Multi-source remote sensing rare earth ore prospecting method and system fused with pteridophyte ore showing

Abstract

The invention discloses a multi-source remote sensing rare earth ore prospecting method and system fused with pteridophyte ore showing, and belongs to the technical field of mineral resource exploration. Weathering crust type rare earth ore mineralization factor remote sensing features are obtained through multi-source remote sensing data, a potential area is preliminarily delineated, an unmanned aerial vehicle carrying a VNIR camera is deployed to obtain a high-resolution image, and the high-resolution image is subjected to image mapping; and comprehensively extracting fern distribution information according to texture, form and spectral indexes, and down-sampling a fern coverage result to a resolution consistent with that of other features to form a fern coverage layer. And constructing a multi-source remote sensing matrix fusing the fiddlehead coverage map layer and the mineralization factor characteristics, and predicting the rare earth mineralization probability and optimizing the potential region selection through sample training and global reasoning. According to the method, the positioning precision and exploration efficiency of the weathering crust rare earth ore are remarkably improved through coupling of the biological indicator factor (pteridophyte) and the traditional rock-mineral-structure-landform ore showing factor.

Description

Technical Field

[0001] The present invention belongs to the technical field of mineral resource exploration, and specifically relates to a multi-source remote sensing rare earth ore prospecting method and system integrating fern ore indication, mainly for remote sensing mineral exploration of weathered crust rare earth ore. Its core innovation lies in taking ferns rich in rare earth elements (such as Dicranopteris dichotoma, Blechnum orientale, etc., with the total rare earth content reaching percentage level) as biological ore indication marks, combining ore-forming factors such as rock-mineral-structure-geomorphology, and implementing efficient and accurate prediction and positioning of potential rare earth ore areas through multi-source remote sensing information recognition and intelligent algorithms. Background Art

[0002] Weathered crust rare earth ore is mostly formed in the weathered residual layers of granite, carbonatite-alkali rock or acidic volcanic rock. Rare earth elements (REE) are often in the form of ion exchange states such as Nd 3+ , Dy 3+ , Y 3+ and are hosted in altered clay minerals such as kaolinite, illite, montmorillonite, etc. Therefore, existing exploration methods often rely on the identification of these clay minerals to determine potential mineralized areas. However, when relying on geological mapping or a single remote sensing index, it is easy to cause misjudgment or missed judgment due to factors such as ground coverage and weak spectra, and it is difficult to quickly screen large areas of weathered crust mineralization potential areas due to limited investigation scope and high cost.

[0003] In recent years, studies have found that some ferns (such as Dicranopteris dichotoma, Blechnum orientale, etc.) can adsorb and enrich rare earth elements up to the percentage level in the weathered crust environment, with significant ore indication characteristics, providing new possibilities for biological ore indication. However, only biological indication is not enough to comprehensively reveal the mineralization distribution, and ore indication factors such as the lithology of the parent rock, clay mineral alteration, terrain undulation, and tectonic fracture zones need to be considered simultaneously. Through multi-source remote sensing data fusion, integrating the above ore-forming factors and fern distribution into the prediction model of machine learning algorithms (especially deep learning) can take into account local biological indication at the regional scale and significantly improve the efficiency and accuracy of weathered crust rare earth ore prospecting.

[0004] Therefore, there is an urgent need to develop a multi-source remote sensing prospecting technology that synchronously integrates the biological ore indication characteristics of ferns rich in rare earth elements with traditional ore indication factors such as rock-mineral-structure-geomorphology, and uses advanced machine learning algorithms to achieve high-precision identification and positioning of weathered crust rare earth ore. Summary of the Invention

[0005] The present invention provides a multi-source remote sensing rare earth ore prospecting method and system integrating fern ore indication, which uses the remote sensing characteristics of ore-forming factors such as "rock-mineral-structure-terrain" for preliminary screening of large-scale weathered crust mineralization, and then conducts refined biological ore indication identification in key areas, so as to efficiently predict and locate weathered crust rare earth ore.

[0006] A multi-source remote sensing rare earth ore prospecting method integrating fern indication of ore deposits, characterized by comprising the following steps:

[0007] (1) Obtain multi-source remote sensing data, perform atmospheric and geometric correction on the images, and extract ore-forming factor indication factors, including rock-mineral, rare earth tracer elements, topographic and geomorphic features, and tectonic fracture zone information;

[0008] The multi-source remote sensing data includes optical images, radar images, and DEM data;

[0009] The optical images include Sentinel-2, Landsat-8, ASTER, and GF5;

[0010] The radar images include ALOS PALSAR or GF-3. The ALOS PALSAR or GF-3 data is used to identify topographic and geomorphic factors and tectonic fracture zone factors, improving the identification accuracy of weathered crust rare earth ore areas.

[0011] The extraction methods of ore-forming factor indication factors include:

[0012] (1.1) Combine band ratio and K-means algorithm to obtain rock-mineral factors;

[0013] (1.2) Use the Constrained Energy Minimization (CEM) algorithm to extract altered mineral abundance information;

[0014] (1.3) Use Principal Component Analysis (PCA) to extract rare earth tracer element characteristics;

[0015] (1.4) Analyze the slope, aspect, cutting depth, and terrain variation coefficient of the DEM to obtain topographic factors;

[0016] (1.5) Use Sobel and Laplacian operator thresholds to detect tectonic fracture zone information.

[0017] (2) After initially delineating the potential areas of weathered crust rare earth ore areas based on ore-forming factor indication factors, use an unmanned aerial vehicle (UAV) equipped with a visible light-near infrared camera to fly at low altitude over the potential areas, collect high-resolution images, and identify the distribution of ferns through texture, morphology, and spectral indices;

[0018] During the low-altitude flight of the UAV, it also includes arranging ground control points (GCPs) and performing orthophoto mosaicking to align the UAV images with the multi-source remote sensing data in the same coordinate system.

[0019] Identifying the distribution of ferns includes the following sub-steps:

[0020] (2.1) Extract the color and texture features of visible light images;

[0021] (2.2) Distinguish ferns from other vegetation using spectral indices (NDVI or red-edge index);

[0022] (2.3) Calculate the fragmentation or roundness morphological characteristics of fern leaves based on object-oriented segmentation method;

[0023] (2.4) Output the fern distribution map through a machine learning classifier (SVM or RF) or a lightweight deep learning network.

[0024] (3) Downsample the obtained fern distribution to the same resolution as other remote sensing features to form a fern coverage layer;

[0025] The downsampling step in step (3) is carried out using neighborhood statistics, specifically including:

[0026] (3.1) Select the grid size corresponding to the main remote sensing data;

[0027] (3.2) Count the proportion of fern pixels in the centimeter-level image within each grid;

[0028] (3.3) Convert the proportion result into a fern coverage layer between 0 and 1.

[0029] (4) Integrate the fern coverage layer with ore-indicating factors to construct a multi-source feature matrix;

[0030] (5) Input the feature matrix into a deep learning model for training and prediction, output a rare earth mineralization probability map and select high-potential areas.

[0031] The deep learning model is a convolutional neural network CNN. The input of CNN is a multi-source feature matrix: rock-mineral factors, rare earth tracer elements, topographic structure factors, and fern distribution information; the output is a mineralization probability map.

[0032] The training steps of the deep learning model include:

[0033] (5.1) Collect known mineralized points as positive samples and collect non-mineralized or background areas as negative samples;

[0034] (5.2) Register the multi-source feature matrix to correspond the sample labels with the pixel features;

[0035] (5.3) Set the network structure and hyperparameters, and use the cross-entropy loss function for iterative optimization;

[0036] (5.4) Evaluate the model performance based on the confusion matrix and AUC evaluation index and save the optimal model parameters.

[0037] When applying the trained deep learning model to the entire feature matrix, the mineralization probability of each pixel is output, potential weathered crust rare earth ore areas are screened based on the probability threshold, and final confirmation is carried out in combination with ground verification.

[0038] Its core includes the following:

[0039] (1) Extraction of multi-source remote sensing information of ore-indicating factors

[0040] Collect multi-source optical remote sensing data including Sentinel-2, Landsat-8, ASTER, GF5, etc. and DEM. After completing atmospheric correction and geometric correction, use methods such as band ratio, principal component analysis, and altered mineral identification to obtain remote sensing characteristics of ore-forming factors such as rocks-minerals-structures-geomorphology, and preliminarily delineate the weathered crust rare earth ore areas.

[0041] (2) Drone fern monitoring

[0042] In the potential area, fly low with a drone equipped with a visible light-near infrared camera to obtain high-resolution images. Use methods such as texture (gray-level co-occurrence matrix), morphology (object-oriented segmentation), and spectral indices (NDVI, RedEdgeIndex) to identify ferns (such as Dicranopteris dichotoma, Blechnum orientale, etc.) that can enrich rare earth elements, downsample them to the same grid resolution as other remote sensing characteristics, and generate a fern coverage map. Comprehensively discriminate the distribution of ferns to provide high-confidence biological ore-indicating factors for the deep learning model.

[0043] (3) Deep learning prediction

[0044] Construct a multi-source feature matrix with fern coverage and information such as rock-mineral factors, abundance of altered minerals, rare earth element absorption characteristics, terrain factors, and tectonic fracture zones, use convolutional neural network (CNN) for training and inference, output a mineralization probability map and optimize high-potential areas, so as to efficiently and accurately complete the identification and positioning of weathered crust rare earth ores.

[0045] A rare earth ore prospecting system for implementing the above method, the system composition includes:

[0046] Multi-source data processing unit: used to acquire and manage multi-source remote sensing images (optical, DEM, radar, etc.), and perform preprocessing (atmospheric correction, geometric correction, cloud cover filtering, image cropping, etc.) and extract ore-indicating factors of ore-forming factors;

[0047] Drone aerial photography unit: includes a drone equipped with a visible light-near infrared camera, ground control points (GCP) and flight planning software for low-altitude image acquisition and fern identification in the potential area;

[0048] Feature Fusion and Deep Learning Unit: Synthesize a multi-source feature matrix by integrating fern coverage (FernCover) with ore-indicating factors such as lithology, minerals, structures, and topography, and perform CNN training and inference in a high-performance environment such as GPU to output a mineralization probability map;

[0049] Result Visualization and Decision-making Unit: Used to generate a mineralization probability map and screen high-potential areas to guide the prospecting and exploration of weathered crust rare earth ores. Use a GIS platform or WebGIS to display layers such as mineralization probability, fern coverage, and lithological background, altered minerals, geomorphology, and tectonic fracture zones for users to interactively query and guide subsequent on-site verification and drilling deployment.

[0050] In summary, through the above methods and systems, the present invention can quickly delineate potential enrichment areas of weathered crust rare earth ores in a large area, integrate fern biological ore-indicating information and metallogenic factors in key areas, and use deep learning models to improve the accuracy and efficiency of exploration, reducing blind ground surveys and drilling costs. Description of the Drawings

[0051] The drawings described herein are used to provide a further understanding of the present invention, form a part of the present invention, and the schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0052] Figure 1 is a flow schematic diagram of the present invention;

[0053] Figure 2 is an indication diagram of lithology classification and ore-hosting parent rock distribution provided for the embodiment;

[0054] Figure 3 is an indication diagram of altered mineral distribution provided for the embodiment;

[0055] Figure 4 is an indication diagram of the band indicating rare earth minerals hosted by PCA analysis provided for the embodiment;

[0056] Figure 5 is an indication diagram of the topographic and geomorphic features of the weathered crust type deposit provided for the embodiment;

[0057] Figure 6 is an indication diagram of the tectonic fracture zone information detected by edge detection using a convolution kernel provided for the embodiment;

[0058] Figure 7 is an indication diagram of rare earth potential areas under multi-factor constraints provided for the embodiment;

[0059] Figure 8 is a schematic diagram of a fern plant sample provided for an embodiment of the present invention;

[0060] Figure 9Schematic diagram of the modules of the rare earth ore prospecting system provided by an embodiment of the present invention. Detailed implementation manners

[0061] As Figure 1 shown in the process, which includes the following steps:

[0062] Step 1: Multi-source remote sensing data analysis (preliminarily obtaining the potential areas of weathered crust rare earth mineralization)

[0063] In this step, multi-source remote sensing data (optical remote sensing, microwave remote sensing, thermal infrared remote sensing, hyperspectral remote sensing, etc.) are mainly used. Through a series of methods such as image processing, feature extraction, classification and recognition, and multi-factor constraint, the potential areas of weathered crust rare earth mineralization are identified. This method conducts systematic analysis from multiple angles such as lithological background, altered minerals, rare earth key tracer elements, topography and geomorphology, and tectonic fracture zones, and preliminarily delineates high-potential sample areas.

[0064] 1.1 Multi-source data acquisition and preprocessing

[0065] 1.1.1 Multi-source data acquisition

[0066] 1) Optical remote sensing data

[0067] Landsat-8 OLI / TIRS: The data provider is the United States Geological Survey (USGS). The resolution of this data in the visible-light short-wave infrared (VNIR-SWIR) is 30 m, and the thermal infrared (TIR) is 100 m (resampled to 30 m). It is used to identify the large-scale rock mass characteristics and hydrothermal alteration areas in the study area.

[0068] ASTER: ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) was jointly developed by the Japan Aerospace Exploration Agency (JAXA) and NASA of the United States and launched on the Terra satellite on December 18, 1999. This data is available for global users to download on the NASA EARTHDATA platform. It covers the VNIR, SWIR, and TIR bands and can be used for lithology identification, vegetation index calculation, and mineral alteration analysis. The TIR band is sensitive to the thermal emission characteristics of different rocks and helps to distinguish alkaline rocks, granite, and gabbro.

[0069] Sentinel-2B MSI provides high spatial resolution and can be used to extract the spectral characteristics of ferns and extract key REE tracer elements. The red-edge bands (705nm, 740nm, 783nm) can enhance vegetation monitoring capabilities and also monitor the absorption characteristics of the key tracer element Nd (at 744, 802, and 865nm bands). The SWIR band can identify altered minerals such as kaolinite and illite. The data provider is the European Space Agency (ESA). The spatial resolution of the VNIR band (443 - 865nm) of this data is 10m, and the spatial resolution of the SWIR band (1373 - 2190nm) is 20m.

[0070] The hyperspectral data of the High-Resolution Five (GF-5) AHSI provides ultra-high spectral resolution and can accurately extract the characteristic absorption peaks of the rare earth tracer element Nd. It is suitable for the identification of the rare earth tracer element Nd and can be used as a supplement and verification for Sentinel-2B MSI to improve the accuracy of prospecting. The data provider is the China National Space Administration. The spectral resolution: visible light (5nm), short-wave infrared (10nm), and the spatial resolution: 30m. 3+

[0071] 2) Radar remote sensing data

[0072] The High-Resolution Three (GF-3) SAR uses multi-polarized SAR data, which can penetrate vegetation to obtain surface structure information and identify fault zones. Combining with a direction filter enhances the ability to identify linear structures. The data provider is the China National Space Administration. The resolution of this data: HH+HV polarization mode, resolution 10m, swath width 100km.

[0073] 3) Digital Elevation (DEM) data

[0074] ALOS PALSAR SAR is used to generate DEM. The final DEM data resolution is about 12.5m. It is suitable for the micro-topography identification of weathered crust mining areas. The data provider: Japan Aerospace Exploration Agency (JAXA), resolution single polarization (FBS): 10m, multi-polarization (FBD): 20m, polarization mode (PLR): 30m.

[0075] NASA DEM is used to fill data gaps and complement ALOS PALSAR SAR for calculating weathered crust development indices (slope, aspect, terrain undulation, cutting depth, etc.). Combining with mineralization factors, it delineates the most suitable weathered crust rare earth metallogenic areas. The resolution, NASA DEM: 30m.

[0076] 1.1.2 Data preprocessing

[0077] (1) Radiometric correction

[0078] Atmospheric correction is performed using FLAASH. The influence of atmospheric scattering and aerosols is removed to restore the true spectral reflectance of the surface. The spectral responses of multi-source data are unified to improve data consistency.

[0079] (2) Geometric correction

[0080] Ensure the spatial matching of multi-source remote sensing data to avoid the influence of error superposition on the analysis results. A unified coordinate system is adopted for all data. Among them, GF5 needs to be orthorectified, and the reference image is Landsat8. The geometric calibration elevation of GF3 data uses ALOS PALSAR DEM.

[0081] (3) Image cropping

[0082] Use the vector boundary of the study area to crop the image. Remove the redundant data outside the study area, reduce the calculation amount, and improve the processing efficiency.

[0083] (4) Cloud cover filtering

[0084] Adopt the interest area threshold screening and mask technology to remove clouds and shadows, avoid cloud interference, and ensure data integrity.

[0085] Through the above steps, a high-quality, multi-source remote sensing data set (radiometric correction, geometric correction, cropping, cloud cover filtering have been performed) is output to ensure the accuracy of subsequent analysis. DEM data, SAR data, and spectral data are registered with each other, providing a data basis for the identification of weathered crust rare earth mineralization potential areas.

[0086] 1.2 Lithological background feature extraction module

[0087] This module needs to use the ASTER thermal infrared band and Landsat8 remote sensing data. Combining spectral indices, principal component analysis (PCA), and support vector machine methods, the main lithological units in the study area are identified to provide lithological background constraints for mining area identification.

[0088] 1.2.1 Band ratio method (BR)

[0089] The band ratio method can enhance the spectral characteristics of specific minerals and improve the accuracy of lithology identification. Input the ASTER bands and calculate the resulting ratio image to improve the separability of different rock units.

[0090] ASTER-TIR quartz index QRI: (band 10 / band 12)×(band 13 / band12)

[0091] ASTER-TIR mafic mineral index MRI: (band 12 / band 13)×(band 13 / band 14)

[0092] ASTER-TIR mafic mineral index MRI: (band 12 / band 13)×(band 13 / band 14)

[0093] ASTER-TIR carbonate rock index CRI: b13 / b14

[0094] ASTER-SWIR-TIR alkali feldspar index ORI: (b4×b10) / (b6×b11)

[0095] ASTER-SWIR dolomite index DRI: (band7 + band9) / band8

[0096] 1.2.2 Principal Component Analysis (PCA)

[0097] Principal Component Analysis uses Landsat-8 OLI / TIRS data, selects the OLI-TIR full spectral bands, and uses the PCA method in the ENVI software's Transform Method. Principal Component Analysis will generate several principal component (PC) bands, which are linear combinations of the original spectral bands and are uncorrelated with each other. The number of output principal component bands is the same as the number of input spectral bands. The first principal component (PC1) will contain the largest amount of data variance, the second principal component (PC2) contains the second largest data variance, and so on. The "transformation parameters" (Transform Params) are set to covariance and eigenvectors. Select PC1, PC2, and PC3 from the obtained results for false color composition to enhance lithological features.

[0098] 1.2.3 Unsupervised Classification (K-means)

[0099] (1) Combine the bands obtained from the above band ratios, the ASTER shortwave infrared bands, and the principal component analysis bands into a composite image, including the following lithological spectral feature indices and bands: QRI, MRI, FRI, CRI, ORI, DRI, B02, B04, B05, B06, B07, B08, B09, PC1 - PC7. This is used to enhance the clustering separability.

[0100] (2) Extract random pixel samples from the image for training the clusterer, and each sample contains the pixel values of all the above bands.

[0101] (3) K-Means clustering algorithm: divide the pixels into 4 lithology clustering categories. Take (2) 50,000 random samples as input for clustering training; apply the clustering model to the entire image, generate a classified image, customize the color of each cluster, and enhance the map interpretation. The clustering formula means finding a division method among all possible clustering methods C so that the sum of the square distances of all points to the center of the cluster to which they belong is minimized, so as to achieve a clustering effect of "high intra-class consistency". The formula is as follows:

[0102]

[0103] Where: argmin C : It means finding an optimal clustering method among all possible clustering partitions C.

[0104] K: number of clusters. For example, 4 clusters means it is divided into 4 lithological units.

[0105] i=1...K: the i-th class (cluster).

[0106] C i : Belongs to the i-th cluster pixel set (i.e. “cluster”).

[0107] x∈C i : Indicates that the sample point x belongs to the i-th cluster.

[0108] μ i : The “mean vector” of the i-th class (i.e. the center point of all points in the cluster).

[0109] ||x-μ i || 2 : Sample point x to its cluster center μ i The square of the Euclidean distance, which represents the "clustering error" or "closeness".

[0110] (4) According to the lithology of the weathering crust type and the parent rock where rare earths are often enriched, the lithology clustering categories related to rare earth mineralization (numbered 1 and 2) are extracted as lithology background constraints to provide a reference for subsequent rare earth mineralization prediction, such as Figure 2 .

[0111] 1.3 Alteration mineral information extraction

[0112] This module requires the input of ASTER-VNIR-SWIR data. Extract alteration minerals related to weathering crust rare earth mineralization, including kaolinite, illite, halloysite, montmorillonite, chlorite, calcite, and hematite, to assist in identifying the spatial distribution of weathering crust rare earth mineralization, such as Figure 3 .

[0113] 1.3.1 Spectral matching method

[0114] The constrained energy minimization (CEM) algorithm is used for altered mineral extraction. By designing a finite impulse response (FIR) filter, CEM maximizes the spectral signal of the target mineral while suppressing background noise. Its mathematical expression is as follows:

[0115] Covariance matrix of pixels:

[0116]

[0117] x i : Spectral vector of the i-th pixel in the background.

[0118] N: Number of background sample points.

[0119] x: Background mean spectrum.

[0120] CEM filtering equation:

[0121]

[0122] Where: w is the weight vector of the CEM filter (a column vector with the same number of spectral bands);

[0123] R is the background covariance matrix, used to represent the correlation between background pixels;

[0124] d is the standard spectral vector of the target mineral (i.e., the spectral vectors of minerals such as kaolinite and illite in the spectral library);

[0125] dt: Transpose of vector d;

[0126] R -1 : Inverse matrix of the background covariance matrix.

[0127] Calculation formula for the CEM response value of target mineral pixels (pixel detection): Applying the filter w to the spectral vector x of each pixel to obtain its spectral matching degree with the target mineral d. The higher the Y cem value, the higher the abundance of the target mineral.

[0128] Y cem = W T X

[0129] Where:

[0130] Y cem is the response value of a certain pixel, indicating the possibility of the presence of the target mineral in this pixel;

[0131] x is the spectral vector of the pixel to be measured in ASTER or GF-5 data (i.e., its reflectance or radiance in each band).

[0132] w T : Transpose of the filter weights

[0133] Operation process: (1) Select the standard spectral curves of kaolinite, illite, halloysite, montmorillonite, chlorite, calcite, and hematite from the USGS spectral library; (2) Perform spectral resampling to match the ASTER VNIR - SWIR bands; (3) Calculate the background covariance matrix R; (4) Calculate the CEM filter weight vector w; (5) Calculate the CEM response value y of the pixel; (6) Generate the CEM grayscale map of each mineral, and the higher the response value, the higher the abundance of the target mineral.

[0134] 1.3.2 Comprehensive alteration mineral abundance information

[0135] Construct a rare earth - related alteration mineral model to obtain comprehensive alteration mineral abundance information. Based on the regression coefficients, the alteration mineral model related to the Nd 3+ value is:

[0136] Y Nd = 0.046321x1 + 0.019022x2 + 0.007328x3 + 0.001308x4 + (-0.001450)x5 +

[0137] (-0.009700)x6 + (-0.040145)x7

[0138] where Y Nd is the comprehensive alteration mineral abundance information. x1, x2, x3…xn are the normalized mineral characteristic values (kaolinite, illite, halloysite, montmorillonite, chlorite, calcite, and hematite)

[0139] Apply this model to image calculation to obtain the comprehensive abundance grayscale map of alteration minerals related to Nd 3+ .

[0140] 1.4 REE mineral Nd 3+ information extraction

[0141] 1.4.1 Characteristic band selection

[0142] Nd 3+ (Neodymium), as an important component of rare earth elements, has obvious absorption characteristics in the near - infrared VNIR band and is a key tracer factor for identifying ion - adsorbed weathered crust rare earth ores. The typical absorption bands (resulting from the composite absorption of Nd 3+ in various alteration minerals) 744 nm, 802 nm, and 865 nm are the main absorption peaks of Nd 3+ in the VNIR band.

[0143] Identifying the absorption characteristics of Nd in the weathered layer using hyperspectral data (Sentinel-2B and GF5), and reducing the dimensionality through principal component analysis (PCA) to screen out the high-abundance distribution areas of direct tracer factors for rare earth mineralization in the weathered crust of the study area. 3+ For Sentinel-2B MSI (10 - 30m resolution): Select VNIR bands such as B5 (703nm), B6 (740nm), B7 (782nm), and B8A (865nm) to cover the main absorption characteristics of Nd.

[0144] 3+ 3+ For GF-5 hyperspectral data (5 - 10nm spectral resolution): Select bands B83 (741nm), B97 (800nm), B112 (865nm) corresponding to the absorption of Nd, as well as shoulder bands B74, B90, B104, and B121 for feature comparison.

[0145] 3+ 3+ 3+

[0146] 1.4.2 Nd 3+ Characteristic Bands and Principal Component Analysis (PCA)

[0147] Principal component analysis is a linear dimensionality reduction method. The goal is to transform high-dimensional spectral data into a set of new orthogonal variables (principal components), which can represent most of the information (i.e., variance) of the original data. The specific steps are as follows:

[0148] (1) Construct a data matrix

[0149] Assume that the hyperspectral image has m pixels, and each pixel has n bands. Construct the following data matrix:

[0150]

[0151] Each row represents the spectrum of a pixel, and each column represents a band.

[0152] (2) Centralization processing

[0153] The centralization operation is to subtract this mean vector from each row:

[0154]

[0155] where,

[0156] X ′ : The centralized data matrix, with dimensions m×n. is a row vector representing the mean of each band.

[0157] (3) Calculate the covariance matrix

[0158] Calculate the covariance between every two bands using the centralized data of all pixels, thereby establishing a correlation matrix between bands and revealing the change correlation between bands.

[0159]

[0160] C is an n×n covariance matrix that describes the correlation between bands.

[0161] X′ is the centralized data matrix with dimensions m×n.

[0162] X′ T is the transpose matrix of X′ with dimensions n×m.

[0163] (4) Calculate eigenvalues and eigenvectors

[0164] Perform eigenvalue decomposition on the covariance matrix C to obtain eigenvalues and eigenvectors, and extract the most useful information:

[0165] C·e i =λ i ·e i

[0166] λ i : The i-th eigenvalue, representing the variance contribution of the i-th principal component.

[0167] e i : The corresponding eigenvector, indicating the weight of the original band in the linear transformation.

[0168] A total of n pairs of (λ i , e i ) are obtained. Sort the eigenvalues in descending order (λ1 > λ2 >... > λ n ), and sort all e i in descending order according to the corresponding λ i . Select the first K principal component directions (eigenvectors) to form the transformation matrix E:

[0169] E = [e1, e2,..., e k

[0170] Calculate the principal component image. By projecting into the principal component space, reduce the dimension and highlight the main change direction, and project the original spectrum into the principal component space:

[0171] Y PC =X′·E

[0172] Y PC is the data after dimensionality reduction, a matrix with shape m×K. Each column represents the projection score on the corresponding principal component, that is, a principal component. Select the one corresponding to Nd 3+ ​Characteristic components with consistent absorption and reflection characteristics bands. Such as Figure 4 。

[0173] 1.5 Extraction of topographic and geomorphic factor information

[0174] Calculate multiple topographic factors (slope, aspect, topographic relief, cutting depth, roughness, topographic variation coefficient, etc.) from DEM data, and combine each topographic factor into a comprehensive evaluation index (CSn) according to a predetermined weight coefficient. Such as Figure 5 ,and generate a "favorable weathering crust" topographic mask map through normalization and threshold segmentation.

[0175] The detailed steps are as follows:

[0176] (1) Load DEM data ALOS PALSAR Hi-Res Terrain Corrected

[0177] (2) Based on the method of GIS spatial analysis technology, combined with DEM data, generate and analyze key topographic factors through the surface analysis tool of ArcGIS software, including elevation, slope, aspect, surface cutting depth, topographic relief, as well as roughness and topographic variation coefficient.

[0178] (3) Comprehensive evaluation model CSn CS n =0.21768x1 + 0.324627x2 + 0.213371x3 + 0.326055x4 + 0.340596x5 + 0.322556x6 + 0.364317x7 where x1, x2, x3…x n are the standardized factor values (elevation, slope, aspect, surface cutting depth, topographic relief, as well as roughness and topographic variation coefficient), and CS n is the comprehensive score of the Nth pixel.

[0179] 1.6 Extraction of structural information

[0180] Identify the edge features of the tectonic fracture zone on the remote sensing image I (SAR data), and combine the Sobel and Laplacian edge detection results: Sobel can extract the boundaries in the horizontal or vertical directions; Laplacian is sensitive to edges in all directions. Screen the stronger edge responses through thresholds and regard them as possible fracture - fracture zone areas. Such as Figure 6 。Select ALOS / PALSAR SAR data; Band: HH (horizontal transmit - horizontal receive polarization).

[0181] 1.6.1 Convolution kernel and edge detection:

[0182] (1) Sobel operator

[0183] Sobel is mainly divided into Gx (detecting edges in the x direction) and Gy (detecting edges in the y direction), and its 3×3 convolution kernel is defined as:

[0184]

[0185] Using G x The convolution kernel is applied to the ALOS / PALSAR SAR image I, and the horizontal gradient I is obtained by convolution. x , using G y The convolution kernel is applied to the image, and the vertical gradient I is obtained by convolution. y :

[0186] I x =I*G x ,I y =I*G y

[0187] The comprehensive edge strength is:

[0188]

[0189] (2) Laplacian operator

[0190] The Laplacian kernel will perform second-order differentiation on the image, which is suitable for detecting edges or grayscale mutations in all directions. The 3×3 template is:

[0191]

[0192] Calculation formula: LaplacianEdge = I*L. A value greater than 0 indicates a grayscale mutation. The larger the value, the more drastic the change.

[0193] 1.6.2 Threshold Setting and Binarization

[0194] According to experiments or experience, sobel_threshold and laplacian_threshold need to be given to filter out significant edge pixels. In the experimental example, sobel_threshold = 11917; laplacian_threshold = 20853; the area greater than the threshold is set to 1, named SobelMask and LapMask respectively. As long as one of SobelMask or LapMask is 1, it is regarded as the "broken zone edge".

[0195] 1.7 Establishing rare earth tracer element Nd with multi-factor constraints 3+ High value area

[0196] Multi-factor constraint analysis (weighted superposition) to obtain Nd 3+High-value mask: Superimpose the output results of modules such as lithological background, altered minerals, Nd 3+ tracer, terrain, and tectonic fracture zones using multiplicative logical operations; achieve multi-factor conditional constraints, and finally extract the target areas with rare earth mineralization potential to obtain the preliminary rare earth mineralization potential areas, such as Figure 7 .

[0197] Step 2: Use drones to precisely identify fern plants and build a precise sample library

[0198] Based on the highly potential areas of weathering crust rare earth ores delineated by "multi-factor constraint analysis" in this step, deploy drones to obtain high-resolution VNIR images and further identify the distribution of fern plants. Use it as a more precise indicator of rare earth mineralization, and then integrate it with the existing feature matrix to provide a new feature band of "fern plants" for the subsequent CNN model.

[0199] 2.1 Precise target area and drone planning

[0200] Potential area screening: Based on the highly potential areas obtained in Step 1 (lithology, altered minerals, element tracer Nd 3+ , terrain, tectonic fracture zones), select the key areas (A, B, C...) that have been screened for drone flights.

[0201] Drone flight path design: According to the area of the highly potential area, terrain undulation, and vegetation coverage, plan the optimal flight altitude (such as 100 - 200m) and overlap rate (heading overlap ≥ 70%, side overlap ≥ 60%).

[0202] Ground marking and coordinate control: Set ground control points (GCP) in the highly potential area to ensure the geometric accuracy of drone images.

[0203] 2.2 Drone image acquisition (VNIR)

[0204] RGB images: Obtain high-resolution visible light images (resolution 10cm / pixel), record the morphology, color, and texture details of vegetation. Through orthomosaic stitching, obtain the integrated visible light image.

[0205] VNIR multispectral images (400 - 1000nm): Facilitate the extraction of vegetation indices (NDVI, PRI, etc.) and capture the spectral characteristics of fern plants (red edge shift, etc.). Use atmospheric correction and geometric correction, and finally output spectral orthoimages.

[0206] 2.3 Fern plant identification (texture + morphology + band index)

[0207] In the UAV images of high-potential areas, in order to obtain more accurate distribution and samples of ferns, visible light RGB images can be combined with simple spectral indices (such as NDVI, red edge), supplemented by object-level features such as texture and morphology, and identified through machine learning methods such as "support vector machine (SVM)". The specific steps are as follows:

[0208] (1) Feature extraction

[0209] Color features: Calculate the means of R, G, B, as well as Hue, Saturation S, etc. from the UAV RGB or HSV space;

[0210] Texture features: Use GLCM (Gray Level Co-occurrence Matrix) to extract parameters such as contrast, entropy, and uniformity within the pixel neighborhood; Use the NIR and red edge channels to calculate NDVI or RedEdgeIndex;

[0211] Morphological features (object-level): When the resolution is relatively high (10 cm / pixel), the image can be segmented object-oriented, and the roundness and fragmentation indices of the region can be calculated to capture the unique pinnate structure of fern leaves.

[0212] (2) Sample data collection:

[0213] Ground field survey: Select several points in the high-potential area to confirm the growth of ferns, such as Figure 8 . Use GPS to record the coordinates and locate them in the UAV images; ROI annotation: On the UAV orthophoto image, combine the above coordinates and human eye recognition to manually circle the fern area as a Polygon (positive sample), and select non-fern vegetation or bare land as negative samples;

[0214] Generate the training set: Extract the above color, texture, index, and morphological features for each annotated pixel (or object); Mark its category = fern (1) or non-fern (0).

[0215] (3) SVM classification

[0216] Model training: Use the annotated ROI positive and negative sample sets to train the classifier using SVM;

[0217] Verification and parameter tuning: Evaluate the Kappa coefficient through cross-validation;

[0218] Classification output: Apply the trained SVM to the entire UAV image to generate a fern distribution map FernMap (1 = fern, 0 = non-fern).

[0219] (4) Downsampling and merging

[0220] Since the main feature matrix is at a resolution of 10 - 30m, FernMap needs to be downsampled from centimeter-level resolution. Neighborhood statistics (majority voting or coverage calculation) can be used to form a "Fern_Cover" layer (0 - 1) aligned with the main features. This enables the accurate characterization of fern distribution and prosperity within the potential area. Aggregate statistics of detailed information (such as "the proportion of ferns in this 10m pixel = 40%") are still useful new information for the CNN, without having millions of fragmented pixels with a large data volume on a large scale.

[0221] (5) Results: Precise fern distribution and samples

[0222] FernMap (Fern Distribution Map): Presents the locations of ferns within the potential area at high resolution;

[0223] High-quality samples: Fern / non-fern pixels with high confidence, which can be exported and used as "precise samples" for CNN training in Step 3;

[0224] FernCover band: Merged with other features (lithology, alteration, Nd 3+ , topography, structure, etc.) to form a new feature matrix for final mineralization prediction.

[0225] 2.4 Indicative significance of ferns for rare earth mineralization

[0226] If FernMap = 1 and the health is high (high VNIR vegetation index), it is more likely to absorb rare earth elements, indicating greater potential;

[0227] If FernMap = 0 but the multi-factor mineralization index is high, it may still have medium potential due to plant species differences;

[0228] If FernMap = 0 and the multi-factor mineralization index is low, it can be used as a low-potential exploration area.

[0229] Final output: High-resolution "fern distribution band" from the drone, forming a new mask or coverage map for CNN model training.

[0230] Step 3: CNN modeling and identification of high-potential areas for rare earth mineralization

[0231] After integrating the "ore-indicating multi-factor features", the "fern plant Fern_Cover" band is newly added to the feature matrix.

[0232] 3.1 Feature matrix integration

[0233] Existing multi-factor features (lithology, alteration minerals, Nd 3+Absorption, terrain CSn, tectonic fracture zones, etc.); New fern bands: Fern_Cover (0 - 1), including the health index (NDVI_Fern).

[0234] 3.2 CNN Model Training

[0235] The training samples include positive samples and negative samples. Positive samples: Sampling in known rare earth ore spots or Nd 3+ high - value areas + fern - covered areas; Negative samples: Sampling in non - mineralized areas or background areas;

[0236] The input CNN structure for each pixel: Treat the geographical grid data as a multi - channel image and use 2D convolution. Output: Mineralization probability (0 - 1). Parameters such as learning rate, batch size, and number of iterations are set during the training process, and finally, AUC and confusion matrix accuracy evaluations are performed.

[0237] 3.3 Application of the Model and Extraction of High - Potential Areas

[0238] Input the trained CNN model to predict the feature matrix of the entire area; Output the mineralization probability map: Each pixel gets a probability value; Use threshold segmentation to determine "high - medium - low potential areas"; If the mineralization probability is high and the fern coverage is high, ground drilling sampling is preferentially deployed.

[0239] A weathered - crust rare earth ore prospecting system based on the fusion of ferns and multi - source remote sensing data, such as Figure 9 :

[0240] (I) The content of the prospecting system includes:

[0241] 1. Multi - source remote sensing data fusion and preliminary screening: Comprehensive integration of multiple factors such as lithological background, altered minerals, Nd 3+ element tracing, topography and geomorphology, and tectonic fracture zones to preliminarily delineate rare earth mineralization potential areas.

[0242] 2. UAV fine identification of ferns: Deploy UAVs in the preliminarily delineated high - potential areas to identify fern distribution as a "biological indicator factor"; Output "Fern_Cover" to provide more accurate training samples and characteristic bands for subsequent models.

[0243] 3. CNN modeling and terminal prediction: Combine Fern_Cover with other multi - factor features to train CNN; Conduct final mineralization probability prediction for the study area, output the rare earth mineralization probability map, and finally delineate high - mineralization potential areas.

[0244] (II) System Functional Modules and Equipment

[0245] 1. Multi - source Remote Sensing Data Analysis Module

[0246] Function: Receive various remote sensing data such as optical, microwave, thermal infrared, hyperspectral and DEM, and complete preprocessing, feature extraction and preliminary delineation of potential areas.

[0247] Main equipment / software:

[0248] (1) Data acquisition workstation

[0249] Used to download and manage data from official platforms such as Landsat-8, ASTER, Sentinel-2B, GF-5, ALOS PALSAR, GF-3SAR, etc. The storage disk is ≥10TB for storing multi-temporal images.

[0250] (2) Data preprocessing server

[0251] Equipped with atmospheric correction tools (such as ENVI / IDL+FLAASH, or Python), and image processing software (ArcGIS / QGIS).

[0252] The GPU or CPU cluster can be determined according to the project scale and can process a large number of images in parallel;

[0253] (3) Feature extraction and analysis software

[0254] ENVI / IDL, ArcGIS, Python, used to implement: Band Ratio (BR), PCA (Principal Component Analysis), K-Means (unsupervised classification), CEM (altered mineral extraction), DEM-derived terrain analysis (slope, aspect, incision degree, etc.), extraction of tectonic fracture zones (Sobel+Laplacian edge detection)

[0255] (4) Result output

[0256] Generate a potential area mask (Raster) and provide it for the next step of UAV planning and deployment.

[0257] 2. UAV high-resolution monitoring and fern recognition module

[0258] Deploy UAVs within the potential area to collect cm-level RGB / VNIR images, and use methods such as texture + morphology + simple spectroscopy to identify ferns; and downsample to form the Fern_Cover band as an accurate indicator factor for CNN training.

[0259] Main equipment / software:

[0260] (1) UAV platform: quad-rotor or fixed-wing UAV, equipped with a high-resolution RGB camera and a multispectral (400-1000nm) sensor; configuration includes resolution: 10cm / pixel, flight time: ≥30 minutes; payload: multispectral camera (including visible light RGB, NIR or red edge band); with real-time image transmission or waypoint planning function.

[0261] (2) Ground Control Points (GCP) and Surveying Equipment

[0262] RTK GPS or total station is used to calibrate ground coordinates; a number of GCPs are distributed to ensure that the image error after ortho-stitching is ≤1 pixel.

[0263] UAV Ground Station (GCS) and image processing: plan routes (altitude, overlap rate), manage flight missions, receive telemetry data, complete orthomosaic stitching, and output orthophotos (RGB + multispectral channels).

[0264] (3) Fern recognition module

[0265] Feature extraction: color (RGB / HSV), texture (GLCM), morphology (object-oriented segmentation), spectral index (NDVI, RedEdgeIndex);

[0266] Supervised learning classifier: SVM / RF (CPU environment is sufficient), generate FernMap (resolution 3-10cm), 1 = fern, 0 = other;

[0267] Downsampling statistics: Merge cm-level FernMap into 10m-30m "FernCover" (0-1).

[0268] Output: (a) high-resolution fern distribution map (FernMap); (b) downsampled Fern_Cover, which is subsequently incorporated into the CNN feature matrix; (c) precise fern sample points (such as high-confidence ROI) for ground verification or CNN training.

[0269] 3.CNN deep learning prediction module

[0270] Combine "Fern_Cover" with the above-mentioned multiple factors (lithology, alteration, Nd 3+ , topography, structure, etc.) to form the final feature matrix; train the CNN model to output the probability distribution of rare earth mineralization and determine the high potential areas.

[0271] Main equipment / software:

[0272] (1) High-performance computing equipment: GPU servers / workstations (such as NVIDIA RTX / Tesla series, 16-64GB video memory); used for deep learning training and inference.

[0273] (2) Deep learning framework: Python environment, used for building 2D convolution; encapsulating geospatial raster processing, feature matrix generator, aligning and stacking all bands (lithology index, CEM mineral abundance, Nd 3+ PCA, terrain CSn, fracture zone mask, Fern_Cover, etc.) rasters; forming N+1 channels (10 - 30m grid);

[0274] (3) CNN training subsystem: collecting positive and negative samples (known mineralized points vs. background); designing network structure (input (N+1) channels); adjusting hyperparameters such as learning rate, batch size, etc.; evaluating performance metrics such as AUC, confusion matrix, etc.; inference and result visualization, inputting the global features into the trained CNN, outputting mineralization probability (0 - 1); extracting high-potential areas based on a threshold (e.g., p>0.6); rendering and overlaying fern distribution, actual geological map, etc. in GIS for expert decision-making.

[0275] 4. Other auxiliary modules and devices

[0276] (1) Data storage management: RAID disks, used for storing multi-temporal images, DEM, UAV orthophotos, feature matrices, etc.; it is recommended to have a capacity of ≥10TB.

[0277] (2) Front-end UI and GIS visualization, WebGIS (such as Geoserver+Leaflet) or Desktop GIS (ArcGIS / QGIS), showing potential area maps, FernCover distribution, mineralization probability maps, etc.;

[0278] (3) Functions such as interactive query, legend, threshold adjustment, etc.

[0279] (4) Ground verification equipment: XRF analyzer, drilling equipment or surface sampling tools, used for field verification of CNN prediction results to form an exploration closed-loop.

Claims

1. A multi-source remote sensing rare earth ore prospecting method integrating fern ore indication, characterized in that: The following steps are involved: (1) Acquire multi-source remote sensing data, perform atmospheric and geometric correction on the images, and extract ore-forming factors, including rock-minerals, rare earth tracer elements, topography, and tectonic fracture zone information; (2) After preliminary delineation of the potential area of ​​the weathering crust rare earth mineral area based on the mineralization factors, a drone equipped with a visible-near infrared camera was used to fly at low altitude in the potential area to collect high-resolution images and identify the distribution of ferns through texture, morphology and spectral index; (3) Down-sampling the obtained fern distribution to a resolution consistent with other remote sensing features to form a fern coverage layer; (4) integrating the fern coverage layer with the mineralization factor to construct a multi-source feature matrix; (5) Input the feature matrix into the deep learning model for training and prediction, output the rare earth mineralization probability map and select high potential areas.

2. The multi-source remote sensing rare earth ore prospecting method integrating fern ore indication according to claim 1 is characterized in that: The multi-source remote sensing data includes optical images, radar images and DEM data; Optical images include Sentinel-2, Landsat-8, ASTER, and GF5; Radar images include ALOS PALSAR or GF-3. ALOS PALSAR or GF-3 data are used to identify topographic factors and structural fracture zone factors to improve the recognition accuracy of weathering crust rare earth mining areas.

3. The multi-source remote sensing rare earth ore prospecting method integrating fern ore indication according to claim 1 is characterized in that: Identifying the distribution of ferns in step (2) includes the following sub-steps: (2.1) Extract color and texture features of visible light images; (2.2) Use spectral indices to distinguish ferns from other vegetation; (2.3) Calculate the split degree or roundness morphological characteristics of fern leaves based on object-oriented segmentation method; (2.4) Outputting the fern distribution map through a machine learning classifier or a lightweight deep learning network.

4. The multi-source remote sensing rare earth ore prospecting method integrating fern ore indication according to claim 1 is characterized in that: The downsampling step in step (3) is performed using neighborhood statistics, specifically including: (3.1) Select the grid size corresponding to the main remote sensing data; (3.2) Count the percentage of fern pixels in centimeter-level images in each grid; (3.3) Convert the percentage results into a fern coverage layer between 0 and 1.

5. The multi-source remote sensing rare earth ore prospecting method integrating fern ore indication according to claim 1 is characterized in that: In step (5), the deep learning model is a convolutional neural network (CNN), the input of which is a multi-source feature matrix: rock-mineral factors, rare earth tracer elements, topographic structural factors and fern distribution information; and the output is a mineralization probability map.

6. The multi-source remote sensing rare earth ore prospecting method integrating fern ore indication according to claim 1 is characterized in that: The extraction method of the ore-indicating factor of the ore-forming factor in step (1) includes: (1.1) The band ratio is combined with the K-means algorithm to obtain the rock-mineral factor; (1.2) Constrained energy minimization CEM algorithm extracts alteration mineral abundance information; (1.3) Principal component analysis (PCA) to extract the characteristics of rare earth tracer elements; (1.4) Analyze the slope, aspect, cutting depth and terrain variation coefficient of DEM to obtain terrain factors; (1.5) Use Sobel and Laplacian operator threshold detection to construct broken zone information.

7. The multi-source remote sensing rare earth ore prospecting method integrating fern ore indication according to claim 1 is characterized in that: In step (2), during the low-altitude flight of the UAV, ground control points (GCPs) are deployed and orthophoto stitching is performed to align the UAV image with the multi-source remote sensing data in the same coordinate system.

8. The multi-source remote sensing rare earth ore prospecting method integrating fern ore indication according to claim 1 is characterized in that: The training steps of the deep learning model in step (5) include: (5.1) Collect known mineralized points as positive samples and collect non-mineralized or background areas as negative samples; (5.2) registering the multi-source feature matrix to correspond sample labels to pixel features; (5.3) Set the network structure and hyperparameters, and use the cross entropy loss function for iterative optimization; (5.4) Evaluate the model performance based on the confusion matrix and AUC evaluation indicators and save the optimal model parameters.

9. The multi-source remote sensing rare earth ore prospecting method integrating fern ore indication according to claim 1 is characterized in that: In step (5), when the trained deep learning model is applied to the entire feature matrix, the mineralization probability of each pixel is output, and potential weathering crust rare earth mining areas are screened based on the probability threshold, and finally confirmed in combination with ground verification.

10. A multi-source remote sensing rare earth ore prospecting system integrating fern ore indication, used to implement a multi-source remote sensing rare earth ore prospecting method integrating fern ore indication as claimed in any one of claims 1 to 9, characterized in that: The system includes: Multi-source data processing unit, used to acquire and pre-process optical images, radar images and DEM data and extract mineralization factors and mineralization indicators; Aerial photography unit of drones, used to collect low-altitude images and identify ferns in potential areas; Feature fusion and deep learning unit, which constructs a multi-source feature matrix based on fern coverage, rock-mineral factors, rare earth tracer elements, and topographic structural factors, and trains a CNN model; The result output and visualization unit is used to generate mineralization probability maps and screen high potential areas to guide the prospecting and exploration of weathering crust rare earth deposits.

Images