Granite alteration classification system based on mineral spectral signatures

Abstract

This invention discloses a granite alteration degree grading system based on mineral spectral characteristics, belonging to the field of hyperspectral image processing technology. It includes a remote sensing data acquisition module, an image processing module, an equipment scheduling module, a spectral recognition module, and an alteration grading module. By fusing multi-source remote sensing images acquired by satellites, UAVs, and ground equipment, combined with NDVI mutation analysis and image texture feature extraction, it achieves accurate identification of blind spots and scheduling of replenishment mining. Mineral absorption band features are extracted using methods such as the first derivative method, Voigt curve fitting, and spectral angle mapping, and matched with a fingerprint database to complete mineral type identification and alteration degree classification. The system supports outputting results as a structured two-dimensional layer, recording the coordinates, mineral type, alteration type, spectral angle, and intensity level of each pixel, facilitating subsequent analysis and geological modeling. It possesses high-precision recognition capabilities and broad application value.

Description

Technical Field

[0001] This invention relates to the field of hyperspectral image processing technology, specifically to a granite alteration degree grading system based on mineral spectral characteristics. Background Technology

[0002] Existing methods for classifying the degree of alteration in granites mainly rely on geologists' visual observation of rocks in the field and laboratory analysis results, supplemented by optical microscopy, X-ray diffraction (XRD), or electron probe microanalysis to obtain alteration information. However, these traditional methods suffer from problems such as low efficiency, high subjectivity, and limited sample coverage, making them unsuitable for the needs of rapid understanding and high-precision identification of large-scale areas.

[0003] In recent years, with the development of hyperspectral remote sensing and mineral spectral analysis technologies, although existing granite alteration degree classification technologies have begun to incorporate hyperspectral remote sensing and mineral reflectance analysis methods, they still have significant shortcomings overall: At the remote sensing data acquisition level, conventional satellite imagery is easily limited by cloud cover, terrain undulations, and imaging angles, resulting in numerous blind spots and a lack of effective blind spot perception and re-acquisition mechanisms, affecting the integrity and consistency of regional data. In terms of mineral spectral feature identification, traditional processing methods often rely on single bands or a few absorption peaks, making it difficult to accurately reconstruct the combined features of complex altered minerals. Furthermore, the lack of multi-parameter fusion and standard fingerprint matching mechanisms leads to insufficient classification accuracy and a high misjudgment rate. At the result expression and management level, existing systems mainly output static images or tables, lacking structured layer representation and scalable data organization methods, making it difficult to support subsequent intelligent analysis and geological modeling needs. Summary of the Invention

[0004] To address the aforementioned shortcomings, a granite alteration degree classification system based on mineral spectral characteristics is proposed.

[0005] The objective of this invention can be achieved through the following technical solutions: A granite alteration degree grading system based on mineral spectral characteristics includes: a graphics processing module, an equipment scheduling module, a spectral feature recognition module, and an alteration grading module. The graphics processing module performs unified formatting and standardization on the received image data, completes geometric registration, radiometric correction, atmospheric correction, image fusion and slicing operations, and generates a high-quality input image set. The equipment scheduling module is used to dynamically schedule satellite image angle adjustment, UAV aerial photography or ground spectral equipment acquisition tasks according to the identification blind spots and the level of acquisition difficulty, and to progressively supplement the acquisition of unidentifiable areas. The spectral feature recognition module extracts and analyzes spectral features from the preprocessed image data to identify spectral feature information related to granite alteration. The alteration grading module identifies the alteration type and intensity level of the target area based on spectral feature information, and outputs the degree of alteration of the granite.

[0006] In a preferred embodiment of the present invention, a remote sensing data acquisition module is also included; The remote sensing data acquisition module is used to acquire multi-source remote sensing data, including satellite remote sensing images, UAV aerial images, and images acquired by ground equipment. It prioritizes the processing of satellite remote sensing images from multiple time periods. During the identification process, if there are blind spots due to regional occlusion, reflection interference, or insufficient resolution, the area is marked as a satellite-unidentifiable area and transmitted to the equipment scheduling module.

[0007] In a preferred embodiment of the present invention, the process for identifying areas that cannot be identified by satellites is as follows: The blind spot perception and discrimination model based on multidimensional image quality indicators specifically includes: NDVI is calculated for multi-temporal images of the same area. If, within a set time period, the NDVI value is significantly lower than the preset value over a large area, a preliminary occlusion signal is generated. This may be due to occlusion objects blocking surface vegetation, resulting in a significant reduction in near-infrared reflection. Occlusion objects are then identified based on the preliminary occlusion signal. Image texture extraction methods are used to calculate the spatial edge intensity and direction changes of the image. If the edge texture is smooth and has no texture features, the image is determined to be covered by clouds or fog. Conversely, if the edge texture has clear texture features, the image is determined to be covered by dense forest. Based on the multi-temporal image NDVI calculation and image texture extraction methods employed, the analysis results are fused. For any image region R, if the NDVI change rate at the current observation time t compared to the previous time t1 satisfies: |NDVI t (R)-NDVI t1 (R) |>θ1, where θ1 is the set threshold; at the same time, the texture gradient magnitude G(R) of the region satisfies: G(R)<θ2, where G(R) can be extracted by the gray-level co-occurrence matrix, and θ2 is the lower limit of edge intensity, representing that the texture of the region tends to be smooth; If region R simultaneously meets the following conditions: NDVI amplitude exceeds the set threshold; and texture gradient is below the set level, then the region is determined to be an occlusion blind zone and marked as a satellite-unrecognizable area in the occlusion mask map.

[0008] In a preferred embodiment of the present invention, the process by which the device scheduling module determines the blind spot identification and the difficulty level of data acquisition is as follows: Based on the location of the unidentifiable area, the digital elevation model data is called to calculate the slope of the blind area and output the slope value. Access historical remote sensing image archives, search for the frequency of image distortion, blanking, or substandard quality in the past multiple time phases of the region, and calculate the historical imaging failure rate Fr; The calculated slope value (Slope) and historical imaging failure rate (Fr) are then transformed into a continuous mapping function using the S-shaped response curve of the logistic regression model: k1 represents the kurtosis of the response function to changes in slope; t1 represents the inflection point of the slope; f1 represents the influence factor of slope on difficulty. Historical imaging failure rate F r Modeled using a nonlinear surge function: f2(F r )=1-(1-Fr) n1 Where n1 is the power of the amplification sensitivity, which can accelerate the response in areas with high failure rates; f2 is the imaging failure rate response value; the two response values ​​f1 and f2 are fused into the final acquisition difficulty index D, by adopting the joint probability complement model: D=1-(1-f1)(1-f2), and the final acquisition difficulty index D; if the final acquisition difficulty index D is greater than the preset difficulty threshold, it is determined that there is no space for imaging attitude adjustment.

[0009] As a preferred embodiment of the present invention, the process of progressive supplementary data collection for unidentifiable areas is as follows: After detecting unidentifiable areas in the remote sensing image; if some areas are found to have blind spots due to cloud cover, terrain obstruction, or viewing angle limitations, the satellite's shooting angle will be dynamically adjusted based on existing orbital resources to perform supplementary image acquisition, specifically: The system performs intersection analysis between the blind zone and the current orbit to obtain the geographical location of the blind zone. Based on the dual-track orbit elements and satellite imaging parameters, it calls the orbit prediction engine to perform calculations and simulates the trajectory of the target satellite over the next 72 hours using a simulation platform. For each satellite, it checks whether a visible window is formed within a given time. If the trajectory of the target satellite over 72 hours passes through the geographical location of the blind zone and there is room for imaging attitude adjustment, then a satellite-region angular reachability index is generated. Then, based on the satellite-region angular reachability index, structured scheduling instructions are generated for each viewport, including: the latitude and longitude coordinates of the imaging area; and the imaging timestamp T. i Desired imaging pose θ i , Imaging configuration parameters and priority labels; The task schedule of the current satellite is queried according to the priority label and the priority is sorted. If the current satellite is idle during the time period, a structured scheduling instruction is issued. If it cannot be scheduled, it is marked as a scheduling failure and transferred to the UAV platform for processing. The UAV platform collects images by remotely controlling the UAV for aerial photography.

[0010] As a preferred embodiment of the present invention, the process of generating a high-quality input image set is as follows: The image file is received and its metadata is parsed; the image format is converted using GDAL and unified to GeoTIFF format; the band order is uniformly adjusted to the standard order and the pixel value range is standardized to [0,1]; the data is written to a unified standardized image buffer to generate a standardized image set. SIFT feature extraction is performed on standardized and historical images, with a feature point count of ≥2000. Affine transformation moments T(x,y) are fitted based on the RANSAC model, with a tolerance for mismatches of ≤15%. The images are then spatially aligned through affine resampling. If the image tilt angle is ≥20°, an SRTM-1 DEM is introduced for orthorectification. The RMSE is calculated and is required to be ≤0.75 pixels. Image features are then output. Orthorectification: Image features are input into the Py6S model; the digital density (DN) of the remote sensing image is converted into top atmospheric reflectance (TOA); the true surface reflectance after atmospheric influence is estimated using a radiative transfer model: TOA reflectance is used as the observed value, and the atmospheric profile constructed by Py6S is used as the model input; the atmospheric transmittance, atmospheric path radiance, and ground reflectance of each pixel are simulated and inverted using Py6S; the true surface reflectance image is output; the error limit is ≤5%; NDVI is calculated on the orthorectified true surface reflectance image; if the region contains water bodies, the MODIS spectral library is used for reflectance correction, and a high-quality input image set is output.

[0011] In a preferred embodiment of the present invention, the process of identifying spectral feature information related to granite alteration is as follows: Extract the reflectance values ​​[R] of each pixel p(x,y) in the image across all wavelength bands. λ1 ,R λ2 ,...,R λn ], forming a spectral vector Sub-region analysis is used to average the pixels of the 9×9 neighborhood window to form the regional spectral feature vector. The minimum position of the reflectance variation curve is detected using the first derivative method, which represents the center of the absorption band. The absorption band is modeled by fitting the Voigt curve to the spectrum, and the absorption depth, bandwidth and center wavelength are extracted. Typical absorption features are identified. By setting a threshold, if the absorption depth D ≥ 0.03 and the bandwidth W ≥ 40 nm, the spectral feature is valid.

[0012] As a preferred embodiment of the present invention, the process of outputting a complete granite alteration degree grade is as follows: For all pixels in the image that have undergone atmospheric correction and spectral feature extraction, extract their spectral vectors. ∈Rn and compared with known standard spectra in the alteration mineral fingerprint library. Similarity matching is performed, and the matching angle is calculated using the spectral angle mapping algorithm: Output spectral angle ε. If the spectral angle value of a pixel and a mineral satisfies ε≤0.1rad, then the pixel is determined to be highly matched with the mineral type. Sort all mineral matching results and select the one with the smallest angle as the dominant mineral response. Map the matched dominant minerals to the corresponding alteration types; After identifying the specific alteration type, the degree of alteration is graded based on the absorption band depth and bandwidth in the spectral characteristics: if the absorption depth D ≥ 0.08, the bandwidth W ≥ 60 nm, and ε ≤ 0.06, the alteration degree is determined to be strong, and the grade is 3; if the absorption depth D is between 0.05 and 0.08, the bandwidth W is between 40 and 60 nm, and ε ≤ 0.08, the alteration degree is determined to be medium, and the grade is 2; if the absorption depth D is between 0.03 and 0.05, the bandwidth is between 30 and 40 nm, and ε ≤ 0.1, the alteration degree is determined to be weak, and the grade is 1; if the absorption depth D < 0.03, the bandwidth W is any value, and the match fails, the alteration degree is determined to be non-existent, and the grade is 0.

[0013] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention introduces a multi-source remote sensing data acquisition method combining satellite remote sensing, UAV aerial photography, and ground equipment. It integrates NDVI mutation analysis and image texture feature extraction technology to construct a blind zone perception and discrimination model, accurately identifying image recognition blind zones caused by cloud cover, terrain obstruction, etc. Simultaneously, by calculating slope and modeling historical imaging failure rates to form an acquisition difficulty index, and coordinating with trajectory prediction and equipment scheduling strategies, it dynamically supplements data acquisition and closes the information loop for unidentifiable areas, improving the spatiotemporal coverage integrity and continuity of remote sensing data.

[0014] 2. This invention employs high-precision spectral analysis methods, including the first derivative of reflectance, Voigt curve fitting, and spectral angle mapping, to accurately extract substances such as Al-OH and Fe. 2+ The center wavelength, depth, and bandwidth of the absorption bands of key minerals are determined and matched with a standard fingerprint database. By using multi-dimensional parameters such as matching angle ε, absorption depth D, and bandwidth W, the alteration type and degree level are determined collaboratively, with a gradient classification from weak to strong, which improves the accuracy and reliability of granite alteration identification.

[0015] 3. In terms of outputting grading results, this invention organizes all pixels in the form of a two-dimensional raster layer, clearly recording core attributes such as image coordinates, dominant mineral type, alteration type, matching angle, and intensity level (alteration degree), and manages them in a structured manner through layer encoding. This improves the readability and spatial representation of the results, facilitates subsequent statistical analysis, model training, and engineering decision support, and enhances the breadth of applications and integration capabilities. Attached Figure Description

[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 This is a schematic diagram showing the connections of the various modules of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be understood that the terms “comprising” and “including” used in this disclosure and claims indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0020] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure. As used in this disclosure and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this disclosure and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0021] Please see Figure 1 As shown, the granite alteration degree classification system based on mineral spectral characteristics includes: a remote sensing data acquisition module, a graphics processing module, an equipment scheduling module, a spectral feature recognition module, and an alteration degree classification module.

[0022] The remote sensing data acquisition module is used to acquire multi-source remote sensing data, including satellite remote sensing images, UAV aerial images, and images acquired by ground equipment, and to uniformly identify and archive data from different sources.

[0023] The graphics processing module performs preprocessing operations on the acquired image data, including radiometric correction, atmospheric correction, geometric correction, image fusion, and region segmentation, to generate a high-quality input image set.

[0024] The equipment scheduling module is used to dynamically schedule satellite image angle adjustment, UAV aerial photography, or ground spectral equipment acquisition tasks based on the identification blind spots and the level of acquisition difficulty, so as to realize progressive supplementary acquisition of unidentifiable areas.

[0025] The spectral feature recognition module extracts and analyzes spectral features from the preprocessed image data, identifying key spectral features related to granite alteration, such as hydroxyl absorption bands and iron ion characteristics.

[0026] The alteration grading module identifies the alteration type and degree of alteration in the target area based on spectral recognition results, outputs a complete granite alteration degree grade, and supports dynamic updates and visualization of the results.

[0027] In one example, the granite alteration degree classification process based on mineral spectral characteristics is as follows: The remote sensing data acquisition module receives data sources from different remote sensing platforms and prioritizes processing satellite remote sensing images from multiple time phases. During the identification process, if there are blind spots due to regional occlusion, reflection interference, or insufficient resolution, the area is marked as an area that cannot be identified by the satellite and transmitted to the equipment scheduling module.

[0028] The graphics processing module performs unified formatting and standardization on the received image data, completes geometric registration, radiometric / atmospheric correction, and image fusion and slicing operations, and generates a high-quality input image set.

[0029] When there are unidentifiable areas, the equipment scheduling module assesses the difficulty of data collection in the area based on the characteristics of the blind spot (such as slope and degree of terrain obstruction), and dynamically selects the appropriate data supplementation method, including: prioritizing drone aerial photography for medium-difficulty areas, and scheduling ground equipment for supplementary photography for high-difficulty areas, and issuing collection instructions to the relevant sub-modules.

[0030] All processed remote sensing image data will be transmitted to the spectral feature recognition module. This module extracts spectral features from each pixel or sub-region in the image, identifies the location and intensity of feature bands, matches the alteration fingerprint spectrum of granite, and extracts features such as Al-OH and Fe. 2+ Absorption characteristics of key minerals such as OH-.

[0031] The extracted results will be input into the alteration grading module. By setting the spectral feature threshold, mineral assemblage pattern and regional statistical parameters, different regions will be graded and the complete granite alteration degree level will be output. Layered visualization management and overlay display are also supported.

[0032] The process for identifying areas undetectable by satellites involves constructing a blind zone perception and discrimination model based on multi-dimensional image quality indicators, specifically including: Calculate NDVI values ​​for multiple temporal images of the same region, i.e. If, within a set time period (e.g., t0), the NDVI value is significantly lower than the preset value over a large area (e.g., the NDVI of an image of a mountainous area is approximately 0.6 under normal conditions (closed forest), but becomes 0.1 over a large area during image acquisition), a preliminary occlusion signal is generated. This may be due to occlusion objects (e.g., clouds or closed forest) covering the surface vegetation, resulting in a significant reduction in near-infrared reflection. Further occlusion object determination is performed based on the preliminary occlusion signal: image texture extraction methods (e.g., Sobel edge detection or gray-level co-occurrence matrix) are used to calculate the spatial edge intensity and direction changes of the image. If blurred or disappeared edge textures are detected (i.e., the image exhibits "smooth, textureless" characteristics), the image is determined to be covered by clouds or fog. Conversely, if clear texture features are detected at the edges, the image is determined to be covered by closed forest.

[0033] Based on the multi-temporal image NDVI calculation and image texture extraction methods, the analysis results are fused. For any image region R, if the rate of change of NDVI at the current observation time t compared to the previous time t1 satisfies: |NDVI| t (R)-NDVI t1 (R)∣>θ1, where θ1 is a set threshold (e.g., 0.3), indicating that the vegetation index in this area has decreased abnormally; at the same time, the texture gradient magnitude G(R) of this area satisfies: G(R)<θ2, where G(R) can be extracted by the Sobel operator or gray-level co-occurrence matrix, and θ2 is the lower limit of edge intensity (e.g., 10% of the average texture value of the image), representing that the texture of the area tends to be smooth.

[0034] If region R simultaneously meets both of the above conditions, namely: the NDVI change exceeds the set threshold; and the texture gradient is lower than the set level, then the region is determined to be an occlusion blind spot and marked as a satellite-unrecognizable area in the occlusion mask map.

[0035] The equipment scheduling module analyzes the difficulty of data collection in a region based on blind spot characteristics (such as slope and degree of terrain obstruction) and dynamically selects an appropriate data supplementation method. The specific process is as follows: Based on the location of the unidentifiable area, DEC (Digital Elevation Model) data is used to calculate the slope of the blind area: Output the slope value Slope; used to measure the degree of ground inclination, Slope∈[0,1]; where z is the elevation value and (x,y) are pixel coordinates.

[0036] Then, access historical remote sensing image archives and retrieve the frequency of image distortion, blank areas, or substandard quality in this area across multiple historical timeframes. Calculate the historical imaging failure rate Fr: Fr = n fail / n total The output historical imaging failure rate Fr ranges from 0 to 1; where n fail This represents the number of times the area has been deemed unsuccessful in past data collection attempts; n total This represents the total number of past data collection attempts (including successful and failed attempts) in this area.

[0037] The calculated slope value and historical imaging failure rate Fr are then transformed into a continuous mapping function using the S-shaped response curve of the logistic regression model: Where k1 is the "steepness" of the response function to the change in slope, with a recommended value of 0.2 to 0.4; t1 represents the inflection point of the slope (recommended to be set to 25°, i.e., a rapid rise from a moderate slope); f1∈(0,1) represents the influence factor of slope on difficulty.

[0038] Then, the historical imaging failure rate F r ∈[0,1], directly modeled using a nonlinear surge function: f2(F r )=1-(1-Fr) n1 Where n1 is the power of the amplification sensitivity (preferably 2 or 3), which can accelerate the response in high failure rate regions; f2∈[0,1] is the imaging failure rate response value.

[0039] The two response values ​​f1 and f2 are fused into the final acquisition difficulty index D. The joint probability complement model is adopted: D=1-(1-f1)(1-f2), and the final acquisition difficulty index D∈[0,1]. If the final acquisition difficulty index D is greater than the preset difficulty threshold (such as D0), it is determined that there is no imaging attitude adjustment space.

[0040] After detecting unidentifiable areas in the remote sensing image; if some areas are found to have blind spots due to cloud cover, terrain obstruction, or viewing angle limitations, the satellite's shooting angle will be dynamically adjusted based on existing orbital resources to perform supplementary image acquisition, specifically: The intersection analysis between the blind zone and the current orbit is performed to obtain the geographical location of the blind zone. Based on the two-line orbit elements (TLE) and satellite imaging parameters, the orbit prediction engine is called to perform calculations, and the simulation platform simulates the trajectory of the target satellite in the next 72 hours. For each satellite, it is analyzed whether a visible window is formed within a given time. If the trajectory of the target satellite in the 72 hours passes through the geographical location of the blind zone and there is room for imaging attitude adjustment, a satellite-region angle reachability index is generated for use in the subsequent inversion stage.

[0041] Then, based on the satellite-region angular reachability index, structured scheduling instructions are generated for each viewport, including: the latitude and longitude coordinates of the imaging area; and the imaging timestamp T. i Desired imaging pose θ i , Imaging configuration parameters and priority labels; priority labels are preset with levels based on task attributes: when the task attribute is natural disaster emergency or national-level task, the priority is level 1; when the task attribute is natural disaster emergency or national-level task, the priority is level 2; when the task attribute is natural disaster emergency or national-level task, the priority is level 3; when the task attribute is natural disaster emergency or national-level task, the priority is level 4; when the task attribute is natural disaster emergency or national-level task, the priority is level 5; ...

[0042] The task schedule of the current satellite is queried according to the priority label to sort the priority; if the current satellite is idle during the time period, a structured scheduling instruction is issued; if it cannot be scheduled, it is marked as a scheduling failure and transferred to the UAV platform processing flow; the UAV platform collects images by remotely controlling the UAV for aerial photography.

[0043] The graphics processing module, after acquiring the structured scheduling instructions, performs unified formatting and standardization processing on the imaging data and the original acquired image data to generate a high-quality input image set; specifically: Image files (such as JPEG2000, HDF, GeoTIFF) are received and their metadata (including shooting time, sensor type, shooting angle, resolution, and band information) are parsed; GDAL is used to convert the image format to GeoTIFF format (image spatial resolution is normalized to 0.5 m / pixel); the band order is uniformly adjusted to the standard order (such as RGB-NIR), and the pixel value range is normalized to [0,1]; the data is written to a uniform normalized image buffer to generate a normalized image set.

[0044] SIFT / SURF feature extraction is performed on standardized and historical images (spatial resolution not less than 1m / pixel), with a feature point count of ≥2000. The affine transformation moment T(x,y) is fitted based on the RANSAC model, with a tolerance for mismatches of ≤15%. The images are then spatially aligned through affine resampling. If the image tilt angle is ≥20° (derived from pose metadata), an SRTM-1 DEM (accuracy ≤30m) is introduced to perform orthorectification. The RMSE is calculated, requiring ≤0.75 pixels, and the image features are output. Otherwise, if the limit is exceeded, re-registration is required.

[0045] Orthorectification: Image features (regional altitude, visibility, and aerosol type) are input into the Py6S model.

[0046] The digital data (DN) of the remote sensing image is then converted into top atmospheric reflectance (TOA). Since TOA reflectance is still affected by the atmosphere, it is not equal to the true surface reflectance. The true surface reflectance after atmospheric influence is estimated using a radiative transfer model: TOA reflectance is used as the observed value, and the atmospheric profile constructed by Py6S is used as the model input. The atmospheric transmittance, atmospheric path radiation, and ground reflectance of each pixel are simulated and inverted using Py6S. The true surface reflectance image is output with an error limit of ≤5%.

[0047] Calculate NDVI from the orthorectified true surface reflectance image; if the region contains water (NDVI≤0.1), import the MODIS spectral library for reflectance correction and output a high-quality input image set.

[0048] The extraction of spectral features from each pixel or sub-region in the image specifically involves: Extract the reflectance values ​​[R] of each pixel p(x,y) in the image across all wavelength bands. λ1 ,R λ2 ,...,R λn ], forming a spectral vector Sub-region analysis is used to average the pixels of the 9×9 neighborhood window to form the regional spectral feature vector.

[0049] The location of the minimum value of the reflectance variation curve was detected using the first derivative method, representing the center of the absorption band; the absorption band was modeled by fitting the Voigt curve to the spectrum, and the absorption depth, bandwidth, and center wavelength were extracted; typical absorption features were identified. Al-OH absorption band: 2200±20nm (muscovite, sericite, etc.); Fe²⁺ absorption band: 1000±30nm (chlorite); OH⁻ absorption band: 1400–1900 nm (bound water, clay); by setting a threshold, if the absorption depth D ≥ 0.03 and the bandwidth W ≥ 40 nm, then the spectral characteristics are valid.

[0050] The output of the complete granite alteration degree level is specifically as follows: For all pixels in the image that have undergone atmospheric correction and spectral feature extraction, extract their spectral vectors. ∈R n and compared with known standard spectra in the alteration mineral fingerprint library. Similarity matching is performed, and the matching angle is calculated using the spectral angle mapping algorithm: Output the spectral angle ε. If the spectral angle value of a pixel and a mineral satisfies ε≤0.1rad, then the pixel is determined to be a high match for the mineral type. Sort all mineral matching results and select the one with the smallest angle as the dominant mineral response.

[0051] The matched dominant minerals (such as kaolinite, sericite, chlorite, etc.) are mapped to the corresponding alteration types; for example, kaolinite exhibits kaolinization alteration.

[0052] After identifying the specific alteration type, the degree of alteration is further graded based on the absorption band depth (D) and bandwidth (W) in the spectral characteristics: if the absorption depth D ≥ 0.08, the bandwidth W ≥ 60 nm, and ε ≤ 0.06, the degree of alteration is determined to be strong, and the grade is 3; if the absorption depth D is between 0.05 and 0.08, the bandwidth W is between 40 and 60 nm, and ε ≤ 0.08, the degree of alteration is determined to be medium, and the grade is 2; if the absorption depth D is between 0.03 and 0.05, the bandwidth is between 30 and 40 nm, and ε ≤ 0.1, the degree of alteration is determined to be weak, and the grade is 1; if the absorption depth D < 0.03, the bandwidth W is any value, and the match fails, the degree of alteration is determined to be non-existent, and the grade is 0.

[0053] All pixels are organized into a two-dimensional raster layer based on their identified etch type and etch severity level: Each pixel stores the following fields: (x,y): image coordinates; mineral type: such as "sericite"; alteration type: such as "sericite"; ε angle: such as 0.057; alteration degree level: such as 3.

[0054] Layer coding: For example, "M2-L3" indicates mineral type 2 (sericite) and alteration level 3.

[0055] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A granite alteration degree grading system based on mineral spectral characteristics, including: The graphics processing module, equipment scheduling module, spectral feature recognition module, and alteration grading module are characterized by: The graphics processing module performs unified formatting and standardization on the received image data, completes geometric registration, radiometric correction, atmospheric correction, image fusion and slicing operations, and generates a high-quality input image set. The equipment scheduling module is used to dynamically schedule satellite image angle adjustments, UAV aerial photography, or ground-based spectral equipment acquisition tasks based on the identification blind spots and the level of acquisition difficulty. The progressive supplementary acquisition process for unidentifiable areas is as follows: After detecting unidentifiable areas in the remote sensing image; if some areas are found to have blind spots due to cloud cover, terrain obstruction, or viewing angle limitations, the satellite's shooting angle will be dynamically adjusted based on existing orbital resources to perform supplementary image acquisition, specifically: The system performs intersection analysis between the blind zone and the current orbit to obtain the geographical location of the blind zone. Based on the dual-track orbit elements and satellite imaging parameters, it calls the orbit prediction engine to perform calculations and simulates the trajectory of the target satellite over the next 72 hours using a simulation platform. For each satellite, it checks whether a visible window is formed within a given time. If the trajectory of the target satellite over 72 hours passes through the geographical location of the blind zone and there is room for imaging attitude adjustment, then a satellite-region angular reachability index is generated. Then, based on the satellite-region angular reachability index, a structured scheduling instruction is generated for each visual window, including: latitude and longitude coordinates of the imaging area, imaging timestamp, desired imaging attitude, imaging configuration parameters, and priority label; The task schedule of the current satellite is queried according to the priority label and the priority is sorted; if the current satellite is idle during the time period, a structured scheduling instruction is issued; if it cannot be scheduled, it is marked as scheduling failure and transferred to the UAV platform processing flow; the UAV platform collects images by remotely controlling the UAV for aerial photography. The spectral feature recognition module extracts and analyzes spectral features from the preprocessed image data to identify spectral feature information related to granite alteration. The alteration grading module identifies the alteration type and classifies the degree of alteration in the target area based on spectral feature information, and outputs the degree of alteration of the granite.

2. The granite alteration degree grading system based on mineral spectral characteristics according to claim 1, characterized in that, It also includes a remote sensing data acquisition module; The remote sensing data acquisition module is used to acquire multi-source remote sensing data, including satellite remote sensing images, UAV aerial images, and images acquired by ground equipment. It prioritizes the processing of satellite remote sensing images from multiple time periods. During the identification process, if there are blind spots due to regional occlusion, reflection interference, or insufficient resolution, the area is marked as a satellite-unidentifiable area and transmitted to the equipment scheduling module.

3. The granite alteration degree grading system based on mineral spectral characteristics according to claim 2, characterized in that, The process of identifying areas that cannot be identified by satellite is as follows: The blind spot perception and discrimination model based on multidimensional image quality indicators specifically includes: NDVI values ​​are calculated for multiple temporal images of the same area. If, within a set time period, the NDVI value falls below a preset value over a large area, a preliminary occlusion signal is generated. The occlusion object is then determined based on this preliminary occlusion signal. Image texture extraction methods are used to calculate the spatial edge intensity and direction changes of the image. If the edge texture is smooth and has no texture features, the image is determined to be covered by clouds or fog. Conversely, if the edge texture has clear texture features, the image is determined to be covered by dense forest. Based on the multi-temporal image calculation of NDVI value and image texture extraction method, the analysis results are fused. For any image region R, if the NDVI change rate at the current observation time t compared to the previous time t1 satisfies: |NDVI t (R)-NDVI t1 (R) |>θ1, where θ1 is the set threshold; at the same time, the texture gradient magnitude G(R) of the region satisfies: G(R)<θ2, where G(R) is extracted by the gray-level co-occurrence matrix, and θ2 is the lower limit of edge intensity, representing that the texture of the region tends to be smooth; If an image region R simultaneously meets the following conditions: the NDVI change exceeds a set threshold; and the texture gradient is below a set level, then the region is determined to be an occlusion blind spot and marked as a satellite-unrecognizable area in the occlusion mask map.

4. The granite alteration degree grading system based on mineral spectral characteristics according to claim 1, characterized in that, The device scheduling module schedules devices based on identified blind spots and data acquisition difficulty levels. The specific process is as follows: Based on the location of the unidentifiable area, the digital elevation model data is called to calculate the slope of the blind area and output the slope value. Access historical remote sensing image archives, search for the frequency of image distortion, blanking, or substandard quality in the past multiple time phases of the region, and calculate the historical imaging failure rate Fr; The calculated slope value (Slope) and historical imaging failure rate (Fr) are then transformed into a continuous mapping function using the S-shaped response curve of the logistic regression model: k1 represents the kurtosis of the response function to changes in slope; t1 represents the inflection point of the slope; f1 represents the influence factor of slope on difficulty. Historical imaging failure rate F r Modeled using a nonlinear surge function: f2(F r )=1-(1-Fr) n1 Where n1 is the power of the magnification sensitivity; f2 is the imaging failure rate response value; the two response values ​​f1 and f2 are fused into the final acquisition difficulty index D, and the joint probability complement model is adopted: D=1-(1-f1)(1-f2); if the final acquisition difficulty index is greater than the preset difficulty threshold, it is determined that there is no space for imaging attitude adjustment.

5. The granite alteration degree grading system based on mineral spectral characteristics according to claim 1, characterized in that, The process of generating a high-quality input image set is as follows: The image file is received and its metadata is parsed; the image format is converted using GDAL and unified to GeoTIFF format; the band order is uniformly adjusted to the standard order and the pixel value range is standardized to [0,1]; the data is written to a unified standardized image buffer to generate a standardized image set. SIFT feature extraction is performed on standardized and historical images, with a feature point count of ≥2000. The affine transformation moment T(x,y) is fitted based on the RANSAC model, with a tolerance for mismatches of ≤15%. The images are then spatially aligned through affine resampling. If the image tilt angle is ≥20°, an SRTM-1 DEM is introduced to perform orthorectification. Calculate the RMSE (Recovery Mean Squared) with a requirement of ≤0.75 pixels, and output the image features. Orthorectification: Image features are input into the Py6S model; the digital density (DN) of the remote sensing image is converted into top atmospheric reflectance (TOA); the true surface reflectance after atmospheric effects is estimated using a radiative transfer model: TOA reflectance is used as the observed value, and the atmospheric profile constructed by Py6S is used as the model input; the atmospheric transmittance, atmospheric path radiance, and ground reflectance of each pixel are simulated and inverted using Py6S; the true surface reflectance image is output; the error limit is ≤5%; NDVI is calculated on the orthorectified true surface reflectance image; If the region contains water, the MODIS spectral library is used for reflectance correction, and a high-quality input image set is output.

6. The granite alteration degree grading system based on mineral spectral characteristics according to claim 1, characterized in that, The process of identifying spectral features related to granite alteration is as follows: Extract the reflectance values ​​[R] of each pixel p(x,y) in the image across all wavelength bands. λ1 ,R λ2 ,...,R λn ], forming a spectral vector Sub-region analysis is used to average the pixels of the 9×9 neighborhood window to form the regional spectral feature vector. The minimum position of the reflectance variation curve is detected using the first derivative method, which represents the center of the absorption band. The absorption band is modeled by fitting the Voigt curve to the spectrum, and the absorption depth, bandwidth, and center wavelength are extracted. Typical absorption features are identified. If the absorption depth D ≥ 0.03 and the bandwidth W ≥ 40 nm, the spectral features are valid.

7. The granite alteration degree grading system based on mineral spectral characteristics according to claim 6, characterized in that, The process of outputting a complete granite alteration grade is as follows: For all pixels in the image that have undergone atmospheric correction and spectral feature extraction, extract their spectral vectors. ∈R n and compared with known standard spectra in the alteration mineral fingerprint library. Similarity matching is performed, and the matching angle is calculated using the spectral angle mapping algorithm: Output spectral angle ε. If the spectral angle value of a pixel and a mineral satisfies ε≤0.1rad, then the pixel is determined to be highly matched with the mineral type. Sort all mineral matching results and select the one with the smallest angle as the dominant mineral response. The matched dominant minerals are mapped to the corresponding alteration types. After identifying the specific alteration type, the intensity is graded according to the absorption band depth and bandwidth in the spectral characteristics: if the absorption depth D ≥ 0.08, the bandwidth W ≥ 60 nm and ε ≤ 0.06, the alteration degree is determined to be strong, and the grade is three; if the absorption depth D ∈ (0.05–0.08), the bandwidth W ∈ (40–60 nm) and ε ≤ 0.08, the alteration degree is determined to be medium, and the grade is two; if the absorption depth D ∈ (0.03–0.05), the bandwidth W ∈ (30–40 nm) and ε ≤ 0.1, the alteration degree is determined to be weak, and the grade is one; if the absorption depth D < 0.03, the bandwidth W is any value, and the matching fails, the alteration is determined to be non-existent, and the grade is zero.

Images