Fluorite ore wall rock alteration detection system based on pattern recognition

Abstract

The application provides a fluorite ore wall rock alteration detection system based on pattern recognition, and relates to the technical field of geological exploration data processing. Firstly, hyperspectral, RGB texture and laser radar data are spatially corrected and radiometrically calibrated to obtain an aligned spectral cube. After spectral smoothing and multi-order differential enhancement of weak alteration mineral characteristics, a deep unmixing network is used to decompose mixed pixels into each end-member mineral abundance. In the alteration intensity quantification stage, the mineral interaction and synergy characteristics and phase diversity entropy weight are introduced, and a nonlinear coupling fusion model is established to solve the problem of weak alteration zone recognition accuracy decay caused by the linear cumulative model ignoring the paragenetic relationship of minerals. Finally, the structure tensor is used to perform spatial gradient vectorization analysis on the alteration index, and the adaptive threshold segmentation and gradient flow constraint boundary evolution algorithm are combined to output the alteration detection results with intensity grading and accurate boundary positioning, providing pixel-level precision decision support for deep concealed vein exploration.

Description

Technical Field

[0001] This application relates to the field of geological exploration data processing technology, and more specifically, to a pattern recognition-based system for detecting alteration of fluorite ore surrounding rocks. Background Technology

[0002] As an important strategic non-metallic mineral resource, the efficiency of fluorite deposit exploration directly affects the raw material security of the fluorochemical industry chain. In fluorite mining practice, the spatial distribution characteristics of wall rock alteration are key geological indicators of the ore body's location and grade changes. Accurately detecting the extent and intensity of alteration zones can provide a scientific basis for drilling direction. However, wall rock alteration in mine working faces usually exhibits a continuous distribution characteristic of gradual transition from unaltered wall rock to the ore body core. The edges of alteration zones are extremely thin, and the mineral composition shows slight gradual changes. This poses a severe challenge to achieving high-precision alteration detection based on pattern recognition technology.

[0003] Existing methods for detecting alteration in surrounding rocks primarily rely on hyperspectral or multispectral remote sensing data, combined with traditional classification algorithms, to identify altered regions. The core drawback of these methods lies in the fact that the spatial resolution of the sensors is far greater than the microcrystalline scale of the altered minerals. This results in a single pixel containing mixed signals from multiple trace altered minerals. Traditional hard classification models forcibly categorize each spatial unit as altered or unaltered, ignoring the essential characteristic of alteration as a continuous chemical evolution process. Consequently, weak alteration signals are filtered out as noise in strong background surrounding rocks, leading to severely blurred alteration boundaries. Furthermore, even when using spectral unmixing techniques to obtain mineral abundance information and quantify alteration intensity, existing methods generally employ linear accumulation models to integrate multiple mineral contributions. This fails to identify the geochemical synergistic and repulsive relationships between indicator minerals during diagenesis, and cannot capture the ore-forming fluid activity characteristics represented by strong coupling between sericitization and silicification. This results in a significant decrease in identification accuracy in weak alteration zones and mineral evolution transition zones, making it difficult to meet the needs of precise location of deep, concealed veins in industrial settings.

[0004] Therefore, an optimized pattern recognition-based system for detecting alteration of fluorite ore surrounding rocks is desired. Summary of the Invention

[0005] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a pattern recognition-based system for detecting alteration of fluorite ore surrounding rocks, comprising: The data alignment module is used to perform spatial geometric correction and radiometric calibration on the raw sensor data stream to obtain an aligned spectral cube. The raw sensor data stream includes hyperspectral data, RGB texture images, and lidar depth data. The spectral enhancement module is used to perform spectral feature enhancement and multi-order differential preprocessing on the aligned spectral cube to obtain the enhanced feature tensor. The endmember extraction module is used to extract endmember components from the enhanced feature tensor based on a deep unmixing network to obtain a mineral abundance map. The alteration intensity quantification module is used to perform alteration intensity quantification modeling on mineral abundance maps to obtain an alteration index matrix characterizing the alteration intensity of the surrounding rock; The trend analysis module is used to perform alteration trend analysis on the alteration index matrix based on the spatial gradient tensor to obtain the alteration gradient vector field; The region localization module is used to perform region segmentation and boundary localization on the alteration gradient vector field and alteration index matrix to obtain the final alteration detection map.

[0006] Compared with existing technologies, the pattern recognition-based alteration detection system for fluorite ore surrounding rocks provided in this application first performs spatial geometric correction and radiometric calibration on hyperspectral data, RGB texture images, and lidar depth data to obtain aligned spectral cubes in a unified coordinate system. Then, it enhances the absorption peak characteristics of weakly altered minerals through spectral smoothing and multi-order differential operations. Next, it utilizes a deep unmixing network to decompose mixed pixels into the abundance distribution of each end-member mineral, achieving quantitative characterization of mineral components at the sub-pixel level. Based on this, it introduces mineral interaction and synergistic features and alteration phase diversity weights, and establishes an alteration intensity index model through nonlinear coupling fusion, solving the problem of the linear accumulation model's inability to capture the symbiotic relationship of minerals, which leads to the attenuation of the recognition accuracy of weak alteration zones. Furthermore, it uses the structural tensor to vectorize and analyze the spatial gradient of the alteration index to obtain the direction and intensity of alteration evolution. Finally, it combines an adaptive threshold segmentation and gradient flow constraint-based boundary evolution algorithm to output detection results with alteration intensity grading and precise boundary positioning, providing pixel-level precision technical support for exploration decisions of deep concealed veins in fluorite ore deposits. Attached Figure Description

[0007] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0008] Figure 1 This is a system block diagram of a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application; Figure 2 This is a schematic diagram of the data flow of a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application; Figure 3 This is a block diagram of the end-member extraction module in the pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application; Figure 4 This is a block diagram of the alteration quantification module in the pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application; Figure 5 This is a block diagram of an intensity analysis unit in a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application; Figure 6 This is a block diagram of the trend analysis module in the pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application. Detailed Implementation

[0009] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0010] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" are not specifically singular and may include plural forms. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0011] While this application makes various references to certain modules of the systems according to embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The modules described are merely illustrative, and different aspects of the systems and methods may use different modules.

[0012] This application uses system block diagrams and data flow diagrams to illustrate the operations performed by the system according to embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from them.

[0013] In the detection of alteration in fluorite ore host rocks, the spatial resolution of sensors is much larger than the microcrystalline scale of altered minerals. A single pixel contains mixed signals from multiple trace altered minerals. Traditional hard classification models cannot quantify the continuous distribution of alteration intensity, leading to the submergence of weak alteration signals and severely blurred boundaries. Furthermore, existing intensity evaluation methods use linear accumulation models, neglecting the geochemical symbiotic relationships between indicator minerals, resulting in a significant decrease in the accuracy of weak alteration zone identification. Therefore, this application proposes a pattern recognition-based fluorite ore host rock alteration detection system. Specifically, it first performs spatial geometric correction and radiometric calibration on hyperspectral data, RGB texture images, and lidar depth data to establish an aligned spectral cube. After enhancing the absorption peak response of weak altered minerals through spectral smoothing and multi-order differential operations, a deep unmixing network is used to extract endmember components from the mixed pixels, decomposing the pixels into the abundance distribution of each mineral, achieving quantitative characterization of sub-pixel-level components. In the alteration intensity quantification stage, mineral interaction synergy characteristics and phase diversity entropy weights are introduced. Nonlinear coupling fusion modeling replaces linear accumulation to capture the mineral symbiotic enhancement effect and the phase complexity of the alteration front. Furthermore, the alteration index is analyzed by spatial gradient vectorization using the structural tensor. Combined with an adaptive threshold segmentation and gradient flow constraint-based boundary evolution algorithm, detection results with intensity grading and precise boundary positioning are output, providing pixel-level precision decision support for deep concealed vein exploration.

[0014] Figure 1 This is a system block diagram of a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application. Figure 2 This is a schematic diagram of the data flow in a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application. Figure 1 and Figure 2 As shown, the fluorite ore surrounding rock alteration detection system 100 based on pattern recognition according to an embodiment of this application includes: a data alignment module 110, used to perform spatial geometric correction and radiometric calibration processing on the original sensor data stream to obtain an aligned spectral cube, the original sensor data stream including hyperspectral data, RGB texture images and lidar depth data; a spectral enhancement module 120, used to perform spectral feature enhancement and multi-order differential preprocessing on the aligned spectral cube to obtain an enhanced feature tensor; an endmember extraction module 130, used to extract endmember components from the enhanced feature tensor based on a deep unmixing network to obtain a mineral abundance map; an alteration intensity quantization module 140, used to perform alteration intensity quantization modeling on the mineral abundance map to obtain an alteration index matrix characterizing the alteration intensity of the surrounding rock; a trend analysis module 150, used to perform alteration trend analysis on the alteration index matrix based on a spatial gradient tensor to obtain an alteration gradient vector field; and a region positioning module 160, used to perform region segmentation and boundary positioning on the alteration gradient vector field and the alteration index matrix to obtain a final alteration detection map.

[0015] In the aforementioned pattern recognition-based fluorite ore surrounding rock alteration detection system 100, the data alignment module 110 is used to perform spatial geometric correction and radiometric calibration processing on the original sensor data stream to obtain an aligned spectral cube. The original sensor data stream includes hyperspectral data, RGB texture images, and lidar depth data. It should be noted that due to the uneven three-dimensional topological structure of the rock wall surface at the underground working face of a fluorite mine, and the differences in installation pose between the hyperspectral sensor, RGB camera, and lidar, the directly acquired original sensor data stream has pixel-level offsets in spatial coordinates. Different modal data cannot be correlated pixel-by-pixel. Furthermore, the original hyperspectral data is recorded as digital quantization values ​​rather than true reflectance, making it severely affected by fluctuations in light source illuminance. Therefore, the technical solution of this application first performs spatial geometric correction and radiometric calibration processing on the original sensor data stream to obtain an aligned spectral cube. Through this processing, heterogeneous sensor data can be unified to the same physical coordinate system, and the spectral signal can be converted into physically meaningful reflectance data, providing a spatially consistent and radiometrically accurate data foundation for subsequent spectral feature enhancement and mineral identification.

[0016] More specifically, in a concrete example of this application, firstly, depth-constrained geometric correction is performed on the RGB texture image in the original sensor data stream based on LiDAR depth data to obtain a geometrically corrected image. This process parses the LiDAR depth data from the original sensor data stream, generates a 3D topological structure of the rock surface using a point cloud reconstruction algorithm, and then projects the pixels of the RGB texture image onto this 3D topological structure according to the relative extrinsic parameter matrix between the RGB camera and the LiDAR. Bilinear interpolation is then used to eliminate perspective distortion caused by the shooting angle and rock surface undulations, outputting a geometrically corrected image with a uniform spatial scale. Subsequently, cross-modal spatial pixel-level registration is performed on the hyperspectral data in the original sensor data stream based on the geometrically corrected image to obtain a registration spectral cube. This process extracts the brightness channel of the geometrically corrected image and calculates the mean image of the hyperspectral data across all bands. Using the mutual information between the two as a similarity metric, the affine transformation matrix is ​​solved through iterative optimization, and the hyperspectral data is spatially resampled to ensure that each spectral pixel is perfectly aligned with the geometrically corrected image in physical coordinates. The calculation of mutual information is expressed as follows: in, The mutual information between the geometrically corrected image and the hyperspectral mean image. The joint probability distribution of the two images. and These represent the edge probability distributions of the corresponding images. and These represent the pixel gray levels in the corresponding image. Finally, the registration spectral cube undergoes spectral dimension radiometric calibration to obtain the aligned spectral cube. This process calls the sensor's preset wavelength-related gain coefficient and offset parameters to convert the original digital quantization values ​​in the registration spectral cube into physically meaningful reflectance data, and performs standard whiteboard calibration correction to eliminate the influence of uneven illumination from artificial light sources in the mine. Its radiometric conversion is expressed as: in, To be at wavelength Spatial coordinates The reflectivity value, To register the corresponding original digital quantization values ​​in the spectral cube, For the sensor at wavelength The radiation gain coefficient at that location. For the sensor at wavelength The radiation offset at that location. After the above three stages of processing, a spatially aligned and radiation-corrected aligned spectral cube is finally obtained.

[0017] In the aforementioned pattern recognition-based fluorite ore host rock alteration detection system 100, the spectral enhancement module 120 is used to perform spectral feature enhancement and multi-order differential preprocessing on the aligned spectral cube to obtain an enhanced feature tensor. It should be noted that due to the high humidity and low illumination environment in mines, the original reflectance curve in the aligned spectral cube has a low signal-to-noise ratio, and the absorption peaks of indicator minerals such as sericite and kaolinite in the fluorite host rock alteration zone are weak and easily masked by background noise. Directly using them for subsequent mineral identification would result in feature submersion. Therefore, the technical solution of this application further enhances the spectral features of the aligned spectral cube and performs multi-order differential preprocessing to obtain an enhanced feature tensor. Through this processing, the spectral curvature features of the altered minerals can be amplified while suppressing random noise, and the original reflectance information and differential features can be integrated into a multi-dimensional unified representation, providing high-quality input data for subsequent deep unmixing networks.

[0018] More specifically, in a concrete example of this application, the aligned spectral cube is first subjected to spectral smoothing to obtain smoothed spectral data. This process involves extracting the reflectance curve along the wavelength dimension for each spatial pixel in the aligned spectral cube, setting the sliding window length and polynomial fitting order, performing least-squares polynomial convolution within the window using a Savitzky-Golay filter, and then weighting the original spectrum using a convolution weight matrix to filter out random high-frequency noise generated by mine illumination fluctuations, outputting smoothed spectral data. The smoothing calculation is expressed as follows: in, To smooth out the first Reflectance values ​​at each band To align the spectral cube within the window range The original reflectivity of each band, The pre-defined convolution coefficients for the Savitzky-Golay filter. Let the radius be the sliding window. This is the normalization factor. Subsequently, the smoothed spectral data undergoes multi-order differential extraction based on the central difference to obtain differential spectral components. This process performs a second-order differential operation along the wavelength axis on the smoothed spectral data to eliminate the influence of linear baseline shift and amplify the absorption peak curvature characteristics of altered minerals at specific wavelengths. This makes weak absorption valleys, which were originally difficult to identify in the original reflectance curve, stand out, generating differential spectral components. The second-order differential calculation is expressed as: in, To be at wavelength The second-order spectral differential value at that point, , , These represent the reflectance values ​​of the smoothed spectral data at adjacent spectral band positions. The spectral sampling band interval is defined. Finally, the smoothed spectral data and the differential spectral components are recombined in a spatiotemporal-spectral feature dimension to obtain an enhanced feature tensor. This process uses the smoothed spectral data as the basic spatial information layer to preserve the original reflectance intensity features, and the differential spectral components as a subtle feature enhancement layer. The two are superimposed along the feature dimension through channel stitching, and the multidimensional features after stitching are subjected to scale standardization to ensure that features of different dimensions have a balanced contribution weight in subsequent pattern recognition, ultimately generating an enhanced feature tensor.

[0019] In the aforementioned pattern recognition-based fluorite ore host rock alteration detection system 100, the endmember extraction module 130 is used to extract endmember components from the enhanced feature tensor using a deep unmixing network to obtain a mineral abundance map. It should be noted that, given the gradual distribution of alteration zones in fluorite ore host rocks, the sensor's spatial resolution is much larger than the microcrystalline scale of altered minerals. The signal within a single pixel is essentially a convolutional superposition of the reflectances of multiple altered minerals such as sericite, calcite, and kaolinite, rather than a pure single mineral response. Traditional hard classification models forcibly classify each spatial unit as altered or unaltered, ignoring the essence of alteration as a continuous chemical evolution process. This causes weak alteration signals to be filtered out as anomalous noise in strong background host rocks, making quantitative characterization of mineral components impossible. Therefore, the technical solution of this application further extracts endmember components from the enhanced feature tensor using a deep unmixing network to obtain a mineral abundance map. Through the above processing, the mixed pixels can be decomposed into the abundance ratio of each end-member mineral in the current pixel, realizing the quantitative extraction of mineral components at the sub-pixel level, and providing input data with clear physical meaning for the subsequent continuous quantitative modeling of alteration intensity.

[0020] Figure 3 This is a block diagram of the endmember extraction module in a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application. Figure 3 As shown, the endmember extraction module 130 includes: a non-negative encoding unit 131, used to perform non-negative encoding feature mapping processing on the enhanced feature tensor to obtain a latent abundance representation; a linear reconstruction unit 132, used to perform linear reconstruction based on endmember spectral constraints on the latent abundance representation to obtain an original abundance estimate; and a normalization unit 133, used to perform abundance summation constraint normalization processing on the original abundance estimate to obtain a mineral abundance map.

[0021] In the aforementioned pattern recognition-based fluorite ore wall rock alteration detection system 100, the non-negative coding unit 131 is used to perform non-negative coding feature mapping processing on the enhanced feature tensor to obtain a latent abundance representation. It should be noted that, due to the high spectral dimension of each pixel in the enhanced feature tensor, directly estimating mineral abundance in the original high-dimensional space is easily affected by dimensional redundancy and noise interference, and the contribution rate of minerals in mixed pixels is physically not allowed to be negative. Based on this, the technical solution of this application further performs non-negative coding feature mapping processing on the enhanced feature tensor to obtain a latent abundance representation. Through the above processing, high-dimensional spectral features can be compressed into a low-dimensional latent space matching the number of mineral endmembers, while ensuring that the feature components satisfy the non-negativity constraint, providing a physically reasonable initial abundance estimate for subsequent linear reconstruction based on the endmember spectral library.

[0022] More specifically, in a concrete example of this application, a multi-layer fully connected neural network is constructed as an encoder to project the enhanced feature tensor from the high-dimensional spectral feature space to the low-dimensional latent space. The number of layers and neurons in each layer of the encoder are set according to the number of alteration mineral types to be identified in the fluorite ore host rock, so that the output dimension is consistent with the number of endmember minerals. A non-negative constraint activation function is applied to the output layer of the encoder to restrict all output components to the non-negative interval, ensuring that the extracted features physically satisfy the property that the mineral contribution rate cannot be negative. Furthermore, the contextual spectral features of adjacent pixels are aggregated through a spatial neighborhood weight function, and the spatial continuity of adjacent pixels in the alteration zone in terms of mineral composition is used to suppress the interference of complex downhole lighting conditions on single-point feature extraction, ultimately generating a latent abundance representation.

[0023] Specifically, the decoder of the deep unmixing network adopts a symmetrical structure design with the encoder. Its weight matrix is ​​directly constructed from a pre-set standard spectral library rather than obtained through training. The standard spectral library stores the standard reflectance curves of key indicator minerals in the alteration process of fluorite ore host rocks. Each column corresponds to the spectral response characteristics of an endmember mineral across the entire spectral range, and the number of columns is consistent with the encoder output dimension, i.e., the number of endmembers to be identified. During forward propagation, the decoder performs matrix multiplication on the latent abundance representation and the standard spectral library weight matrix. This is physically equivalent to linearly superimposing the endmember spectra according to the proportions of each component in the latent abundance representation to reconstruct the simulated observation spectrum of the current pixel. Since the decoder weights remain fixed and unupdated during training, the network's optimization gradient only applies to the encoder parameters. This forces the latent abundance representation learned by the encoder to accurately reconstruct the original input under the constraint of known endmember spectra, thus ensuring that the abundance decomposition results have clear mineralogical interpretability.

[0024] In the aforementioned pattern recognition-based fluorite ore host rock alteration detection system 100, the linear reconstruction unit 132 is used to perform linear reconstruction of the potential abundance representation based on endmember spectral constraints to obtain the original abundance estimate. It should be noted that since the potential abundance representation is only an abstract mapping result of the encoder in a low-dimensional space, and has not yet established a correspondence with the physical spectral characteristics of specific minerals, it is impossible to directly determine the mineral type and abundance accuracy represented by each component. Based on this, the technical solution of this application further performs linear reconstruction of the potential abundance representation based on endmember spectral constraints to obtain the original abundance estimate. Through the above processing, physical constraints can be applied to the abundance decomposition process using the standard spectral characteristics of known minerals, ensuring that the unmixing result has a clear mineralogical meaning, thereby transforming abstract low-dimensional features into interpretable mineral component proportions.

[0025] More specifically, in a concrete example of this application, a pre-defined standard spectral library is used as the weight matrix of the decoding layer. This spectral library contains the standard reflectance features of key indicator minerals in the alteration process of fluorite ore host rocks, such as fluorite, sericite, calcite, and kaolinite. Each column corresponds to the spectral curve of a mineral endmember. Using the logic of a linear mixture model, the latent abundance representation is multiplied by the standard spectral library to reconstruct the simulated observation spectrum of the current pixel. This simulated spectrum is physically equivalent to the theoretical response after linearly superimposing each endmember mineral according to the proportions in the latent abundance representation. Based on this, a reconstruction loss function is established. By minimizing the Frobenius norm error between the actual observed spectrum and the simulated reconstructed spectrum in the enhancement feature tensor, and introducing a sparse regularization term to constrain the sparsity of the abundance distribution, the encoder parameters are iteratively optimized in reverse, so that the abundance estimate gradually approaches the true mineral component proportions, ultimately generating the original abundance estimate. The reconstruction loss function is expressed as follows: in, Let $\mathbf{ ... To enhance the observations in the feature tensor, The endmember matrix is ​​composed of a standard spectral library. Representing the current potential abundance, The sparsity penalty coefficient, This represents the total number of end-member minerals. For the first The absolute value of the abundance component of a mineral.

[0026] In the aforementioned pattern recognition-based fluorite ore host rock alteration detection system 100, the normalization unit 133 is used to perform abundance summation constraint normalization on the original abundance estimate to obtain a mineral abundance map. It should be noted that since the original abundance estimate is a numerical solution obtained through network optimization iteration, the sum of abundance of each mineral component at the same pixel may deviate from 1 due to the influence of reconstruction residuals and model gains. This does not satisfy the physical constraint that the sum of the proportions of all mineral components in the mixed pixel must be strictly equal to 1, which will lead to distortion of the contribution ratio of each mineral during subsequent alteration intensity quantification. Based on this, the technical solution of this application further performs abundance summation constraint normalization on the original abundance estimate to obtain a mineral abundance map. Through the above processing, it can be ensured that the mineral abundance allocation of each spatial sampling point satisfies the physical conservation condition, so that the abundance values ​​can be directly used to characterize the true proportion relationship of minerals, providing a proportionally accurate input for quantitative modeling of alteration intensity.

[0027] More specifically, in a concrete example of this application, for each spatial pixel in the original abundance estimate, the sum of all endmember mineral abundance components is calculated along the mineral channel dimension. Then, the abundance value of each mineral at that pixel is divided by this sum, ensuring that the sum of all mineral components after normalization is strictly equal to 1. This process corrects for scaling biases caused by changes in illumination intensity or model gain while maintaining the relative abundance relationships between minerals. The normalization calculation is expressed as follows: in, For mineral abundance maps at spatial points First The final normalized abundance value of the mineral, This is the initial estimate corresponding to the original abundance estimate. To detect the total number of mineral endmembers involved, This serves as an index variable for mineral types. After normalization, the corrected data is remapped to a spatial coordinate system, ultimately generating a mineral abundance map.

[0028] In the aforementioned pattern recognition-based fluorite ore host rock alteration detection system 100, the alteration quantification module 140 is used to perform quantification modeling of alteration intensity on the mineral abundance map to obtain an alteration index matrix characterizing the alteration intensity of the host rock. It should be noted that, given that the mineral abundance map records the abundance proportion of each end-member mineral such as sericite, calcite, and kaolinite at each spatial pixel in a multi-channel format, but the degree of host rock alteration is not determined by the content of a single mineral, but rather by the comprehensive evolutionary state of multiple indicator minerals under the action of ore-forming hydrothermal fluids, different minerals have different indicative meanings for alteration intensity. Directly using discrete multi-channel abundance data cannot provide a unified intensity measure for subsequent spatial gradient analysis and regional segmentation. Based on this, the technical solution of this application further performs quantification modeling of alteration intensity on the mineral abundance map to obtain an alteration index matrix characterizing the alteration intensity of the host rock. Through the above processing, based on the differences in geological contributions of each indicator mineral in the genetic model of fluorite deposits, multidimensional mineral abundance information can be integrated into a single continuous intensity scalar, so that each spatial sampling point has a comparable numerical characterization of alteration degree, providing a physically meaningful input for gradient analysis of alteration trends and boundary location of alteration areas.

[0029] Figure 4 This is a block diagram of the alteration quantification module in a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application. Figure 4As shown, the alteration differentiation module 140 includes: a weight mapping unit 141, used to perform mineral component weight mapping processing on the mineral abundance map through a weight vector to obtain weighted abundance components; an intensity analysis unit 142, used to perform alteration intensity analysis on the weighted abundance components to obtain the original index matrix; and a range normalization unit 143, used to perform intensity dynamic range normalization on the original index matrix to obtain the alteration index matrix.

[0030] In the aforementioned pattern recognition-based fluorite ore wall rock alteration detection system 100, the weight mapping unit 141 is used to perform mineral component weight mapping processing on the mineral abundance map through weight vectors to obtain weighted abundance components. It should be noted that since the abundance values ​​of each end-member mineral in the mineral abundance map only reflect its component proportion within a pixel, and in the genetic evolution of fluorite deposits, different alteration minerals have different indicative significance for the wall rock alteration intensity. For example, the presence of sericitization often indicates the core area of ​​ore-forming hydrothermal activity, and its contribution weight to alteration intensity should be higher than that of associated minerals such as calcite. If the abundance of each mineral is weighted equally, the alteration intensity evaluation will deviate from the geological reality. Based on this, the technical solution of this application further performs mineral component weight mapping processing on the mineral abundance map through weight vectors to obtain weighted abundance components. Through the above processing, the abundance values ​​can be proportionally scaled according to the differences in the geological contribution of each indicator mineral in the alteration zoning of the fluorite deposit, so that the participation degree of each mineral in the subsequent intensity accumulation calculation matches its actual indicative significance.

[0031] More specifically, in a specific example of this application, a weight vector is preset based on the zoning pattern of fluorite deposits. Each element in this weight vector corresponds to the contribution coefficient of an alteration indicator mineral to the alteration intensity of the surrounding rock. The value of the contribution coefficient is determined based on the formation order and indication intensity of the minerals during the hydrothermal alteration process. Minerals with a higher correlation to the mineralization core area are assigned larger weight values. Subsequently, mineral abundance channel data corresponding one-to-one with each component in the weight vector are extracted from the mineral abundance map. The abundance value of each channel at each spatial pixel is then subjected to a pixel-by-pixel Hadamard product operation with its corresponding weight coefficient. That is, the abundance value and weight value at the same spatial location are multiplied element-wise, so that the abundance value of each mineral channel is scaled according to its geological contribution ratio, and finally a weighted abundance component is generated.

[0032] In the aforementioned pattern recognition-based alteration detection system 100 for fluorite ore surrounding rock, the intensity analysis unit 142 is used to perform alteration intensity analysis on the weighted abundance components to obtain the original index matrix. It should be noted that, given that the weighted abundance components record the abundance information of each mineral after scaling by geological contribution weights in a multi-channel format, they are still discrete multi-dimensional data and have not yet been integrated into a unified scalar capable of characterizing the comprehensive alteration degree of a single spatial sampling point. Therefore, they cannot be directly used for subsequent spatial gradient calculation and region segmentation. Based on this, the technical solution of this application further performs alteration intensity analysis on the weighted abundance components to obtain the original index matrix. Through the above processing, the weighted contributions of multiple minerals can be aggregated into a single alteration intensity value at each pixel, generating a continuously distributed original index matrix, providing a numerical basis for subsequent dynamic range normalization and alteration trend analysis.

[0033] Figure 5 This is a block diagram of the intensity analysis unit in a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application. Figure 5 As shown, the intensity analysis unit 142 includes: a collaborative extraction subunit 1421, used to extract mineral interaction collaborative features from the weighted abundance components to obtain collaborative interaction features; an entropy weight estimation subunit 1422, used to estimate the alteration phase diversity weights of the weighted abundance components to obtain alteration entropy weights; and a nonlinear fusion subunit 1423, used to perform nonlinear coupling alteration index fusion of the weighted abundance components and collaborative interaction features based on the alteration entropy weights to obtain the original index matrix.

[0034] In the aforementioned pattern recognition-based fluorite ore wall rock alteration detection system 100, the collaborative extraction subunit 1421 is used to extract mineral interaction collaborative features from the weighted abundance components to obtain collaborative interaction features. It should be noted that, given that existing solutions only use linear summation to accumulate alteration intensity for the weighted abundance components, their inherent limitation lies in failing to identify the geochemical symbiotic and repulsive relationships between indicator minerals during the diagenesis of fluorite deposits. Wall rock alteration is not a simple physical superposition of mineral abundances, but rather a dynamic product of complex chemical replacement between ore-forming hydrothermal fluids and the wall rock. Linear models neglect this nonlinear interaction effect and cannot capture nodes of intense ore-forming fluid activity, such as the strong coupling between sericitization and silicification. Therefore, the technical solution of this application further extracts mineral interaction collaborative features from the weighted abundance components to obtain collaborative interaction features. Through the above processing, by performing pairwise cross-modeling of different mineral channels, the signal of mineral synergistic growth induced by ore-forming fluid activity can be captured, the characteristic response of the ore-forming center area can be enhanced, thereby improving the positioning accuracy of the concealed strong alteration center and the mineralization core area.

[0035] More specifically, in a concrete example of this application, a nonlinear cooperative operator is constructed to perform pairwise cross-modeling of different mineral channels in the weighted abundance components. For the weighted abundance components... An indicator mineral is used to traverse all unique mineral pair combinations. For each mineral pair, the geometric mean of its abundance values ​​is calculated. This geometric mean calculation produces a strong response when both mineral abundances are high simultaneously, while the response naturally decays when the abundance of either mineral approaches 0, thus reflecting the nonlinear enhancement effect of inter-mineral synergy. Based on this, a preset geological synergy coefficient is introduced to weight the geometric mean of each mineral pair. This coefficient is based on the mineral... With minerals The intensity of geochemical co-existence during ore-forming hydrothermal metasomatism is determined, with mineral pairs exhibiting higher co-existence correlations being assigned larger synergy coefficients. The weighted geometric mean of all mineral pairs is summed to generate the final synergistic interaction characteristic. Its calculation is expressed as follows: in, For the generated collaborative interaction features in coordinates The value at that location, To indicate the total number of mineral types, Geological coherence coefficient, representing minerals With minerals The intensity of geochemical symbiosis, and These are the first two components in the weighted abundance component. Article and Section The abundance values ​​of each channel. For example, the alteration detection of fluorite ore host rocks involves three indicator minerals: sericite, calcite, and kaolinite. The value is 3, and the mineral pair combinations to be calculated include three groups: sericite and calcite, sericite and kaolinite, and calcite and kaolinite. Since sericitization and silicification often exhibit a strong coupled symbiotic relationship in areas of intense hydrothermal activity during mineralization, the synergy coefficient between sericite and calcite is... The synergy coefficient between calcite and kaolinite was set to a relatively high value. The abundance of sericite and calcite is relatively low. In the core mineralization region, the abundance of both sericite and calcite is high. Their geometric mean term generates a strong response and is amplified by a high synergy coefficient, making the synergistic interaction characteristic value of this region much higher than that of the background region with only a single mineral high abundance. This effectively distinguishes the active alteration front with multiple minerals alternating violently from the background region with a single mineral high abundance.

[0036] In the aforementioned pattern recognition-based alteration detection system 100 for fluorite ore surrounding rock, the entropy weight estimation subunit 1422 is used to estimate the alteration phase diversity weight of the weighted abundance component to obtain the alteration entropy weight. It should be noted that, given that the chemical substitution reactions at the edge of the alteration zone are more chaotic and active than in the central area, multiple minerals exhibit a highly mixed distribution in the alteration front region, while the mineral composition in the background surrounding rock region is relatively simple. Simple content statistics cannot reflect this dynamic difference in phases. If the complexity of mineral distribution is not quantified, subsequent alteration index fusion will be unable to distinguish between the background area and the active alteration front. Based on this, the technical solution of this application further estimates the alteration phase diversity weight of the weighted abundance component to obtain the alteration entropy weight. Through the above processing, the mineral distribution state of spatial sampling points can be evaluated using information entropy theory, and the background area and the active alteration front can be distinguished by entropy values, thereby effectively suppressing environmental background noise and highlighting weak alteration veins, ensuring that the detection boundary remains sharp in complex downhole environments.

[0037] More specifically, in a concrete example of this application, for each spatial pixel in the weighted abundance component, the weighted abundance values ​​of each mineral channel at that pixel are first normalized, and the probability distribution ratio of each mineral in the total amount of minerals in the current pixel is calculated. Then, based on information entropy theory, the entropy value of this probability distribution is calculated. When only a single mineral dominates at a pixel, its probability distribution is highly concentrated, and the information entropy tends to 0, indicating that the mineral composition in this area is simple and belongs to a stable region of background rock or alteration core. When multiple minerals coexist in nearly equal proportions, the information entropy tends to its maximum value, indicating that this area is at the active alteration front of drastic mineral replacement. Based on this, the calculated information entropy is standardized and its complement is taken, so that areas with more uniform mineral distribution receive lower weight values, and areas with more concentrated mineral distribution receive higher weight values, ultimately generating alteration entropy weights. The calculation is expressed as follows: in, The generated alteration entropy weights in coordinates The value at that location, For the first The probability distribution of a particular mineral within the total number of minerals in the current pixel is obtained by normalizing the weighted abundance components. To indicate the total number of minerals, This is the maximum entropy normalization factor, used to map entropy values ​​to the range of 0 to 1. Assuming that at a certain spatial pixel, the weighted abundance of sericite is 0.9, and calcite and kaolinite are 0.05 respectively, the mineral distribution is highly concentrated in sericite, and the information entropy is close to 0. After complementation, the alteration entropy weight is close to 1, indicating that this region is an alteration core area with a clearly defined mineral composition, and will obtain a strong intensity response in subsequent fusion. At another pixel on the edge of the alteration zone, the proportions of sericite, calcite, and kaolinite are 0.4, 0.35, and 0.25 respectively, the mineral distribution tends to be uniform, and the information entropy is close to its maximum value. After complementation, the alteration entropy weight tends to 0, indicating that this region is at the transitional front of drastic mineral replacement. In subsequent fusion, its intensity response will be moderately suppressed, thus avoiding misjudging a transitional region with active chemical replacement but not yet stable alteration as a high-intensity alteration core.

[0038] In the aforementioned pattern recognition-based fluorite ore host rock alteration detection system 100, the nonlinear fusion subunit 1423 is used to perform nonlinear coupling alteration index fusion on weighted abundance components and cooperative interaction features based on alteration entropy weights to obtain the original index matrix. It should be noted that, given that the aforementioned steps have already generated cooperative interaction features reflecting the symbiotic relationship between minerals and alteration entropy weights quantifying the complexity of mineral distribution, it is necessary to perform deep unified modeling of static mineral abundance and dynamic geochemical indicators to construct a nonlinear evaluation system capable of mapping the hydrothermal dynamic evolution process. Based on this, the technical solution of this application further performs nonlinear coupling alteration index fusion on weighted abundance components and cooperative interaction features based on alteration entropy weights to obtain the original index matrix. Through the above processing, the ore-forming core can be enhanced through the cooperative term, and the alteration front can be sensitively corrected through the entropy weight term, achieving nonlinear amplification of the intensity of active alteration zones, outputting an intensity characterization with geological evolution logic, and improving the reliability and detection sensitivity of fluorite ore detection in complex mineralization backgrounds.

[0039] More specifically, in a concrete example of this application, weighted abundance components, synergistic interaction features, and alteration entropy weights are input simultaneously. First, the weighted abundance components are linearly accumulated along the mineral channel dimension to obtain a basic linear term. This linear term reflects the static alteration intensity after the minerals are superimposed according to their geological contribution weights. Subsequently, the synergistic interaction features are multiplied by a synergistic term adjustment coefficient and added to the basic linear term to form a coupling term. This coupling term superimposes the synergistic effect of mineral symbiosis onto the static intensity, allowing the mineralization core area where multiple minerals coexist at high abundance to obtain additional intensity gains. Based on this, the coupling term is nonlinearly exponentially modulated using the alteration entropy weight multiplied by an entropy weight amplitude adjustment coefficient as the power of an exponential function. When the alteration entropy weight is high, the exponential amplification effect is enhanced; when the alteration entropy weight is low, the exponential amplification effect is weakened, thereby achieving differentiated response control for areas with different alteration activity levels, ultimately generating the original exponential matrix. Its calculation expression is as follows: in, The original exponent matrix generated in the final coordinate system Pixel value at that location, The first weighted abundance component One portion, To indicate the total number of minerals, As a feature of collaborative interaction, For alteration entropy weights, This is a synergy adjustment coefficient used to control the intensity of mineral synergy in the fusion process. The entropy-weighted amplitude adjustment coefficient controls the degree of nonlinear amplification of the final exponent by the complexity of the alteration phase. For example, in a certain core mineralization region of a fluorite ore host rock, the weighted abundances of sericite and calcite are both high, and their basic linear cumulative terms already have high values. Simultaneously, due to the strong symbiotic relationship between the two minerals, the synergistic interaction characteristics also produce a strong response. After being amplified by the synergistic term adjustment coefficient, this is superimposed on the linear term, further increasing the coupling term. The alteration entropy weight in this region is close to 1. After the entropy-weighted amplitude adjustment coefficient, the exponential function exhibits a strong amplification effect, and the original exponential matrix in this region shows a higher intensity value than the linear cumulative result. However, in the background host rock region far from the ore body, only a small amount of calcite is distributed, the linear term value is low, and the synergistic interaction characteristics tend to be 0 due to the lack of mineral symbiotic pairs. The alteration entropy weight is also low, and the exponential amplification effect is weak. The original exponential matrix in this region is close to the result of the basic linear cumulative sum. Through this nonlinear coupling mechanism, the intensity contrast between the ore-forming core area and the background surrounding rock is effectively stretched, enabling subsequent regional segmentation and boundary positioning to more accurately extract weak alteration anomalies.

[0040] In the aforementioned pattern recognition-based fluorite ore wall rock alteration detection system 100, the range normalization unit 143 is used to normalize the intensity dynamic range of the original index matrix to obtain the alteration index matrix. It should be noted that, because the numerical range of the original index matrix after nonlinear coupling fusion is affected by multiple factors such as mineral abundance distribution, synergy coefficient, and index amplitude adjustment, there are differences in numerical scale between different working faces or different collection batches. Furthermore, local extreme values ​​may compress the effective dynamic range of the overall data, making it impossible to compare on a unified numerical benchmark during subsequent spatial gradient calculation and threshold segmentation. Based on this, the technical solution of this application further normalizes the intensity dynamic range of the original index matrix to obtain the alteration index matrix. Through the above processing, the values ​​of the original index matrix can be mapped to a standardized 0-1 interval, eliminating scale differences under different collection conditions, making the alteration intensity values ​​comparable across scenarios, and providing numerically stable input for subsequent gradient tensor analysis and region segmentation.

[0041] More specifically, in a concrete example of this application, firstly, all pixels in the original exponent matrix are traversed, and their global maximum and minimum values ​​are calculated to determine the dynamic range of the current working surface alteration intensity. Then, a maximum-minimum normalization operator is used to scale each pixel value in the original exponent matrix, linearly mapping it to the interval between 0 and 1, where the pixel with the lowest alteration intensity is mapped to 0, and the pixel with the highest alteration intensity is mapped to 1. The normalization calculation is expressed as follows: in, The alteration index matrix generated after normalization is in spatial coordinates The value at that location, These are the corresponding pixel values ​​in the original exponent matrix. This represents the global maximum value in the original exponent matrix. This represents the global minimum value in the original index matrix. After the above normalization process, we finally obtain an alteration index matrix with a uniform numerical range and a clear distribution of alteration intensity.

[0042] In the aforementioned pattern recognition-based fluorite ore surrounding rock alteration detection system 100, the trend analysis module 150 is used to perform alteration trend analysis on the alteration index matrix based on spatial gradient tensors to obtain an alteration gradient vector field. It should be noted that, given that the alteration index matrix records the alteration intensity value of each spatial sampling point in scalar form, but surrounding rock alteration is not uniformly distributed in space, but rather exhibits a gradual attenuation trend from the ore core to the periphery, static intensity values ​​alone cannot reveal the rate of change and evolution direction of alteration intensity in space, and cannot provide directional guidance for mining operations towards high-quality ore cores. Based on this, the technical solution of this application further performs alteration trend analysis on the alteration index matrix based on spatial gradient tensors to obtain an alteration gradient vector field. Through the above processing, the spatial gradient of alteration intensity can be extracted, obtaining a vectorized representation containing the degree of alteration severity and evolution direction, realizing the transformation from static content detection to dynamic trend prediction, providing directional constraints for boundary evolution in subsequent area segmentation, and providing vector-level guidance for on-site mining direction decisions.

[0043] Figure 6 This is a block diagram of the trend analysis module in a pattern recognition-based fluorite ore surrounding rock alteration detection system according to an embodiment of this application. Figure 6 As shown, the trend analysis module 150 includes: a spatial convolution unit 151, used to perform spatial convolution on the alteration index matrix to obtain partial derivative components; a tensor construction unit 152, used to perform second-order moment combination and Gaussian smoothing based on the partial derivative components to obtain a structure tensor matrix; and a vector synthesis unit 153, used to perform evolution trend vector analysis and synthesis on the structure tensor matrix to obtain an alteration gradient vector field.

[0044] In the aforementioned pattern recognition-based fluorite ore surrounding rock alteration detection system 100, the spatial convolution unit 151 is used to perform spatial convolution on the alteration index matrix to obtain partial derivative components. It should be noted that since the values ​​in the alteration index matrix are distributed in scalar form on a two-dimensional spatial grid, to obtain the spatial trend of alteration intensity, it is necessary to calculate its local rates of change in the horizontal and vertical directions, i.e., spatial partial derivatives. However, discretized matrix data cannot be directly analyzed for differentiation and requires numerical approximation using a discrete convolution operator. Based on this, the technical solution of this application further performs spatial convolution on the alteration index matrix to obtain partial derivative components. Through the above processing, the local rates of change of the alteration index in both the horizontal and vertical directions can be extracted, capturing subtle abrupt changes in the transition of surrounding rock alteration from the unaltered zone to the strongly altered zone, providing basic gradient data for the subsequent construction of the structural tensor matrix.

[0045] More specifically, in a concrete example of this application, the Scharr operator is used as the convolution kernel to perform discrete convolution operations in the horizontal and vertical directions on the alteration index matrix. Compared with the traditional Sobel operator, the Scharr operator has higher rotational symmetry and edge response accuracy, enabling more accurate estimation of local gradients under complex mine rock surface textures. In the horizontal direction, the Scharr horizontal convolution kernel is performed pixel-by-pixel sliding convolution with the alteration index matrix to calculate the rate of change of the alteration index along the horizontal direction at each spatial location, obtaining the horizontal partial derivative component. In the vertical direction, the Scharr vertical convolution kernel is performed in the same convolution operation with the alteration index matrix to calculate the rate of change of the alteration index along the vertical direction at each spatial location, obtaining the vertical partial derivative component. The convolution calculation is expressed as follows: in, and These are the horizontal and vertical derivatives of the partial derivative components, respectively. The input is the alteration index matrix. and These are the Scharr convolution operator kernels for the horizontal and vertical directions, respectively. This is a two-dimensional convolution operator. After convolution operations in the two directions mentioned above, the final result is partial derivative components containing both horizontal and vertical channels.

[0046] In the aforementioned pattern recognition-based fluorite ore surrounding rock alteration detection system 100, the tensor construction unit 152 is used to perform second-order moment combination and Gaussian smoothing based on the partial derivative components to obtain a structural tensor matrix. It should be noted that since the partial derivative components only record the instantaneous rate of change of the alteration index in the horizontal and vertical directions at a single pixel, they are easily affected by noise from mine rock surface texture and local shot noise caused by uneven illumination. The stability of single-point gradient information is insufficient, and it cannot describe the anisotropic distribution characteristics of alteration intensity changes in the local neighborhood. Therefore, the technical solution of this application further performs second-order moment combination and Gaussian smoothing based on the partial derivative components to obtain a structural tensor matrix. Through the above processing, local gradient information can be integrated into a symmetric positive definite matrix describing the anisotropic characteristics of the spatial evolution of the alteration index. Simultaneously, Gaussian smoothing suppresses the interference of noise on gradient estimation, providing a stable tensor representation with directional information for subsequent vector analysis.

[0047] More specifically, in a particular example of this application, the horizontal component of the partial derivative components is... and vertical components Perform a second-order moment tensor combination, and for each spatial pixel position, calculate... The square term, The square term and and The interaction term assembles the three second-order moment components into a second-order symmetric positive definite matrix. The diagonal elements of this matrix reflect the intensity of the alteration index change in its respective direction, while the off-diagonal elements reflect the correlation between the horizontal and vertical gradients. Subsequently, using parameters... The Gaussian window function performs a spatial neighborhood weighted average on the above second-order moment components. By smoothing and aggregating the gradient statistics within the local window, it suppresses gradient jitter caused by rock surface microtexture and acquisition noise, enhancing the robustness of the tensor to the estimation of alteration boundary directions, and finally generating the structure tensor matrix. Its matrix construction is expressed as: in, For the generated structure tensor matrix, For parameters The Gaussian smoothing operator is used to spatially smooth local second moments to enhance robustness. and These are the squared terms of the horizontal and vertical gradients in the partial derivative components, respectively. This is the interaction term between the horizontal and vertical gradients. This is a two-dimensional convolution operator. After the above combination of second-order moments and Gaussian smoothing, the final structure tensor matrix that can stably characterize the local spatial evolution direction and intensity of the alteration index is obtained.

[0048] In the aforementioned pattern recognition-based fluorite ore wall rock alteration detection system 100, the vector synthesis unit 153 is used to perform evolution trend vector analysis and synthesis on the structural tensor matrix to obtain the alteration gradient vector field. It should be noted that since the structural tensor matrix encodes the gradient statistical characteristics of the alteration index in its local neighborhood in the form of a second-order symmetric matrix, its matrix form cannot be directly used to characterize the dominant direction and intensity of alteration evolution. It is necessary to parse vector parameters with clear physical meaning from it so that the boundary evolution in subsequent region segmentation can obtain directional constraints. Based on this, the technical solution of this application further performs evolution trend vector analysis and synthesis on the structural tensor matrix to obtain the alteration gradient vector field. Through the above processing, the tensor matrix can be transformed into a vector representation containing a modulus and an orientation angle, where the modulus reflects the intensity of alteration intensity change, and the orientation angle points to the direction of the fastest increase in alteration intensity, thereby providing vector-level guidance for mining operations towards high-quality ore cores.

[0049] More specifically, in a concrete example of this application, eigenvalue decomposition is performed on the second-order symmetric matrix at each spatial pixel position in the structure tensor matrix to extract principal eigenvalues ​​and secondary eigenvalues. The principal eigenvalues ​​represent the maximum rate of change in alteration intensity, and the secondary eigenvalues ​​represent the rate of change in directions orthogonal to the principal direction. The principal axis direction of alteration evolution is determined based on the eigenvectors corresponding to the principal eigenvalues, and the absolute direction of the vector is determined by combining the sign of the partial derivative components. The direction angle of the gradient is then calculated. The vector magnitude calculation is expressed as follows: in, represents the vector magnitude of each spatial point in the alteration gradient vector field, reflecting the severity of the alteration evolution. and These are the principal and secondary eigenvalues ​​obtained after eigenvalue decomposition of the structure tensor matrix. Their evolution pointing angle is calculated as follows: in, The directional angle of alteration evolution indicates the direction in which the alteration intensity increases most rapidly. This is the interaction term between the horizontal and vertical gradients in the partial derivative components. and These are the squared terms of the horizontal and vertical gradients, respectively. After calculating the modulus and pointing angle, the vector information of each pixel in space is meshed and integrated. Each spatial location corresponds to a two-dimensional vector containing the modulus and pointing angle. At the alteration boundary, the vector modulus is larger and the direction points to the side where the alteration intensity increases. In the alteration core region, the vector modulus is smaller and the direction tends to diverge. In the background surrounding rock region, the vector modulus is close to 0, and finally, the alteration gradient vector field is generated.

[0050] In the aforementioned pattern recognition-based fluorite ore surrounding rock alteration detection system 100, the region positioning module 160 is used to perform region segmentation and boundary positioning on the alteration gradient vector field and alteration index matrix to obtain the final alteration detection map. It should be noted that since the alteration index matrix and alteration gradient vector field record the alteration intensity and its spatial evolution trend in scalar and vector forms respectively, but have not yet been transformed into detection results with clear spatial boundaries and intensity classifications, they cannot be directly used to guide mining decisions at the mine site. Therefore, it is necessary to segment and locate the altered area and output it in a visual form. Based on this, the technical solution of this application further performs region segmentation and boundary positioning on the alteration gradient vector field and alteration index matrix to obtain the alteration detection map. Through the above processing, continuously distributed alteration intensity data can be transformed into a detection map with alteration area boundaries, intensity classifications, and core target point annotations, providing intuitive technical support for exploration decisions of deep concealed veins in fluorite ore deposits.

[0051] More specifically, in a concrete example of this application, the alteration index matrix is ​​first segmented into initial regions based on maximizing inter-class variance to obtain alteration candidate masks. This process involves statistically analyzing the normalized index distribution of all pixels in the alteration index matrix, constructing an intensity histogram, and iteratively traversing all possible thresholds on the histogram using Otsu's method. The inter-class variance between the background and foreground classes is calculated at each candidate threshold. The value that maximizes the inter-class variance is selected as the global adaptive segmentation threshold. Based on this threshold, the alteration index matrix is ​​binarized, and connected pixels with alteration intensities higher than the threshold are marked as foreground, while the rest are marked as background, generating alteration candidate masks. The inter-class variance calculation is expressed as follows: in, This represents the inter-class variance between the background and foreground classes; a larger value indicates a more effective segmentation threshold. and These represent the proportions of background and foreground pixels in the erosion index matrix after thresholding. and These are the average erosion indices for background pixels and foreground pixels, respectively.

[0052] Subsequently, based on the flow direction constraint of the alteration gradient vector field, the alteration candidate mask is subjected to level set boundary evolution to obtain the alteration evolution boundary. This process uses the edge contour of the alteration candidate mask as the initial zero level set of the level set function, introduces the alteration gradient vector field as an external energy constraint term, and drives the zero level set to iteratively evolve along the ridge direction where the gradient vector magnitude changes drastically. This causes the boundary to converge towards the location where the alteration intensity gradient abruptly concentrates, correcting the initial segmentation edge blurring and breakage problems caused by noise. During the evolution process, the direction information of the gradient vector constrains the expansion direction of the boundary, preventing the boundary from excessively expanding into the background region with flat alteration intensity. The magnitude information of the gradient vector controls the convergence speed of the boundary; the boundary quickly locks at alteration edges with larger magnitudes and stops evolving at the boundary in uniform regions with smaller magnitudes. After evolution, a closing operation is performed on the topology to eliminate small holes and burrs, generating a spatially continuous and topologically complete alteration evolution boundary.

[0053] Finally, the alteration evolution boundary and alteration index matrix are overlaid with rendering and pseudo-color encoding to obtain the alteration detection map. This process uses the alteration evolution boundary as the spatially defined range, performing pseudo-color mapping on the alteration index matrix values ​​within it. The alteration index values ​​in the 0-1 range are mapped to a color table, with low-intensity areas mapped to cool tones and high-intensity areas to warm tones, thus visually presenting the spatial distribution of alteration intensity as a color gradient. Based on this, a non-maximum suppression algorithm is used to extract local peak points from the alteration index matrix as core alteration targets for annotation. The alteration evolution boundary, pseudo-color intensity map, and core target markers are then overlaid and composited to finally generate the alteration detection map.

[0054] In summary, the pattern recognition-based alteration detection system for fluorite ore surrounding rocks according to the embodiments of this application is explained. It solves the problems of difficult quantitative decomposition of mixed pixels in the alteration zone and blurred alteration boundaries by constructing a complete processing link from raw data from multiple sensors to alteration detection maps. The system first performs spatial geometric correction and radiometric calibration on hyperspectral data, RGB texture images, and lidar depth data to obtain aligned spectral cubes in a unified coordinate system. Then, it enhances the absorption characteristics of weakly altered minerals through spectral smoothing and multi-order differential operations. Based on this, it uses a deep unmixing network to decompose mixed pixels into abundance maps of each end-member mineral, achieving sub-pixel-level mineral component quantification. Furthermore, it integrates the abundance of multiple minerals into an alteration index matrix through weight mapping and intensity fusion, and introduces nonlinear coupling modeling of mineral interaction synergy features and alteration phase diversity entropy weights to solve the problem of reduced accuracy in identifying weakly altered zones due to the inability of linear models to characterize mineral symbiosis. Next, it extracts the alteration gradient vector field based on the structural tensor to characterize the alteration evolution trend. Finally, it combines adaptive threshold segmentation and gradient flow constraint boundary evolution to achieve accurate positioning and visualization output of the alteration region.

Claims

1. A pattern recognition-based system for detecting alteration of fluorite ore surrounding rocks, characterized in that, include: The data alignment module is used to perform spatial geometric correction and radiometric calibration on the raw sensor data stream to obtain an aligned spectral cube. The raw sensor data stream includes hyperspectral data, RGB texture images, and lidar depth data. The spectral enhancement module is used to perform spectral feature enhancement and multi-order differential preprocessing on the aligned spectral cube to obtain the enhanced feature tensor. The endmember extraction module is used to extract endmember components from the enhanced feature tensor based on a deep unmixing network to obtain a mineral abundance map. The alteration intensity quantification module is used to perform alteration intensity quantification modeling on mineral abundance maps to obtain an alteration index matrix characterizing the alteration intensity of the surrounding rock; The trend analysis module is used to perform alteration trend analysis on the alteration index matrix based on the spatial gradient tensor to obtain the alteration gradient vector field; The region localization module is used to perform region segmentation and boundary localization on the alteration gradient vector field and alteration index matrix to obtain the final alteration detection map.

2. The pattern recognition-based alteration detection system for fluorite ore surrounding rocks according to claim 1, characterized in that, The data alignment module includes: The geometric correction unit is used to perform depth-constrained image geometric correction on the RGB texture image in the original sensor data stream based on lidar depth data to obtain a geometrically corrected image. A pixel registration unit is used to perform cross-modal spatial pixel-level registration of hyperspectral data in the original sensor data stream based on a geometrically corrected image to obtain a registration spectral cube. The radiometric calibration unit is used to perform spectral dimension radiometric calibration on the registered spectral cube to obtain the aligned spectral cube.

3. The pattern recognition-based fluorite ore surrounding rock alteration detection system according to claim 1, characterized in that, The spectral enhancement module includes: The spectral smoothing unit is used to perform spectral smoothing on the aligned spectral cube to obtain smoothed spectral data. The differential extraction unit is used to perform multi-order differential extraction of smooth spectral data based on the central difference to obtain differential spectral components; The feature recombination unit is used to reconstruct the spatiotemporal-spectral feature dimensions of smoothed spectral data and differential spectral components to obtain an enhanced feature tensor.

4. The pattern recognition-based fluorite ore surrounding rock alteration detection system according to claim 1, characterized in that, The end-member extraction module includes: Non-negative coding units are used to perform non-negative coding feature mapping on the augmented feature tensor to obtain the latent abundance representation; Linear reconstruction unit, used to perform linear reconstruction of the latent abundance representation based on endmember spectral constraints to obtain the original abundance estimate; The normalization unit is used to perform full abundance summation constraint normalization on the original abundance estimate to obtain the mineral abundance map.

5. The pattern recognition-based fluorite ore surrounding rock alteration detection system according to claim 1, characterized in that, The erosion quantification module includes: The weighting mapping unit is used to perform mineral component weighting mapping on the mineral abundance map through a weight vector to obtain weighted abundance components. The intensity analysis unit is used to perform alteration intensity analysis on the weighted abundance components to obtain the original index matrix; The range normalization unit is used to perform intensity dynamic range normalization on the original index matrix to obtain the alteration index matrix.

6. The pattern recognition-based fluorite ore surrounding rock alteration detection system according to claim 1, characterized in that, The trend analysis module includes: Spatial convolution unit, used to perform spatial convolution on the alteration index matrix to obtain partial derivative components; Tensor building units are used to obtain the structure tensor matrix by combining second-order moments and Gaussian smoothing based on partial derivative components. The vector synthesis unit is used to perform evolution trend vector analysis and synthesis on the structural tensor matrix to obtain the alteration gradient vector field.

7. The pattern recognition-based fluorite ore surrounding rock alteration detection system according to claim 1, characterized in that, The area positioning module includes: The region segmentation unit is used to perform initial region segmentation on the alteration index matrix based on maximizing the inter-class variance to obtain the alteration candidate mask. Boundary evolution unit, used to perform level set boundary evolution on the alteration candidate mask based on the flow direction constraint of the alteration gradient vector field to obtain the alteration evolution boundary; The rendering and encoding unit is used to perform overlay rendering and pseudo-color encoding on the alteration evolution boundary and alteration index matrix to obtain the final alteration detection map.

8. The pattern recognition-based alteration detection system for fluorite ore surrounding rocks according to claim 5, characterized in that, The intensity analysis unit includes: The collaborative extraction subunit is used to extract mineral interaction collaborative features from the weighted abundance components to obtain collaborative interaction features. The entropy weight estimation subunit is used to estimate the alteration phase diversity weights of the weighted abundance components in order to obtain the alteration entropy weights. The nonlinear fusion subunit is used to perform nonlinear coupling fusion of weighted abundance components and cooperative interaction features based on etch entropy weights to obtain the original index matrix.

Images