Crust key metal element content estimation method and system based on hyperspectrum

Abstract

The invention discloses a method and a system for estimating the content of key metal elements of a crusting based on a hyperspectrum, and belongs to the technical field of deep sea mineral detection. The method comprises the following steps: selecting a crusting sample, obtaining hyperspectral data and chemical data of the crusting sample, establishing a point selection data set at a corresponding position, deleting abnormal sample points, and calculating the content of the key metal elements of the crusting sample; the method comprises the following steps: selecting a characteristic wave band of a crusting sample through preprocessing and CARS to obtain a characteristic wave band data set; and based on an MLP neural network model, performing model training through the data set, and constructing an estimation model for estimating the grade of the crusting sample. Compared with the prior art, the method has high accuracy and high predictive capacity, by inverting chemical data of the whole crusting section and applying the scheme for estimating the content of the key metal elements of the crusting, nondestructive grade and abundance analysis of the crusting sample can be achieved, and a reliable scientific basis is provided for seabed detection and multi-metal mineral exploitation.

Description

Technical Field

[0001] The present invention relates to the field of deep-sea mineral exploration technology, and more specifically, to a hyperspectral-based method and system for estimating the content of key metal elements in crusts. The present invention can be extended to provide a sampling basis for estimating the key metal grade contained in deep-sea crusts and for laboratory key metal grade calculations. Background Art

[0002] The deep sea, rich in minerals, energy, and biological resources, represents the world's largest strategic resource base yet to be discovered and developed. Discovered deep-sea metal minerals include polymetallic nodules, cobalt-rich crusts, polymetallic sulfides, and deep-sea rare earth minerals. Crusts are a type of mineral resource rich in various valuable metals, such as cobalt, nickel, and copper. Their exploration and utilization have become a hot topic in global mining research. Cobalt, in particular, is rich in cobalt, a substance whose cobalt content rivals that of corresponding metals on land, earning it the nickname "cobalt-rich crusts." Cobalt, due to its unique physical and chemical properties, including excellent hardness, corrosion resistance, ferromagnetism, and electrochemical properties, is often used in alloy manufacturing, aerospace, battery manufacturing, petrochemicals, and other fields. Although global cobalt reserves are abundant, current exploration and development technologies and economic conditions have limited the development and utilization of these abundant seabed cobalt resources. The geoeconomic evaluation of cobalt-rich crusts is essential for accurately assessing cobalt resources and delineating mining areas. Analyzing the metal content of cobalt-rich crusts using scientific techniques and methods is a crucial step in understanding and evaluating the abundance and grade of the crusts. Commonly used methods for determining elemental content include atomic absorption spectroscopy (AAS), inductively coupled plasma optical emission spectroscopy (ICP-OES), inductively coupled plasma mass spectrometry (ICP-MS), X-ray fluorescence spectroscopy (XRF), electron microprobe microscopy (EPMA), and laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS). Typical methods for determining the metal grade of crusts involve indoor sampling followed by metal content testing. These methods include radioactive detection techniques such as X-ray fluorescence, X-ray transmission electron microscopy, and neutron activation analysis; optical techniques such as laser-induced fluorescence, laser-induced breakdown spectroscopy, near-infrared spectroscopy, and hyperspectral spectroscopy; electromagnetic detection techniques; and machine vision. These methods are all carried out in the laboratory and have high measurement accuracy, but they require professional operation, have high measurement costs, are complex and time-consuming, and have limited application. Summary of the Invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide a hyperspectral-based estimation technology for the content of key metal elements in crusts, aiming to use hyperspectral technology to achieve non-destructive estimation of key metal element content and metal grade estimation of crust samples, providing support for rapid seabed detection and crust mineral mining.

[0004] To achieve the above technical objectives, the present application provides a method for estimating the content of key metal elements in crusts based on hyperspectral analysis, comprising the following steps: Select crust samples, obtain hyperspectral data and EPMA chemical data of crust samples, construct selected point datasets at corresponding locations, and obtain characteristic band datasets through data preprocessing and screening characteristic bands; Based on the MLP neural network model, the model is trained using the characteristic band dataset to construct an estimation model for estimating the grade of crust samples.

[0005] Preferably, in the process of selecting crust samples, samples collected from Magellan Seamount are used as crust samples for method construction.

[0006] Preferably, during the preprocessing process, the hyperspectral data is smoothed and mean filtered to remove noise, and abnormal samples are removed in combination with chemical values. After the data is transformed by the first-order logarithmic differential, a preprocessed data set is obtained.

[0007] Preferably, in the process of removing abnormal samples, the abnormal sample removal process includes the following steps: Calculate the Mahalanobis distance of the original data to obtain the distance between each sample point and other sample points in the data set; Perform PCA dimensionality reduction on the original data to obtain the principal components containing most of the data change information; In the principal component space, the projection of each sample point is calculated, and the outliers are identified based on the degree of deviation of the sample point on the principal component; Sample points with large deviations are considered as outliers, and the outlier sample points are removed from the original data set to obtain a data set.

[0008] Preferably, in the process of performing logarithmic first-order differential transformation to extract the spectral characteristics of the crust sample, the logarithmic transformation is expressed as: Where, is the radiation brightness value corresponding to the wavelength, is the wavelength. The first-order differential transform is expressed as: Where, is the wavelength The logarithmic first-order differential spectrum at ; 、 is the wavelength 、 The spectral radiance data after logarithmic processing at .

[0009] Preferably, in the process of obtaining the characteristic band data set, based on the preprocessed data, the characteristic bands of the crust sample are selected by a competitive adaptive weighting algorithm CARS to generate the characteristic band data set.

[0010] Preferably, in the process of constructing the estimation model, the percentage contents of Co, Cu, Mn and Ni elements in the cobalt-rich crust are predicted by the estimation model.

[0011] Preferably, in the process of using the estimation scheme for the content of key metal elements in crusts, the grade of the cobalt-rich crusts is estimated by means of the estimation scheme.

[0012] The present invention discloses a hyperspectral-based system for estimating the content of key metal elements in crusts, which is used to implement the above-mentioned estimation method, including: A data acquisition module, for acquiring hyperspectral data and chemical data of the crust sample according to the selected crust sample; A data processing module, configured to obtain hyperspectral data and the EPMA chemical data at corresponding positions of the sample, and obtain a preprocessed data set through preprocessing; The estimation module is used to train the model based on the MLP neural network model through the characteristic band data set to build an estimation model for estimating the grade of the crust sample.

[0013] Preferably, the data processing module is further used to perform smoothing and mean filtering on the hyperspectral data to remove noise, and to remove abnormal samples in combination with chemical values, and then perform transformation by logarithmic first-order differential to obtain the preprocessed data set.

[0014] Preferably, the data processing module is further used to select characteristic bands of the crust sample based on the preprocessed data through a competitive adaptive weighting algorithm CARS to generate a characteristic band data set.

[0015] The present invention also provides a storage medium having a computer program stored thereon, and the computer program executes the inversion prediction method for the metal element content of cobalt-rich crusts when running.

[0016] The present invention discloses the following technical effects: The present invention has the advantages of low cost and high efficiency. At the same time, the present invention has high practicality: the model established using hyperspectral data and chemical data can quickly and accurately predict the content of key metal elements in the crust. For resource exploration, this can improve the efficiency of exploration site work, reduce the workload of laboratory analysis, and provide more reliable sampling support for laboratory analysis. Through model prediction, the dependence on a large number of samples can be reduced, and the model can quickly process a large amount of data, reducing manpower and time costs. The present invention can ultimately generate a visualization of the content of different key metal elements, estimate the grade of the crust, and can effectively improve the efficiency and accuracy of resource assessment, which is of great significance for the assessment and mining of mineral resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 This is the basic process for constructing the element inversion prediction model described in the present invention; Figure 2 Schematic diagram of sampling locations of hyperspectral data and chemical data for constructing a model according to the present invention; Figure 3 This is a result diagram after preprocessing the hyperspectral data of the present invention; Figure 4 This is a graph showing the results of the model inversion of Co, Cu, Mn and Ni elements verified by the Co element according to the present invention; Figure 5 is the sample cross-section vector diagram clipped using the ENVI clipping tool described in the present invention; Figure 6 It is a visualization diagram of the percentage content of the cross section of different elements predicted by the model according to the present invention; Figure 7 It is a schematic flow chart of the method described in the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions 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. Obviously, 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 generally described and shown in the drawings here can be arranged and designed in various different 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 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 are within the scope of protection of this application.

[0020] like Figure 1-7 As shown, the present invention provides a hyperspectral-based estimation technology for the content of key metal elements in crusts, which specifically includes the following contents: 1. Data selection and processing: 1.1. Samples: The internal structure of the cobalt-rich crust samples used in the present invention is generally manifested as a three-layer structure on a macroscopic scale: the lower layer is a bright coal seam with high density, bright black, strong metallic luster, and relatively high hardness; the middle layer is loose, porous, gray-black, loose and fragile, and is strongly affected by sedimentation; the upper layer is a relatively dense layer, brown-black, with high hardness and brightness, representing a relatively complete crust growth sequence.

[0021] 1.2 Hyperspectral Data Hyperspectral imaging (HSI) technology uses a spectrometer system to decompose the light incident on the object being measured into different wavelength spectra. This allows for the acquisition of dozens or even hundreds of continuous narrow bands of information per pixel within a spectral range. This image information can characterize the object's appearance, such as size, shape, and color. Due to its high spectral resolution and integrated image and spectrum, hyperspectral technology offers the ability to precisely detect and analyze component content. In recent years, hyperspectral imaging has seen increasing application in various material inspections as a new nondestructive testing technique. The hyperspectral data used in this paper was collected using the ATH1010K hyperspectral imager from Aopu Tiancheng (Xiamen) Optoelectronics Co., Ltd. in China. Its wavelength range is 391 to 1015 nm, with a spectral resolution better than 1.3 nm and 480 spectral bands. The hyperspectral camera is positioned at a height of 15.5 cm from the sample, with a frame rate of 85 and an exposure time of 8000.

[0022] 1.3 Chemical data: This study used a JEOL (Japan Electron Microscope) JXA-8230 electron probe X-ray microanalyzer to obtain chemical data from crust samples. Equipped with a five-channel spectrometer, the JXA-8230 is capable of high-resolution, high-precision, in-situ, non-destructive, and rapid determination of the B–U elemental content in solid samples. It can perform both line and surface scanning analyses of elemental content, as well as backscattered electron and secondary electron image analysis. The electron probe used in the experiment had an accelerating voltage of 15 kV, a beam current of 20 nA, and an electron beam spot diameter of 3 μm.

[0023] 2. Methods: Hyperspectral imaging technology can capture subtle changes in the composition of a material by measuring the reflection or absorption spectrum of an object at different wavelengths. Since the chemical composition of a substance affects its spectral characteristics, there is an inherent physical correlation between these spectral characteristics and the chemical composition, which makes it possible to parse these relationships through machine learning models to achieve prediction or inversion of chemical composition. Each pixel in a hyperspectral image contains spectral information of the material at that point, which indirectly reflects its chemical and physical properties. Key features that help characterize the chemical composition, such as absorption peaks and reflection peaks, are extracted from hyperspectral data. On this basis, machine learning techniques (such as linear regression, decision trees, neural networks, etc.) can be used to train models based on known chemical compositions (labels), establish a mapping relationship from hyperspectral features to chemical compositions, verify the accuracy of the prediction results through experimental data, and adjust and optimize the model based on feedback. The present invention uses the coefficient of determination (R 2 ), root mean square error (RMSE), relative analytical error (RPD) and mean absolute error (MAE) were used to evaluate the model. 2 The closer it is to 1, the higher the model accuracy and the stronger the estimation ability; the lower the RMSE, the higher the model accuracy, and vice versa, the worse the model accuracy; RPD can further evaluate the quality of the constructed model. When RPD>2, it means that the model can predict the sample well. When 1.4<RPD<2, it means that the model can only roughly estimate the sample. When RPD<1.4, it means that the established estimation model cannot estimate the sample. In machine learning, MAE directly reflects the mean absolute deviation between the predicted value and the actual value. It is often used for performance evaluation in regression tasks. The smaller the MAE value, the more accurate the model prediction and the smaller the prediction error. Their calculation formulas are as follows; in, is the sample size, Indicates the The actual value of the samples, represents the average value of the actual sample values, Indicates the The predicted value of the sample, Represents the absolute difference between the actual value and the predicted value. The parameters of different models can be modified as required to ultimately establish a more ideal model. Finally, the trained model is applied to new hyperspectral data to predict its chemical composition. Based on the proposed estimation scheme for the content of key metallic elements in crusts, the grade of cobalt-rich crusts is estimated.

[0024] 3. Application: The application steps of the present invention are as follows: (1) data quality assessment and screening, and construction of corresponding location selection point data sets; (2) spectral data preprocessing using methods such as multivariate scatter correction (MSC) and logarithmic first-order derivative (LOG-FD); (3) screening of characteristic bands; (4) establishment of inversion models for key metal elements such as Co using the screened hyperspectral characteristic band data sets and chemical data sets; (5) application of the model inversion to predict the content of key metal elements such as Co, Fe, Mn, Cu, and Ni in the crust; (6) application of the estimation scheme for the content of key metal elements in the crust to analyze the metal element content of the crust and estimate its grade. The basic process of the early construction of the Co element inversion model is as follows: Figure 1 .

[0025] 3.1 Data Processing The present invention selects the cross-section position (the red straight line position, such as Figure 2 (a) Hyperspectral data is used as spectral data for constructing the inversion model. The data point spacing is 0.02 cm. The number of electron probe chemical element sampling points is about 3 times that of the hyperspectral data. The sampling spacing is 0.06 cm. The sampling positions are the corresponding profiles ( Figure 2 (b)) Sampling points,.

[0026] 3.2 Data preprocessing: First, the spectral curves are de-noised using a combination of convolution smoothing and mean filtering. The spectral data is then combined with the chemical data to remove outliers. This outlier removal method is as follows: First, the Mahalanobis distance is calculated on the raw data to determine the distance between each sample point and other sample points in the dataset. Then, PCA dimensionality reduction is performed on the raw data to obtain the principal components that contain the majority of the data variation information. In the principal component space, the projection of each sample point is calculated, and outliers are identified based on the degree of deviation of the sample point from the principal component. Sample points with large deviations are considered outliers and removed from the original dataset for subsequent modeling and analysis.

[0027] The data after elimination are used as the original spectral data of the sample, and then the original spectral data are preprocessed in five forms, namely, multivariate scatter correction (MSC), standard normal variable transformation (SNV), standardization (Nor), first-order differential (FD) and logarithmic first-order differential (LOG-FD). After this operation, the purpose of improving the signal-to-noise ratio and enhancing the spectral characteristics can be achieved.

[0028] Multivariate scattering correction (MSC) is a method used to address the multivariate scattering effect in spectral data. In spectroscopy, MSC refers to the multiple scattering of light within a sample, which distorts the absorption peaks and valleys in the spectrum, thereby affecting the accuracy of quantitative analysis. MSC aims to eliminate this multivariate scattering effect, making the spectrum more accurately reflect the characteristics of the sample. Corrected spectra typically more accurately reflect the sample's characteristics, with clearer positions and shapes of absorption peaks and valleys.

[0029] Standard normal variate crossover is used to eliminate spectral intensity variations caused by factors such as light scattering and path length changes in spectral data. After standard normal variate crossover, the spectral intensity distribution of each wavelength point of the spectral data will be more consistent, and the shape difference between the peak and the valley will be reduced, which can improve the comparability of the data and the accuracy of the analysis.

[0030] Spectral data normalization is a specific data preprocessing technique used to scale the spectral intensity values at each wavelength of spectral data to the same scale range, eliminating dimensional and numerical range differences between different wavelengths. Its main purpose is to eliminate the differences in spectral intensity at different wavelengths, ensuring that the spectral characteristics at different wavelengths have a consistent impact on the model. This eliminates the differences in spectral intensity at different wavelengths and improves the performance and stability of the model.

[0031] The first-order differential can largely eliminate linear and atmospheric background noise, and can also enhance subtle changes in the slope of the spectral curve, thereby suppressing useless information and highlighting information of interest. The formula is as follows: Where, is the wavelength First-order differential spectral data at ; 、 is the wavelength 、 Spectral radiance value data at .

[0032] The first-order differential spectrum actually represents the rate of change of the original spectral curve. When the first-order differential value is positive, it means that the original spectral reflectance is increasing, and when it is negative, it means that the original spectral reflectance is decreasing, which can highlight the changes caused by subtle curvature differences.

[0033] The logarithmic first-order difference combines the advantages of logarithmic transformation and first-order difference, and can effectively enhance spectral features, especially in revealing weak absorption peaks, combating environmental noise and instrument instability. Specifically, the logarithmic transformation not only balances the dynamic range of the spectrum but also enhances the visibility of spectral details by compressing the high-intensity region and expanding the expression of the low-intensity region. The subsequent first-order difference step focuses on eliminating baseline drift, highlighting the rapidly changing part of the spectral curve, further refining the characteristic structure, and improving the spectral contrast. This series of operations jointly promotes the highlighting of key information in the spectral data, laying a solid foundation for subsequent feature recognition, classification and model building.

[0034] The data after elimination are used as the original spectrum data of the sample, and the original spectrum data (ORG) curve and the spectrum curve obtained by preprocessing in five transformation forms are plotted. The results are as follows Figure 3 .

[0035] Depend on Figure 3 The original spectral curves exhibit distinct peaks and troughs within certain wavelength ranges. After multivariate scattering correction, standard normal variate transformation, and standardization, the spectral curves exhibit similar trends, with more pronounced peaks and troughs. Furthermore, the spectral curves for the bedrock and shell sampling points exhibit two distinct distribution patterns. The spectral curves after first-order differential and logarithmic first-order differential processing exhibit similar trends, both exhibiting prominent peaks and troughs. The logarithmic first-order differential results in smoother and more uniform spectral curves than the first-order differential.

[0036] 3.3. Screening characteristic bands: The competitive adaptive weighting algorithm (CARS) and the iterative information variable retention algorithm (IRIV) were applied to the five preprocessed data and the raw spectral data to identify characteristic wavelength bands for the sample. This operation effectively reduces data dimensionality, and the selected wavelength bands can also be used as characteristic wavelength bands for the chemical element. Based on the "survival of the fittest" principle in Darwinian theory, CARS is a characteristic wavelength selection method based on Monte Carlo sampling and partial least squares (PLS) regression coefficients. Adaptive weighted sampling is used to retain wavelength variables with large regression coefficient weights in the PLS model, and the PLS model is then re-established. After multiple sampling cycles, the subset with the lowest root mean square error (RMSECV) of cross-validation is selected as the optimal characteristic wavelength variable subset. IRIV randomly combines variables multiple times and establishes a PLS model. The change in prediction error is observed when each variable is included in the model. Based on cluster analysis, variables are classified into strong, weak, uninformative, and nuisance variables. After analyzing each variable individually and removing uninformative and nuisance variables, multiple iterations are performed until the remaining variables are all strong and weak informative. The variables that are finally retained are the desired characteristic variables.

[0037] 3.4 Predictive Modeling Five types of predictive modeling were selected, including linear modeling (partial least squares regression (PLSR)), and nonlinear modeling (support vector machines (SVM), random forest (RF), multilayer perceptron (MLP), and gradient boosting decision tree (GBDT). The accuracy of the models was verified using validation metrics including coefficient of determination, root mean square error (RMS), relative analytical error, and mean absolute error (MAE). PLSR considers the correlation between predictor and response variables to identify latent variables that are highly correlated with both. This paper aims to address regression problems with multicollinearity and high-dimensional data. SVM is a machine learning algorithm based on statistical learning theory. During modeling, it transforms low-dimensional data into a high-dimensional space using a kernel function to construct an optimal classification hyperplane, capable of achieving optimal solutions with small sample sizes. RF is an ensemble learning method primarily used for classification and regression tasks. MLP is a feedforward artificial neural network consisting of an input layer, several hidden layers, and an output layer. Each layer consists of multiple neurons, which interact with each other through weighted connections. GBDT is a powerful ensemble learning algorithm that corrects the prediction error of the previous tree by gradually adding decision trees. It is suitable for processing complex relationships of structured data and is widely used in classification and regression tasks.

[0038] According to the modeling results, the modeling results of the LOG-FD preprocessing method are better than those of other methods in the data preprocessing step; CARS is better than IRIV in screening characteristic bands, indicating that the CARS characteristic band screening method has great advantages in this invention; comparing the five types of predictive modeling, it is found that the modeling results using the MLP method are stable and excellent. In summary, the LOG-FD-CARS-MLP combination achieved the best results, R 2 is 0.9858, and the RMSE is 0.0267×10 -2 , the RPD value is 8.4749, and the MAE value is 0.0164, indicating that the model under this combination can better predict chemical data, and the model has strong stability and accuracy.

[0039] 3.5. Element inversion results display: The present invention further expands its application to the spectral analysis of Cu, Mn and Ni metal elements. Figure 4 As shown, according to the LOG-FD-CARS-MLP combination, Cu ( Figure 4 (a) ), Cu ( Figure 4 (b)), Mn ( Figure 4 (c)), Ni ( Figure 4 (d) Element inversion results display: R 2 All were greater than 0.98, and RMSE was less than 0.97×10 -2 , RPD values are all greater than 8.47, and MAE values are all less than 0.68, indicating that the model under this combination also achieves excellent performance in inverting other metal elements, providing a scientific basis for exploring the element distribution laws in complex systems.

[0040] 3.6. Prediction results of five element profiles: Extract sample hyperspectral data, such as Figure 5 The data size is 217×330×480, representing the spectral information of 480 bands with 217 data points in the vertical direction and 330 data points in the horizontal direction of the cropped image. Among them, a total of 17,264 data points have a full band data value of 0 (i.e., the black background part). The model constructed according to the present invention predicts the percentage content of Co, Cu, Mn, and Ni in the entire cross section, as shown in Figure 2. Figure 6 As shown, the model can effectively distinguish between the cross-section and shell of a crust sample. The predicted results show that the percentage of metal elements is lower in the bedrock and higher in the shell, which is consistent with the actual chemical content distribution of the crust sample, demonstrating the high accuracy of the model established by this invention. Furthermore, the model can also predict the specific percentage of metal elements at each data point on the cross-section, which is of great significance for analyzing specific crust layers and estimating grade.

[0041] 3.7. Estimation of key metal element content in crusts: The grade of a mineral represents the mass percentage (wt%) of a particular element or mineral in the entire ore. The EPMA data measured in this invention represents the percentage of a single metallic element in the crust sample. The model predicts the percentage of each metallic element at each data point across the entire profile. The average value of the data points across the entire sample profile yields the average elemental percentage, or grade, for the crust sample. The formula for Co elemental grade is as follows: in, The Co elemental content of the i-th data point is expressed as a percentage, and n represents the total number of data points on the profile. Using the above estimation scheme, the Co elemental grade of the crust sample can be determined. The calculation formulas for Cu, Mn, and Ni elemental grades are similar.

[0042] In practical applications, during the preliminary survey phase of the research sea area, optical equipment such as hyperspectral data can be installed on the survey platform. If there are five mining areas with cobalt-rich crusts in the research area, representative cobalt-rich crust samples can be selected from each of the five mining areas, and their metal element content can be inverted using the collected hyperspectral data. The grades of elements such as Co, Cu, Mn, and Ni can be calculated according to the grade estimation scheme, thereby knowing the resource volume of different mining areas. This has important guiding significance for the subsequent assessment and development of mineral resources. Compared with previous methods for evaluating cobalt crust resources, this method has the advantages of being fast, in-situ, and economical. At the same time, because the grade of the crust is calculated based on the metal content of the profile, its results are more reliable and scientific than previous geological experience estimates.

[0043] It can be seen that for the hyperspectral inversion of Co elements in cobalt-rich crusts, the optimal preprocessing method is LOG-FD, the optimal characteristic band screening method is CARS, and the optimal modeling type is MLP, that is, the LOG-FD-CARS-MLP technical solution. This solution can achieve the inversion of the Co element content of the entire profile from 36 characteristic bands and achieve good results. It can quickly and accurately invert the element content within 1 minute, and the amount of inversion data reaches 71,610 (data accuracy 0.20 mm).

[0044] An embodiment of the present invention further provides a storage medium having a computer program stored thereon, which executes a method for inversion prediction of element content in cobalt-rich crusts when the computer program is run.

[0045] The present invention makes it possible to obtain the content of key metal elements in crusts in a non-destructive, rapid and wide-ranging testing environment. It is expected to be matured and improved so that it can be applied in practical work.

[0046] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0047] In the description of the present invention, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of the technical features indicated. Therefore, a feature specified as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, "plurality" means two or more, unless otherwise specifically defined.

[0048] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for estimating the content of key metal elements in crusts based on hyperspectral analysis, characterized in that: The following steps are involved: Select a crust sample, obtain hyperspectral data and electron probe microscopy (EPMA) chemical data of the crust sample, delete abnormal data points and establish a selected point dataset at the corresponding location, select characteristic bands through preprocessing and CARS method, and obtain a characteristic band dataset of key metal elements in the crust; Based on the MLP neural network model, model training is performed using the characteristic band data set to construct an estimation model for estimating the content of key metal elements in the crust.

2. The method for estimating the content of key metal elements in crusts based on hyperspectral analysis according to claim 1, characterized in that: In the process of selecting crust samples, the Magellan Seamount sample was used as the crust sample.

3. The method for estimating the content of key metal elements in crusts based on hyperspectral analysis according to claim 2, characterized in that: During the preprocessing process, the hyperspectral data is smoothed and mean filtered to remove noise, and combined with the chemical data to remove abnormal samples, and then transformed by the first-order logarithmic differential to obtain the preprocessed data set.

4. The method for estimating the content of key metal elements in crusts based on hyperspectral analysis according to claim 3, characterized in that: In the process of removing abnormal samples, the abnormal sample removal process includes the following steps: Calculate the Mahalanobis distance of the original data to obtain the distance between each sample point and other sample points in the data set; Perform PCA dimensionality reduction on the original data to obtain the principal components containing most of the data change information; In the principal component space, the projection of each sample point is calculated, and the outliers are identified based on the degree of deviation of the sample point on the principal component; Sample points with large deviations are considered as outliers, and the outlier sample points are removed from the original data set to obtain the processed data set.

5. The method for estimating the content of key metal elements in crusts based on hyperspectral analysis according to claim 4, characterized in that: In the process of extracting the spectral characteristics of crust samples by performing logarithmic first-order differential transformation, the logarithmic transformation is expressed as: Where, is the radiation brightness value corresponding to the wavelength, is the wavelength. The first-order differential transform is expressed as: Where, is the wavelength The logarithmic first-order differential spectrum at ; 、 is the wavelength 、 The spectral radiance data after logarithmic processing at .

6. The method for estimating the content of key metal elements in crusts based on hyperspectral analysis according to claim 5, characterized in that: In the process of obtaining the characteristic band data set, the characteristic bands of the crust sample are selected based on the preprocessed data through a competitive adaptive weighting algorithm (CARS) to generate the characteristic band data set.

7. The method for estimating the content of key metal elements in crusts based on hyperspectral analysis according to claim 6, characterized in that: In the process of constructing the estimation model, the percentage contents of Co, Cu, Mn and Ni elements in the cobalt-rich crust are predicted by the estimation model. In the process of applying the estimation scheme for the content of key metal elements in crusts, the grade of cobalt-rich crusts is estimated by means of the estimation scheme.

8. A hyperspectral-based system for estimating the content of key metal elements in crusts, characterized in that: include: A data acquisition module, configured to obtain hyperspectral data and chemical data of a selected crust sample; A data processing module, configured to obtain hyperspectral data and the EPMA chemical data at corresponding positions of the sample, and obtain a preprocessed data set through preprocessing; An estimation module is used to perform model training based on an MLP neural network model using the characteristic band data set to construct an estimation model for estimating the grade of the crust sample.

9. The hyperspectral-based system for estimating the content of key metal elements in crusts according to claim 8, characterized in that: The data processing module is further used to perform smoothing and mean filtering on the hyperspectral data to remove noise, and to eliminate abnormal samples in combination with chemical values, and then perform transformation through logarithmic first-order differential to obtain the preprocessed data set.

10. The hyperspectral-based system for estimating the content of key metal elements in crusts according to claim 9, characterized in that: The data processing module is further configured to select characteristic wavebands of the crust sample based on the preprocessed data by using a competitive adaptive weighting algorithm (CARS) to generate the characteristic waveband data set.

Images