GF-5 images is significant, there are still many technical difficulties in the inversion process due to sensor limitations and the influence of natural and human factors. For example, due to the limitation of the spatial resolution of hyperspectral images, soil spectra will show a mixing phenomenon; there will be human interference when measuring the content and spectra of soil samples; while hyperspectral has the advantages of many bands and high spectral resolution, it also brings problems such as a large amount of data redundancy and strong collinearity between bands. In addition, capturing the spectral response of trace elements, especially heavy metals, in soil is still a major challenge. There is still uncertainty in estimating soil heavy metal content using hyperspectral image data. Therefore, there is an urgent need to develop a feasible and reliable method and system for estimating heavy metal concentration using GF-5 AHSI images. Summary of the Invention

[0005] In order to effectively combine ground indoor spectra and GF-5 spectra to achieve the purpose of large-scale and rapid monitoring of soil heavy metal content, the present invention provides a method and system for quantitatively inverting heavy metal content by combining GF-5 remote sensing images and ground soil hyperspectra.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A method for quantitatively retrieving heavy metal content by combining GF-5 remote sensing images and ground soil hyperspectra, comprising the following steps:

[0008] Collect soil samples at preset sampling points and download GF-5 remote sensing images of the sampling points

[0009] Based on the normalized difference vegetation index value of the GF-5 remote sensing image, obtain a bare soil image, and perform DS correction and first derivative spectral transformation on the bare soil image to obtain a bare soil image after first derivative spectral transformation;

[0010] Based on the soil samples, obtain ground hyperspectral reflectance data, soil heavy metal content, and GF-5 remote sensing image hyperspectral reflectance data of the sampling points;

[0011] Taking the ground hyperspectral reflectance data as a standard, perform DS correction on the GF-5 remote sensing image hyperspectral reflectance data, and perform first derivative spectral transformation on the GF-5 remote sensing image hyperspectral reflectance data after DS correction;

[0012] Screen the GF-5 remote sensing image hyperspectral reflectance data after first derivative spectral transformation through the Boruta algorithm, and combine the soil heavy metal content to construct a heavy metal inversion model;

[0013] Based on the heavy metal inversion model, predict the bare soil image after first derivative spectral transformation to obtain a spatial distribution map of heavy metal content pollution, and complete the quantitative inversion of heavy metal content.

[0014] Preferably, the method for performing DS correction on the GF-5 remote sensing image hyperspectral reflectance data includes: obtaining the GF-5 remote sensing image hyperspectral reflectance data after DS correction based on the conversion matrix and residual matrix of the ground hyperspectral reflectance data and the GF-5 remote sensing image hyperspectral reflectance data, and the calculation formula is as follows:

[0015] X Lab =X GF-5 B+E,

[0016] X′ GF-5 =X GF-5 B+E,

[0017] Wherein, X Lab represents the original ground hyperspectral reflectance data of the sample point, X GF-5 represents the GF-5 remote sensing image hyperspectral reflectance data of the sample point, B is the conversion matrix of X Lab and X GF-5 , E is the residual matrix, and X' GF-5 represents the GF-5 remote sensing image hyperspectral reflectance data after DS transformation.

[0018] Preferably, the method for screening the hyperspectral reflectance data of GF-5 remote sensing images after the first derivative spectral transformation by the Boruta algorithm includes:

[0019] Randomly shuffle the hyperspectral reflectance data of GF-5 remote sensing images after the first derivative spectral transformation as the original features to obtain shadow features;

[0020] Based on the shadow features and the original features, construct and train a random forest model;

[0021] Based on the trained random forest model, calculate and compare the importance scores of the original features and the shadow features respectively. When the comparison result of the importance scores meets the preset requirements, obtain the hyperspectral feature bands as important features.

[0022] Preferably, the method for constructing a heavy metal inversion model includes: taking the hyperspectral feature bands as independent variables and the soil heavy metal content as the dependent variable and inputting them into the XGBoost model to construct an XGBoost-based heavy metal inversion model.

[0023] Preferably, the construction method of the XGBoost model includes:

[0024] Based on the actual value of the soil heavy metal content and the predicted value of the soil heavy metal content predicted by the decision tree, establish a loss function;

[0025] Based on the loss function and the regularization term for the decision tree, obtain the objective function of the XGBoost model;

[0026] Use the second-order Taylor expansion based on gradient boosting to approximate the loss function to obtain an approximation of the objective function;

[0027] Based on the approximation of the objective function, calculate the optimal weights of the decision tree leaf nodes in the regularization term, and based on the optimal weights, calculate the leaf node splitting gain to obtain the optimal splitting point, and complete the construction of the XGBoost model.

[0028] A quantitative inversion system for heavy metal content that combines GF-5 remote sensing images and ground soil hyperspectra according to the present invention is used to implement the method, and includes:

[0029] A collection module for collecting soil samples at preset sampling points and downloading GF-5 remote sensing images of the sampling points

[0030] A bare soil image acquisition module for obtaining a bare soil image based on the normalized vegetation index value of the GF-5 remote sensing image, and performing DS correction and first derivative spectral transformation on the bare soil image to obtain a bare soil image after the first derivative spectral transformation;

[0031] A data acquisition module, configured to obtain ground hyperspectral reflectance data, soil heavy metal content, and GF-5 remote sensing image hyperspectral reflectance data of a sampling point based on the soil sample;

[0032] A data transformation module, configured to perform DS correction on the GF-5 remote sensing image hyperspectral reflectance data with the ground hyperspectral reflectance data as a standard, and perform first derivative spectral transformation on the DS-corrected GF-5 remote sensing image hyperspectral reflectance data;

[0033] An inversion model construction module, configured to screen the GF-5 remote sensing image hyperspectral reflectance data after first derivative spectral transformation through the Boruta algorithm, and construct a heavy metal inversion model in combination with the soil heavy metal content;

[0034] A prediction module, configured to predict the bare soil image after first derivative spectral transformation based on the heavy metal inversion model, obtain a heavy metal content pollution spatial distribution map, and complete the quantitative inversion of the heavy metal content.

[0035] Preferably, the inversion model construction module includes a band screening unit, configured to screen the GF-5 remote sensing image hyperspectral reflectance data after first derivative spectral transformation through the Boruta algorithm; the band screening unit includes:

[0036] A shadow feature acquisition sub-unit, configured to randomly shuffle the GF-5 remote sensing image hyperspectral reflectance data after first derivative spectral transformation as original features to obtain shadow features;

[0037] A random forest construction sub-unit, configured to construct and train a random forest model based on the shadow features and the original features;

[0038] An importance evaluation sub-unit, configured to calculate and compare the importance scores of the original features and the shadow features respectively based on the trained random forest model, and obtain the hyperspectral feature bands as important features when the comparison result of the importance scores meets the preset requirements.

[0039] Preferably, in the inversion model construction module, the hyperspectral feature bands are used as independent variables, and the soil heavy metal content is used as a dependent variable and input into the XGBoost model to construct a heavy metal inversion model based on XGBoost.

[0040] Compared with the prior art, the beneficial effects of the present invention are:

[0041] (1) High efficiency and accuracy: By combining GF-5 remote sensing images with ground soil hyperspectral data, the present invention can quickly and accurately invert the heavy metal content in the soil, overcoming the time-consuming and high-cost problems of traditional soil detection methods, thereby realizing large-scale environmental monitoring.

[0042] (2) Feature optimization: The present invention uses the DS algorithm to correct the hyperspectral data and selects key feature bands through the Boruta algorithm to ensure the reliability and accuracy of the inversion results. The Boruta algorithm effectively reduces the influence caused by band redundancy; by correcting the GF-5 image spectrum, the problem of poor GF-5 image spectrum quality is solved.

[0043] (3) Spatial distribution visualization: With the help of the XGBoost model, the present invention can not only predict the heavy metal content in the soil, but also generate a spatial distribution map of heavy metal pollution, providing intuitive and easy-to-understand information for environmental management and ecological restoration, and supporting decision-making. Description of the Drawings

[0044] In order to more clearly illustrate the technical solutions of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the following-described drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0045] Figure 1 It is a flowchart of the quantitative inversion method for heavy metal content by combining GF-5 remote sensing images and ground soil hyperspectral data in the embodiments of the present invention;

[0046] Figure 2 It is an example of the spectral effect before and after DS correction in the embodiments of the present invention; among them, (a) is the original laboratory spectrum; (b) is the original GF-5 spectrum; (c) is the GF-5 spectrum after DS correction;

[0047] Figure 3 It is an example of the first derivative spectral transformation and the result of feature band selection in the embodiments of the present invention;

[0048] Figure 4 It is a spatial distribution map of heavy metal zinc pollution in the embodiments of the present invention;

[0049] Figure 5 It is a spatial distribution map of heavy metal lead pollution in the embodiments of the present invention;

[0050] Figure 6 It is a spatial distribution map of heavy metal nickel pollution in the embodiments of the present invention. Detailed Embodiments

[0051] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0052] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0053] Embodiment 1

[0054] As Figure 1 shown, a method for quantitatively retrieving heavy metal content by combining GF-5 remote sensing images and ground soil hyperspectra includes the following steps:

[0055] S1: Collect soil samples at preset sampling points and download the GF-5 remote sensing images of the sampling points; in this embodiment, taking the area around a certain mining area as the experimental site, 83 surface soils of 5-20 cm were collected in places such as farmland, mountains, and roads that researchers can reach, record the number and GPS coordinates of each sample, bring the collected samples back to the laboratory, and at the same time download the GF-5 remote sensing images covering the sampling point range.

[0056] S2: Based on the normalized difference vegetation index (NDVI) value of the GF-5 remote sensing image, obtain a bare soil image, and perform DS correction and first derivative spectral transformation on the bare soil image to obtain the bare soil image after the first derivative spectral transformation.

[0057] Specifically, after preprocessing the obtained GF-5 remote sensing image using ENVI software for radiometric calibration, atmospheric correction, and orthorectification, taking the original ground hyperspectral reflectance data in step 1 as the standard, resample the GF-5 remote sensing image band range to 400 - 2450 nm; then calculate the normalized difference vegetation index (NDVI) value of the image, and use Python programming to extract the GF-5 image with NDVI < 0.4 as the bare soil area. NDVI is calculated using the following formula:

[0058]

[0059] where Band431 is the near-infrared band and Band297 is the red band;

[0060] Then, according to the GPS coordinates of the sampling points, use ENVI software to extract the hyperspectral reflectance data of the bare soil image at the corresponding positions.

[0061] The bare soil image after DS correction is subjected to first - order derivative spectral transformation in ENVI software to ensure that the inversion image and the modeling band values are of the same order of magnitude. The transformed image is used for subsequent heavy metal content prediction.

[0062] S3: Based on soil samples, obtain the ground hyperspectral reflectance data, soil heavy metal content, and GF - 5 remote sensing image hyperspectral reflectance data at the sampling points.

[0063] Specifically, after removing impurities such as weeds and stones from the collected soil samples, air - dry them naturally, grind them with a grinder and pass through a 100 - mesh sieve. Subsequently, use a ground object spectrometer (ASD) to measure the hyperspectral reflectance data of the soil samples, and use an X - ray fluorescence spectrometer (XRF) to measure the heavy metal content of the soil samples. Exclude the bands at 350 - 399 and 2451 - 2500 nm with large noise and unstable spectral data in the ground hyperspectral reflectance data measured by ASD, and then perform SG smoothing on the remaining 2051 bands. The smoothed data is used as the original ground spectral reflectance data. Conduct a skewness distribution test on the heavy metal content measured by XRF. If the heavy metal content belongs to a skewed distribution, perform skewness distribution correction.

[0064] S4: Using the ground hyperspectral reflectance data as a standard, perform DS correction on the GF - 5 remote sensing image hyperspectral reflectance data, and perform first - order derivative spectral transformation on the DS - corrected GF - 5 remote sensing image hyperspectral reflectance data with the help of Python.

[0065] Figure 2 Among them, (a) is the original laboratory spectrum; (b) is the original GF - 5 spectrum; (c) is the GF - 5 spectrum after DS correction. A further implementation method is that the method for performing DS correction on the GF - 5 remote sensing image hyperspectral reflectance data includes: obtaining the DS - corrected GF - 5 remote sensing image hyperspectral reflectance data based on the conversion matrix and residual matrix of the ground hyperspectral reflectance data and the GF - 5 remote sensing image hyperspectral reflectance data. The calculation formula is as follows:

[0066] X Lab =X GF-5 B + E,

[0067] X′ GF-5 =X GF-5 B + E,

[0068] Among them, X Lab represents the original ground hyperspectral reflectance data of the sample points, X GF-5 represents the GF - 5 remote sensing image hyperspectral reflectance data of the sample points, B is the conversion matrix of X Lab and X GF-5 , and E is the residual matrix, X' GF-5Represents the GF-5 remote sensing image hyperspectral reflectance data after DS conversion. Among them, 70% of the original ground hyperspectral reflectance data and GF-5 hyperspectral reflectance data are selected to calculate DS parameters B and E.

[0069] S5: The hyperspectral reflectance data of GF-5 remote sensing images after first-order derivative spectral transformation were screened by Boruta algorithm, and combined with the soil heavy metal content to construct a heavy metal inversion model.

[0070] like Figure 3 As shown, a further implementation method is that the method of screening the GF-5 remote sensing image hyperspectral reflectance data after the first-order derivative spectrum transformation by the Boruta algorithm includes:

[0071] The hyperspectral reflectance data of GF-5 remote sensing images after first-order derivative spectral transformation are randomly shuffled as original features to obtain shadow features;

[0072] Build and train a random forest model based on shadow features and original features;

[0073] Based on the trained random forest model, the importance scores of the original feature and the shadow feature are calculated and compared respectively. If the importance score of the original feature is significantly higher than the score of the shadow feature, the feature is considered "important". If the importance score of the original feature is not higher than the score of the shadow feature, the feature is considered "unimportant". When the comparison result of the importance score meets the above preset requirements, the hyperspectral feature band as the important feature is obtained, and the unimportant features are eliminated. The present invention uses Python programming to call the boruta package to realize feature band selection.

[0074] A further implementation method is that the method for constructing a heavy metal inversion model includes: inputting the hyperspectral characteristic band as an independent variable and the soil heavy metal content as a dependent variable into the XGBoost model to construct a heavy metal inversion model based on XGBoost. Subsequently, the bare soil image after the first-order derivative transformation is input into the trained heavy metal inversion model, and then the bare soil image is processed in blocks, and the heavy metal content value of the bare soil image is predicted in blocks; finally, the predicted image is imported into ArcMap for thematic mapping of the spatial distribution of heavy metal content pollution.

[0075] A further implementation method is that the method for constructing the XGBoost model includes:

[0076] Based on the actual value of soil heavy metal content and the predicted value of soil heavy metal content predicted by the decision tree, a loss function is established; based on the loss function and the regularization term of the decision tree, the objective function of the XGBoost model is obtained.

[0077] Specifically, the objective function of XGBoost can be summarized as the following formula:

[0078]

[0079] Among them, is the loss function, representing the error between the actual value y i and the predicted value . The predicted value here consists of the predicted value from the previous round and the prediction of the current tree (f t (x i ). Ω(f t ) is the regularization term for the tree, used to control the complexity of the model and prevent overfitting.

[0080] The second-order Taylor expansion based on gradient boosting is used to approximate the loss function to obtain an approximation of the objective function. Specifically, to optimize the objective function, XGBoost uses the second-order Taylor expansion to approximate the loss function. That is, the first-order derivative (Gradient) and second-order derivative (Hessian) of the loss function are used to make a quadratic approximation of the error. This enables the model to quickly find the optimal solution. Then the approximate formula of the objective function is:

[0081]

[0082] Among them, is the first-order gradient, representing the gradient of the loss function; is the second-order gradient, representing the curvature of the loss function; Ω(f t ) is the regularization term of the tree, used to control the complexity of the tree.

[0083] Based on the approximation of the objective function, calculate the optimal weights of the decision tree leaf nodes in the regularization term, and based on the optimal weights, calculate the split gain of the leaf nodes to obtain the optimal split point, completing the construction of the XGBoost model. Specifically, XGBoost regularizes the complexity of each tree, and the form of the regularization term is:

[0084]

[0085] Among them: T is the number of leaf nodes of the tree; w j is the weight of the j-th leaf node; γ controls the penalty for the number of leaf nodes, and λ controls the penalty for the weights.

[0086] According to the approximation formula of the Taylor expansion, XGBoost optimizes the model by calculating the optimal weight w j of each leaf node. The calculation formula for the optimal weight is:

[0087]

[0088] Among them, ∑ i∈j g i is the first-order gradient sum of all samples within the j-th leaf node; ∑ i∈j h i is the second-order gradient sum of all samples within the j-th leaf node; λ is the regularization parameter used to control the weight of the leaf node.

[0089] To select the best split point, XGBoost calculates the change (gain) of the objective function before and after splitting. The formula for the gain is:

[0090]

[0091] Among them, L and R respectively represent the left and right child nodes after splitting. ∑ i∈L g i and ∑ i∈R g i are the first-order gradient sums within the left and right child nodes; ∑ i∈L h i and ∑ i∈R h i are the second-order gradient sums within the left and right child nodes; γ is the leaf node penalty term; this formula indicates that if a split can bring sufficient gain (i.e., reduce the loss), then the split will be executed.

[0092] XGBoost generates new trees to optimize the model through continuous iteration. Each round of iteration includes steps such as calculating residuals, fitting trees, updating leaf node weights, and calculating gains until the loss function can no longer be significantly reduced.

[0093] S6: Based on the heavy metal inversion model, predict the bare soil image after the first-order derivative spectral transformation to obtain the spatial distribution map of heavy metal content pollution, and complete the quantitative inversion of heavy metal content. As Figure 4 , Figure 5 , Figure 6 shown.

[0094] Embodiment 2

[0095] A quantitative inversion system for heavy metal content that combines GF-5 remote sensing images and ground soil hyperspectra of the present invention is used to implement the method described in Embodiment 1, and includes:

[0096] A collection module for collecting soil samples at preset sampling points and downloading GF-5 remote sensing images of the sampling points;

[0097] A bare soil image acquisition module for obtaining a bare soil image based on the normalized vegetation index value of the GF-5 remote sensing image, performing DS correction and first-order derivative spectral transformation on the bare soil image to obtain a bare soil image after the first-order derivative spectral transformation;

[0098] A data acquisition module, configured to obtain ground hyperspectral reflectance data, soil heavy metal content, and GF-5 remote sensing image hyperspectral reflectance data of a sampling point based on soil samples;

[0099] A data transformation module, configured to perform DS correction on the GF-5 remote sensing image hyperspectral reflectance data with the ground hyperspectral reflectance data as a standard, and perform first derivative spectral transformation on the DS-corrected GF-5 remote sensing image hyperspectral reflectance data;

[0100] An inversion model construction module, configured to screen the GF-5 remote sensing image hyperspectral reflectance data after first derivative spectral transformation through the Boruta algorithm, and construct a heavy metal inversion model in combination with the soil heavy metal content;

[0101] A prediction module, configured to predict the bare soil image after first derivative spectral transformation based on the heavy metal inversion model, obtain a heavy metal content pollution spatial distribution map, and complete the quantitative inversion of the heavy metal content.

[0102] A further implementation manner lies in that the inversion model construction module includes a band screening unit, configured to screen the GF-5 remote sensing image hyperspectral reflectance data after first derivative spectral transformation through the Boruta algorithm; the band screening unit includes:

[0103] A shadow feature acquisition subunit, configured to randomly shuffle the GF-5 remote sensing image hyperspectral reflectance data after first derivative spectral transformation as original features to obtain shadow features;

[0104] A random forest construction subunit, configured to construct and train a random forest model based on the shadow features and the original features;

[0105] An importance evaluation subunit, configured to calculate and compare the importance scores of the original features and the shadow features respectively based on the trained random forest model, and obtain the hyperspectral feature bands as important features when the comparison result of the importance scores meets the preset requirements.

[0106] A further implementation manner lies in that in the inversion model construction module, the hyperspectral feature bands are used as independent variables and the soil heavy metal content is used as the dependent variable and input into the XGBoost model to construct a heavy metal inversion model based on XGBoost.

[0107] The above-described embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A quantitative inversion method for heavy metal content combining GF-5 remote sensing images and ground soil hyperspectral, characterized in that: The following steps are involved: Collect soil samples from preset sampling points and download GF-5 remote sensing images of the sampling points; Based on the normalized vegetation index value of the GF-5 remote sensing image, a bare soil image is obtained, and the bare soil image is subjected to DS correction and first-order derivative spectrum transformation to obtain a bare soil image after first-order derivative spectrum transformation; Based on the soil samples, ground hyperspectral reflectance data of the sampling points, soil heavy metal content, and GF-5 remote sensing image hyperspectral reflectance data are obtained; Taking the ground hyperspectral reflectance data as the standard, performing DS correction on the GF-5 remote sensing image hyperspectral reflectance data, and performing first-order derivative spectral transformation on the GF-5 remote sensing image hyperspectral reflectance data after DS correction; The hyperspectral reflectance data of GF-5 remote sensing images after first-order derivative spectral transformation were screened by Boruta algorithm, and combined with the soil heavy metal content to build a heavy metal inversion model; The method of screening the hyperspectral reflectance data of GF-5 remote sensing images after the first-order derivative spectral transformation by Boruta algorithm includes: The hyperspectral reflectance data of GF-5 remote sensing images after first-order derivative spectral transformation are randomly shuffled as original features to obtain shadow features; Based on the shadow features and the original features, a random forest model is constructed and trained; Based on the trained random forest model, respectively calculating and comparing the importance scores of the original feature and the shadow feature, when the comparison result of the importance scores meets the preset requirements, obtaining the hyperspectral feature band as the important feature; The method for constructing a heavy metal inversion model includes: inputting the hyperspectral characteristic band as an independent variable and the soil heavy metal content as a dependent variable into an XGBoost model to construct a heavy metal inversion model based on XGBoost; The XGBoost model construction method includes: A loss function is established based on the actual value of soil heavy metal content and the predicted value of soil heavy metal content predicted by the decision tree; Based on the loss function and the regularization term of the decision tree, an objective function of the XGBoost model is obtained; Approximating the loss function using a second-order Taylor expansion based on gradient boosting to obtain an approximation of the objective function; Based on the approximation of the objective function, the optimal weights of the decision tree leaf nodes in the regularization term are calculated, and based on the optimal weights, the leaf node splitting gains are calculated to obtain the optimal splitting point, and the construction of the XGBoost model is completed; Based on the heavy metal inversion model, the bare soil image after the first-order derivative spectrum transformation is predicted to obtain the spatial distribution map of heavy metal content pollution, and complete the quantitative inversion of heavy metal content.

2. The method for quantitatively inverting heavy metal content by combining GF-5 remote sensing images and ground soil hyperspectral according to claim 1, characterized in that: The method for performing DS correction on the GF-5 remote sensing image hyperspectral reflectance data includes: obtaining DS-corrected GF-5 remote sensing image hyperspectral reflectance data based on the conversion matrix and residual matrix of the ground hyperspectral reflectance data and the GF-5 remote sensing image hyperspectral reflectance data, and the calculation formula is as follows: ， ， in, Represents the original ground hyperspectral reflectance data of the sample point, Represents the GF-5 remote sensing image hyperspectral reflectance data of the sample point, for and The transformation matrix, is the residual matrix, Represents the GF-5 remote sensing image hyperspectral reflectance data after DS conversion.

3. A system for quantitatively inverting heavy metal content by combining GF-5 remote sensing images with ground soil hyperspectral, used to implement any one of the methods of claims 1-2, characterized in that: include: A collection module is used to collect soil samples at preset sampling points and download GF-5 remote sensing images of the sampling points; A bare soil image acquisition module, used to obtain a bare soil image based on the normalized vegetation index value of the GF-5 remote sensing image, and perform DS correction and first-order derivative spectrum transformation on the bare soil image to obtain a bare soil image after first-order derivative spectrum transformation; A data acquisition module, for obtaining ground hyperspectral reflectance data, soil heavy metal content and GF-5 remote sensing image hyperspectral reflectance data of the sampling point based on the soil sample; A data transformation module is used to perform DS correction on the GF-5 remote sensing image hyperspectral reflectance data based on ground hyperspectral reflectance data, and perform first-order derivative spectral transformation on the GF-5 remote sensing image hyperspectral reflectance data after DS correction; The inversion model building module is used to filter the GF-5 remote sensing image hyperspectral reflectance data after the first-order derivative spectrum transformation through the Boruta algorithm, and build a heavy metal inversion model in combination with the soil heavy metal content; The prediction module is used to predict the bare soil image after the first-order derivative spectrum transformation based on the heavy metal inversion model, obtain the spatial distribution map of heavy metal content pollution, and complete the quantitative inversion of heavy metal content.

4. The system for quantitatively inverting heavy metal content by combining GF-5 remote sensing images with ground soil hyperspectral according to claim 3, characterized in that: The inversion model building module includes a band screening unit for screening the GF-5 remote sensing image hyperspectral reflectance data after the first-order derivative spectrum transformation through the Boruta algorithm; the band screening unit includes: The shadow feature acquisition subunit is used to randomly shuffle the hyperspectral reflectance data of the GF-5 remote sensing image after the first-order derivative spectrum transformation as the original feature to obtain the shadow feature; A random forest construction subunit, used to construct and train a random forest model based on the shadow features and the original features; The importance evaluation subunit is used to calculate and compare the importance scores of the original feature and the shadow feature based on the trained random forest model, and when the comparison result of the importance scores meets the preset requirements, the hyperspectral feature band as the important feature is obtained.

5. The system for quantitatively inverting heavy metal content by combining GF-5 remote sensing images with ground soil hyperspectral according to claim 3, characterized in that: In the inversion model construction module, the hyperspectral characteristic band is used as an independent variable, and the soil heavy metal content is used as a dependent variable to input into the XGBoost model to construct a heavy metal inversion model based on XGBoost.

Images