Method and device for identifying rare earth ore soil pollution by using leaf spectrum

Abstract

The invention discloses a method and a device for identifying rare earth ore soil pollution by using a leaf spectrum, a data acquisition device is adopted to obtain reflection spectrum data of vegetation leaves, the reflection spectrum data is input into a computer, and a processed spectrum is obtained in a data preprocessing module of the computer; meanwhile, a surface soil sample is collected, and the heavy metal content in the surface soil sample is obtained; the processed spectrum is decomposed into a plurality of intrinsic mode function components and a residual term in a multivariate empirical mode decomposition module, then correlation analysis is carried out on the plurality of intrinsic mode function components and the heavy metal content in a data processing cabinet, intrinsic mode function component wave bands with weak correlation are removed, and the heavy metal content is obtained; and finally, constructing a non-linear prediction model by adopting a random forest regression algorithm, and outputting a predicted soil heavy metal content value in the non-linear prediction model. According to the method, the rare earth ore soil heavy metal pollution is accurately identified, and the precision higher than that of a current popular method can be obtained.

Description

Technical Field

[0001] This invention belongs to the field of heavy metal content prediction technology in rare earth mining areas, specifically relating to a method and apparatus for identifying rare earth mine soil pollution using leaf spectroscopy. Background Technology

[0002] Ion-adsorption rare earth mines widely employ acidic leaching processes such as pond leaching, heap leaching, and in-situ leaching during mining. This leads to significant migration and enrichment of heavy metals in the mining soil, creating contaminated areas with high concealment, spatial heterogeneity, and complex composition. Such pollution not only severely damages surface ecosystems but also poses a potential threat to regional ecological security and public health. However, current soil pollution monitoring primarily relies on manual soil sample collection and laboratory chemical analysis, which suffers from cumbersome procedures, long cycles, high costs, and difficulty in frequently covering large areas. Especially in rare earth mining areas with extensive reclaimed vegetation cover, contaminated soil is often obscured by vegetation, making it difficult to directly obtain representative samples, further limiting the applicability of traditional methods on a spatial scale.

[0003] Meanwhile, the spectral identification of soil pollution in rare earth mining areas faces multiple technical challenges: First, the spectral response is weak. The effects of heavy metal stress on vegetation physiology are often gradual and insidious, manifesting only as low-amplitude signals such as slight shifts in the red edge, weakening of the green peak, or a slight decrease in near-infrared reflectance, which are easily masked by noise. Second, the environmental interference is complex. Leaf spectra are modulated by multiple factors, including chlorophyll content, water status, nitrogen level, light conditions, and reclamation stage, making pollution characteristics easily obscured by non-polluting factors. Third, the feature correlation is weak. Hyperspectral data has high dimensionality and severe inter-band redundancy, making it difficult for traditional spectral transformation and modeling methods to effectively extract discriminative features strongly correlated with heavy metal pollution. Fourth, the sample size is limited. Acquiring paired samples of "leaf spectra - measured soil heavy metal values" is costly and limited in quantity, resulting in poor generalization ability and overfitting of traditional models, and insufficient predictive stability in complex mining environments. These problems collectively restrict the accuracy and practicality of existing hyperspectral analysis methods in the identification of pollution in rare earth mining areas. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a method for identifying rare earth mineral soil pollution using leaf spectroscopy, comprising the following steps: S1: Collect multiple samples of the same vegetation sample to obtain multiple reflectance spectral data of the vegetation leaves, and then input the multiple reflectance spectral data into the data preprocessing module to obtain the processed spectrum; at the same time, collect surface soil samples below the root system of the vegetation sample, and then obtain the heavy metal content in the surface soil samples through chemical analysis methods. S2: Input the processed spectrum into the multivariate empirical mode decomposition module for adaptive decomposition, decomposing the processed spectrum into multiple intrinsic mode function components and a residual term; S3: Input multiple intrinsic mode function components and the heavy metal content into the correlation analysis module to perform correlation analysis, obtain the correlation between the intrinsic mode function components and the heavy metal content, then remove the feature bands of the intrinsic mode function components with weak correlation, select the feature bands of the intrinsic mode function components with the strongest correlation, and finally obtain the selected intrinsic mode function components and the heavy metal content; S4: Input the selected intrinsic mode function components and the heavy metal content into the nonlinear prediction module. The nonlinear prediction module adopts a random forest regression model and outputs the predicted soil heavy metal content value.

[0005] On the other hand, the present invention provides an apparatus for the aforementioned method of identifying rare earth mineral soil pollution using leaf spectra, comprising a data acquisition device for collecting vegetation leaf reflectance spectral data, the data acquisition device transmitting the vegetation leaf reflectance spectral data to a data processing device, the data processing device being provided with a data preprocessing module and a multivariate empirical mode decomposition module, the data processing device being connected to a sensitive feature screening device, the sensitive feature screening device being provided with a correlation analysis module, the sensitive feature screening device being connected to a nonlinear prediction device, the nonlinear prediction device being connected to a printing output device, the printing output device generating a drawing from the output results of the nonlinear prediction module to visually display the spatial distribution and intensity of the pollution.

[0006] Preferably, the data acquisition device includes a portable hyperspectral analyzer and a calibration whiteboard, wherein the portable hyperspectral analyzer is connected to a tripod and a GPS module for synchronously recording geographical location.

[0007] Preferably, the sensitive feature screening device, the data processing device, and the nonlinear prediction device are all computer-based.

[0008] Preferably, the computer is equipped with data processing software, which performs the operations of the data preprocessing module, the multivariate empirical mode decomposition module, the correlation analysis module, and the nonlinear prediction module.

[0009] The present invention has the following beneficial effects: 1. This invention addresses the problem of weak spectral response mixed with environmental interference. Utilizing leaf spectral data and soil sample data, it introduces multivariate empirical mode decomposition to adaptively decompose and reconstruct the hyperspectral data of reclaimed vegetation leaves at multiple scales. Furthermore, correlation analysis separates environmental fluctuation components unrelated to heavy metal stress, such as high-frequency / low-frequency noise caused by changes in moisture and light, while preserving and enhancing the intrinsic mode function components with high correlation to soil heavy metal content. This significantly improves the correlation and identifiability between spectral features and pollution indicators, providing reliable technical support for assessing the risk of heavy metal accumulation in soil and the stress status of vegetation.

[0010] 2. This invention effectively explores the joint response mechanism of key band combinations to multi-element compound pollution by selectively screening the most relevant feature bands and combining them with advanced regression models to deal with nonlinear relationships. It solves the problems of sample scarcity and high-dimensional feature redundancy in soil pollution identification, improves the accuracy and adaptability of the prediction model, and enhances the generalization ability and prediction stability of the method under different reclamation stages and different pollution intensities.

[0011] 3. This invention establishes a coupled prediction model of leaf spectral characteristics and the content of multiple heavy metals in the soil to achieve simultaneous identification of multi-element composite pollution levels in rare earth mining areas. It does not require direct exposure of polluted soil, but can complete large-scale, high-frequency, and automated pollution screening by relying solely on the vegetation canopy spectrum, providing intelligent technical support for early warning of ecological risks in mining areas, assessment of reclamation effectiveness, and precise restoration. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of the overall process of the method for identifying rare earth mineral soil pollution according to the present invention; Figure 2 This is a comparison diagram showing the nonlinear prediction results and linear prediction results of the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and advantages of the invention clearer, the invention will be further described below with reference to embodiments. Obviously, the described embodiments are only some, not all, of the embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without creative effort are within the scope of protection of the invention.

[0014] The main objective of this invention is to address the core challenges of heavy metal pollution in rare earth mining areas, such as the strong concealment of the pollution, the highly heterogeneous spatial distribution, the difficulty of sampling due to vegetation cover, and the weak and easily disturbed mapping relationship between leaf spectra and soil pollution. This invention proposes a leaf hyperspectral driven method and supporting device for soil pollution identification in vegetation-covered areas.

[0015] like Figure 1 As shown, in one aspect, the present invention provides a method for identifying rare earth mineral soil pollution using leaf spectroscopy, comprising the following steps: (S1) Data Acquisition and Preprocessing The original hyperspectral data of the reclaimed vegetation in the field and the corresponding soil sample data were obtained. The original hyperspectral data of the reclaimed vegetation was obtained by collecting the reflectance spectrum data of the vegetation leaves. Due to the influence of solar altitude angle and environmental scattering, the same vegetation sample needed to be collected multiple times for subsequent noise reduction. The soil sample was obtained from the topsoil below the root system of the vegetation where the leaf spectrum was collected. After the soil sample was air-dried, ground, sieved and digested in the laboratory, the heavy metal content of the soil sample was determined by chemical analysis methods, mainly including the content of lead (Pb) and cadmium (Cd).

[0016] The collected reflectance spectral data is input into the data preprocessing module for the following processing: 1. Construct the central spectrum: Calculate the average value of the reflectance spectral data to serve as the reference central spectrum.

[0017] 2. Define the tolerance band: Use 1.5 times the sample standard deviation as the tolerance band.

[0018] 3. Outlier Spectrum Removal: Data falling outside the tolerance band in the reflectance spectrum data are identified as outliers and removed.

[0019] 4. Valid band selection: To avoid strong absorption noise caused by water vapor and CO2 after the 1350nm band in the reflectance spectrum data, this invention selects 350–1350nm as the valid analysis band.

[0020] The processed spectrum obtained after data preprocessing is more representative and has less environmental interference.

[0021] (S2) Design a multivariate empirical mode decomposition module The processed spectrum is input into the Multivariate Empirical Mode Decomposition (MEMD) module, which decomposes the processed spectrum into multiple intrinsic mode function (IMF) components and a residual term:

[0022] in, This indicates the processed spectrum. For IMF components at different frequency scales, For the residual term; m = 1, 2, ..., M, where M represents the total number of IMF components. Low-frequency IMF components are used to characterize changes in nutrient / mineral composition, while high-frequency IMF components are used for noise suppression.

[0023] (S3) Correlation Analysis Multiple IMF components were used as independent variables. The heavy metal content of soil samples was used as the dependent variable. The two variables are input into the correlation analysis module, and each IMF component is combined with its corresponding heavy metal content to form a sample. The correlation coefficient r between each IMF component and the heavy metal content was calculated using the Pearson correlation coefficient formula:

[0024] in, For the sample The two variable values, , This represents the mean of two variables. Indicates the number of samples. The larger the absolute value, the stronger the linear correlation.

[0025] Specific IMF components often contain red-edge shifts or green-peak changes most relevant to heavy metal stress. Based on the calculated r-value, the feature bands of the most correlated IMF components are selected. Feature bands of weakly correlated IMF components are discarded, outputting only the highly sensitive feature bands. Finally, the selected IMF components and heavy metal content are obtained, thereby reducing data dimensionality and improving computational efficiency. Figure 1 The correlation analysis results show that seven IMF components, IMF1-IMF7, were decomposed. The IMF components showed strong correlation in the 650-750nm band. However, the spectral data (OR) that were not decomposed by the multivariate empirical mode decomposition module did not show a correlation between the IMF components and the heavy metal content after correlation analysis.

[0026] (S4) Design a nonlinear prediction module The selected IMF components and their corresponding heavy metal content data were randomly divided into training and test sets in a 7:3 ratio. A nonlinear prediction model was constructed using the Random Forest Regression (RFR) algorithm. The RFR algorithm uses bootstrap sampling to repeatedly extract random feature subsets from the training set, and searches for the optimal split only from the randomly selected feature subsets when splitting each decision tree, thereby training multiple sets of mutually distinct regression trees. (b=1,2,…,B), where B represents the total number of regression trees for any input. The final prediction result of the RFR algorithm is obtained by averaging the outputs of all trees, which can be expressed as:

[0027] It can effectively characterize the nonlinear relationship between leaf spectra and soil heavy metal content, output predicted soil heavy metal content values, and finally use the predicted data to generate a soil heavy metal pollution prediction map.

[0028] Meanwhile, linear and nonlinear models were used to predict the Pb and Cd contents in soil using different spectral processing methods. These methods included OR (Original Orthogonal Spectrum), Mathematical Transformation (MT), and signal decomposition. MT included FD, SD, and CR methods. MEMD was used for signal decomposition. Partial Least Squares Regression (PLSR) was used for the linear model, and RFR was used for the nonlinear model. The prediction results are shown in Tables 1-2 below, and the coefficient of determination (R²) was used. 2 R0 and root mean square error (RMSE) are used as accuracy indicators, where R0 2 The formula for calculating RMSE is as follows:

[0029]

[0030] in, This indicates the actual value of heavy metal content. This indicates the predicted value of heavy metal content.

[0031]

[0032]

[0033] The PLSR and RFR algorithms were used in different spectral processing methods. The results showed that the best results of the two algorithms for R² and RMSE data of Pb and Cd test sets were in the data after MEMD transformation. MEMD can effectively separate noise energy from the real spectral signal components. MEMD showed better anti-interference ability than other methods.

[0034] like Figure 2As shown in Table 1-2 above, after MEMD transformation, the RFR algorithm has an R² of 0.7813 and an RMSE of 8.0850 for the Pb test set, while the PLSR algorithm has an R² of 0.6906 and an RMSE of 12.9980. For the Cd test set, the R² of the RFR algorithm is 0.8007, and its RMSE decreases to 0.0769, while the PLSR algorithm has an R² of 0.7237 and an RMSE of 0.1348. The comparison shows that the RFR algorithm exhibits better accuracy on the MEMD-processed data, demonstrating superior predictive ability compared to the PLSR algorithm. This indicates that under the complexity and high dimensionality of spectral data, nonlinear regression models can provide more accurate and stable prediction results compared to linear models.

[0035] At the application level, this invention enables large-scale, rapid, and repeatable pollution identification in mining areas without the need for intensive manual sampling, significantly reducing labor costs and testing cycles. Simultaneously, it can dynamically capture pollution gradients and trends, providing real-time guidance for assessing the effectiveness of ecological restoration and adjusting recovery strategies in mining areas. In summary, this invention not only improves the accuracy and spatial continuity of pollution identification but also enhances the applicability and robustness of the model under different ecological backgrounds in mining areas, demonstrating significant value for environmental management and engineering practice.

[0036] On the other hand, the present invention provides an apparatus for the aforementioned method of identifying rare earth mineral soil pollution using leaf spectra, comprising a data acquisition device for collecting reflectance spectral data of vegetation leaves, including a portable hyperspectral analyzer and a calibration whiteboard. The portable hyperspectral analyzer is connected to a tripod and a GPS module for synchronously recording geographical location. Before data acquisition, the portable hyperspectral analyzer is mounted on the tripod and adjusted to a vertical observation angle or a specific observation geometry angle, and radiometric correction is performed using a standard whiteboard to eliminate the influence of light intensity variations. The reflectance spectral data of the leaves is collected using the portable hyperspectral analyzer. During data acquisition, due to the influence of solar altitude angle and environmental scattering, multiple data acquisitions of the same vegetation sample are required for subsequent noise reduction. The portable hyperspectral analyzer transmits the collected reflectance spectral data from vegetation leaves to a data processing device. This device includes a multivariate empirical mode decomposition module, a sensitive feature screening device, a correlation analysis module, and a nonlinear prediction device. The nonlinear prediction device is connected to a printing output device, which generates soil heavy metal pollution prediction results from the output of the nonlinear prediction module. The printing output device then produces a detailed pollution monitoring report for use by environmental management departments, effectively solving the technical challenge of the concealed nature and difficulty in direct monitoring of soil pollution in rare earth mining areas.

[0037] The data processing device, the sensitive feature screening device, and the nonlinear prediction device can all be based on the same computer, which is equipped with data processing software that executes the aforementioned modules.

[0038] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and scope of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying rare earth mineral soil pollution using leaf spectroscopy, characterized in that, Includes the following steps: S1: Collect multiple samples of the same vegetation sample to obtain multiple reflectance spectral data of the vegetation leaves, and then input the multiple reflectance spectral data into the data preprocessing module to obtain the processed spectrum; at the same time, collect surface soil samples below the root system of the vegetation sample, and then obtain the heavy metal content in the surface soil samples through chemical analysis methods. S2: Input the processed spectrum into the multivariate empirical mode decomposition module for adaptive decomposition, and decompose the processed spectrum into multiple intrinsic mode function components and a residual term; S3: Input multiple intrinsic mode function components and the heavy metal content into the correlation analysis module to perform correlation analysis, obtain the correlation between the intrinsic mode function components and the heavy metal content, then remove the feature bands of the intrinsic mode function components with weak correlation, select the feature bands of the intrinsic mode function components with the strongest correlation, and finally obtain the selected intrinsic mode function components and the heavy metal content; S4: Input the selected intrinsic mode function components and the heavy metal content into the nonlinear prediction module. The nonlinear prediction module uses the random forest regression algorithm to output the predicted soil heavy metal content value.

2. The method for identifying rare earth mineral soil pollution using leaf spectroscopy as described in claim 1, characterized in that, The implementation steps of the data preprocessing module are as follows: select 350-1350nm of the reflectance spectral data as the effective analysis band, calculate the average value of multiple reflectance spectral data as the reference center spectrum, calculate the standard deviation of multiple reflectance spectral data based on the reference center spectrum, use 1.5 times the standard deviation as the tolerance range, and remove data that are not in the tolerance range from the multiple reflectance spectral data to obtain the processed spectrum.

3. The method for identifying rare earth mineral soil pollution using leaf spectroscopy as described in claim 1, characterized in that, The processed spectrum is input into the multivariate empirical mode decomposition module, which decomposes the processed spectrum into multiple intrinsic mode function components and a residual term: ； in, This represents the spectrum after the processing. These are the eigenmode function components at different frequency scales. The residual term is m = 1, 2, ..., M; M represents the total number of intrinsic mode function components.

4. The method for identifying rare earth mineral soil pollution using leaf spectroscopy as described in claim 1, characterized in that, The correlation analysis module employs Pearson correlation analysis, using multiple intrinsic mode function components as independent variables. The heavy metal content was used as the dependent variable. Each intrinsic mode function component is combined with its corresponding heavy metal content to form a sample. The correlation r between each intrinsic mode function component and the heavy metal content is calculated using the Pearson correlation coefficient formula: ； in, For the sample The two variable values, , This represents the mean of two variables. Indicates the number of samples. The strength of the correlation is related to the absolute value of r.

5. The method for identifying rare earth mineral soil pollution using leaf spectroscopy as described in claim 4, characterized in that, The nonlinear prediction module randomly divides the selected intrinsic mode function components and heavy metal content data into training and test sets in a 7:3 ratio. It then uses a random forest regression algorithm to construct a nonlinear prediction model. This model utilizes bootstrap sampling to repeatedly extract random feature subsets from the training set, and during each decision tree partition, it searches for the optimal partition only from the randomly selected feature subset, thereby training multiple sets of mutually distinct regression trees. (b=1,2,…,B), where B represents the total number of regression trees for any input. The final prediction result of the random forest regression algorithm is obtained by averaging the outputs of all trees, which can be expressed as: 。 6. An apparatus for a method of identifying rare earth mineral soil pollution using leaf spectroscopy as described in any one of claims 1-5, characterized in that, The system includes a data acquisition device for collecting reflectance spectral data of vegetation leaves. The data acquisition device transmits the reflectance spectral data of the vegetation leaves to a data processing device. The data processing device includes a data preprocessing module and a multivariate empirical mode decomposition module. The data processing device is connected to a sensitive feature screening device, which includes a correlation analysis module. The sensitive feature screening device is connected to a nonlinear prediction device, which is connected to a printing output device. The printing output device generates drawings from the output results of the nonlinear prediction module.

7. The apparatus as claimed in claim 6, characterized in that, The data acquisition device includes a portable hyperspectral analyzer and a calibration whiteboard. The portable hyperspectral analyzer is connected to a tripod and a GPS module for synchronously recording geographical locations.

8. The apparatus as claimed in claim 6, characterized in that, The data processing device, the sensitive feature screening device, and the nonlinear prediction device all employ computers.

9. The apparatus as claimed in claim 6, characterized in that, The computer is equipped with data processing software, which executes the operations of the data preprocessing module, the multivariate empirical mode decomposition module, the correlation analysis module, and the nonlinear prediction module.

Images