Method and system for identifying mineral categories of deep borehole cores, computer storage medium

Abstract

Embodiments of the present application relate to the field of analyzing materials by optical means, and in particular to a method and system for identifying mineral categories of deep drilling cores, and a computer storage medium. The method and system provided by the embodiments of the present application can retain spatial structure information and continuous spectral information of the cores at the same time by obtaining hyperspectral high spatial resolution fusion image data of the deep drilling cores, thereby improving the mineral boundary blurring and spectral mixing problems caused by insufficient spatial resolution. The method and system can reduce boundary fragmentation and calculation redundancy by dividing the fusion image data into connected regions that are more consistent with the physical boundaries of minerals. The method and system can improve the discrimination ability of minerals that are easy to mix in complex geological scenes by aggregating the node features corresponding to the graph structure and the connected regions. Furthermore, the method and system can improve the reliability and interpretability of the mineral identification results by combining a standard mineral feature library and physical spectral rules to constrain and check the identification results.

Description

Technical Field

[0001] The embodiments of this application relate to the field of material analysis using optical means, and particularly to a method and system for identifying mineral types in deep borehole core samples, as well as a computer storage medium. Background Technology

[0002] The statements herein are provided merely as background information in connection with this application and do not necessarily constitute prior art.

[0003] Deep borehole cores are rock samples obtained from deep boreholes. They can directly reflect the original state of the strata. Analyzing the mineral types and spatial distribution of the cores helps to analyze the lithology of the strata, reservoir properties and structural information.

[0004] Existing methods for identifying minerals in core samples mainly include manual visual logging, microscopic observation, X-ray diffraction, and automatic identification methods based on hyperspectral data. Manual logging relies on experience, resulting in low efficiency and high subjectivity. While laboratory testing methods offer high accuracy, they are typically point-based and cannot achieve rapid, continuous, and high-resolution spatial identification of minerals across an entire core sample. Automatic identification methods based on hyperspectral imaging can utilize the characteristic absorption information of minerals in the visible to short-wave infrared bands to achieve non-contact identification, and are widely used in current core mineral identification practices. Summary of the Invention

[0005] A brief overview of this application is provided below to offer a basic understanding of certain aspects thereof. It should be understood that this overview is not an exhaustive summary of the application. It is not intended to identify key or essential parts of the application, nor is it intended to limit its scope. Its purpose is merely to present certain concepts in a simplified form as a prelude to the more detailed description that follows.

[0006] This application provides a method for identifying mineral categories in deep borehole cores, comprising the following steps: S10: acquiring hyperspectral and high spatial resolution fused image data of deep borehole cores; S20: dividing the hyperspectral and high spatial resolution fused image data to obtain multiple connected regions composed of several spatially adjacent pixels with similar spatial and spectral features; S30: determining the graph structure, wherein the graph structure uses connected regions as nodes and includes spatial adjacency relationships, boundary relationships, and spectral similarity relationships between connected regions; S40: determining the node features corresponding to the connected regions, wherein the node features include spectral features and spatial features; S50: aggregating the graph structure and node features to obtain a context-enhanced representation that retains its own spatial and spectral features while incorporating the features of its adjacent nodes; S60: determining the mineral category of each connected region based on the similarity between the context-enhanced representation and a standard mineral feature library, combined with physical spectral rule constraints; S70: determining the mineral category of the deep borehole core based on the mineral category of each connected region.

[0007] Compared to existing pixel-by-pixel recognition methods, the mineral classification method for deep borehole cores provided in this application, by acquiring high-spectral, high-spatial-resolution fused image data of deep borehole cores, can simultaneously retain the spatial structure information and continuous spectral information of the core, fundamentally improving the problems of blurred mineral boundaries and spectral mixing caused by insufficient spatial resolution; by dividing the fused image data into connected regions that better fit the physical boundaries of minerals, it can reduce boundary fragmentation and computational redundancy caused by mechanical slicing or pixel-by-pixel processing; by aggregating information from the graph structure and the node features corresponding to the connected regions, it can utilize the spatial adjacency relationships between connected regions and the mineral co-occurrence context information to improve the ability to distinguish easily confused minerals in complex geological scenes; and by combining the standard mineral feature library and physical spectral rules to constrain and verify the recognition results, it can improve the reliability and interpretability of the mineral recognition results, thereby achieving automated and high-precision identification of mineral categories in deep borehole cores.

[0008] This application, in another aspect, provides a mineral classification system for deep borehole core samples, comprising: a sensor configured to acquire hyperspectral and high spatial resolution fused image data of deep borehole core samples; and a processor configured to: divide the hyperspectral and high spatial resolution fused image data to obtain multiple connected regions composed of several spatially adjacent pixels with similar spatial and spectral features; determine a graph structure, the graph structure using connected regions as nodes and including spatial adjacency, boundary relationships, and spectral similarity relationships between connected regions; determine the node features corresponding to the connected regions, the node features including spectral features and spatial features; aggregate the graph structure and node features to obtain a context-enhanced representation that retains its own spatial and spectral features while incorporating the features of its adjacent nodes; determine the mineral classification of each connected region based on the similarity between the context-enhanced representation and a standard mineral feature library, and in conjunction with physical spectral rules constraints; and determine the mineral classification of the deep borehole core samples based on the mineral classification of each connected region.

[0009] The mineral classification identification system for deep borehole cores provided in this application acquires high-spectral, high-spatial-resolution fused image data of deep borehole cores through sensors. This system simultaneously preserves the spatial structure information and continuous spectral information of the core, fundamentally improving the problems of blurred mineral boundaries and spectral mixing caused by insufficient spatial resolution. By dividing the fused image data into connected regions that better fit the physical boundaries of the minerals through a processor, the system reduces boundary fragmentation and computational redundancy caused by mechanical slicing or pixel-by-pixel processing. By aggregating information from the graph structure and the node features corresponding to the connected regions, the system can utilize the spatial adjacency relationships between connected regions and the mineral co-occurrence context information to improve the ability to distinguish easily confused minerals in complex geological scenes. Furthermore, by combining a standard mineral feature library and physical spectral rules to constrain and verify the identification results, the system can improve the reliability and interpretability of the mineral identification results, thereby achieving automated and high-precision identification of mineral categories in deep borehole cores.

[0010] This application also provides a computer storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, perform the aforementioned method. Attached Figure Description

[0011] To further illustrate the above and other advantages and features of this application, the specific embodiments of this application will be described in more detail below with reference to the accompanying drawings. The drawings, together with the following detailed description, are included in and form a part of this specification. Elements having the same function and structure are indicated by the same reference numerals. It should be understood that these drawings only depict typical examples of this application and should not be considered as limiting the scope of this application.

[0012] Figure 1 This is a schematic diagram of the distribution of mineral categories identified by the identification method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the mineral categories and their corresponding typical reflectance spectral curves of core samples obtained at different depths using the method provided in the embodiments of this application.

[0013] Explanation of reference numerals in the attached figures: 1. Typical reflectance spectrum of kaolinite; 2. Typical reflectance spectrum of carbonates. Detailed Implementation

[0014] Exemplary embodiments of this application will be described below with reference to the accompanying drawings. For clarity and brevity, not all features of actual implementations are described in the specification. However, it should be understood that many implementation-specific decisions must be made in the development of any such actual embodiment to achieve the developer's specific goals, such as complying with constraints related to the system and business, and these constraints may vary depending on the implementation. Furthermore, it should be understood that while development work can be very complex and time-consuming, such development work is merely a routine task for those skilled in the art who benefit from the content of this application.

[0015] It should also be noted that, in order to avoid obscuring this application with unnecessary details, only the equipment structure and / or processing steps closely related to the solution according to this application are shown in the accompanying drawings, while other details that are not closely related to this application are omitted.

[0016] The following disclosure provides several different implementations or examples for carrying out this application. To simplify the disclosure of this application, specific examples of components and methods are described below. Of course, these are merely examples and are not intended to limit this application. In the description of the embodiments of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0017] In deep borehole core mineral identification, traditional pixel-by-pixel hyperspectral identification methods typically suffer from the following problems: Although hyperspectral images possess continuous spectral information, with limited spatial resolution, a single pixel can easily cover multiple minerals, leading to spectral mixing and weakening, shifting, or masking of diagnostic absorption bands; pixel-by-pixel identification is computationally intensive, and fragmentation and misclassification are prone to occur in boundary areas; identification based solely on the spectra of a single pixel or local block makes it difficult to utilize the inherent symbiotic, associated, and zoning patterns of minerals within geological bodies; and relying solely on deep learning models to output category results results in insufficient interpretability and difficulty in unifying with traditional mineral spectroscopic rules. Therefore, currently, there is no deep core mineral identification method that can simultaneously utilize the spatial structure information, continuous spectral information, spatial symbiotic relationships of minerals, and mineral physical spectral rules of core images to improve the accuracy, stability, and interpretability of identification.

[0018] To address the aforementioned problems, embodiments of this application provide a method for identifying mineral categories in deep borehole cores, comprising the following steps: S10: acquiring hyperspectral and high spatial resolution fused image data of deep borehole cores; S20: dividing the hyperspectral and high spatial resolution fused image data to obtain multiple connected regions composed of several spatially adjacent pixels with similar spatial and spectral features; S30: determining a graph structure, wherein the graph structure uses connected regions as nodes and includes spatial adjacency relationships, boundary relationships, and spectral similarity relationships between connected regions; S40: determining the node features corresponding to the connected regions, wherein the node features include spectral features and spatial features; S50: aggregating the graph structure and node features to obtain a context-enhanced representation that retains its own spatial and spectral features while incorporating the features of its adjacent nodes; S60: determining the mineral category of each connected region based on the similarity between the context-enhanced representation and a standard mineral feature library, combined with physical spectral rule constraints; S70: determining the mineral category of the deep borehole core based on the mineral category of each connected region.

[0019] Compared to existing pixel-by-pixel recognition methods, the mineral classification method for deep borehole cores provided in this application, by acquiring high-spectral, high-spatial-resolution fused image data of deep borehole cores, can simultaneously retain the spatial structure information and continuous spectral information of the core, fundamentally improving the problems of blurred mineral boundaries and spectral mixing caused by insufficient spatial resolution; by dividing the fused image data into connected regions that better fit the physical boundaries of minerals, it can reduce boundary fragmentation and computational redundancy caused by mechanical slicing or pixel-by-pixel processing; by aggregating information from the graph structure and the node features corresponding to the connected regions, it can utilize the spatial adjacency relationships between connected regions and the mineral co-occurrence context information to improve the ability to distinguish easily confused minerals in complex geological scenes; and by combining the standard mineral feature library and physical spectral rules to constrain and verify the recognition results, it can improve the reliability and interpretability of the mineral recognition results, thereby achieving automated and high-precision identification of mineral categories in deep borehole cores.

[0020] In some embodiments, in step S10, the hyperspectral high spatial resolution fused image data is the data obtained by fusing hyperspectral image data and high spatial resolution image data. The data includes spatial structure information and continuous spectral information of deep borehole cores. By simultaneously preserving the spatial details and spectral features of the core in the source data, the spectral mixing phenomenon caused by insufficient spatial resolution can be effectively suppressed, which is beneficial for accurately delineating mineral boundaries.

[0021] In some embodiments, step S20 further includes the following steps: S21: extracting principal component pseudo-color features, spectral difference features, and edge texture features from the hyperspectral high spatial resolution fused image data; S22: determining a spatial-spectral joint distance metric based on the principal component pseudo-color features, spectral difference features, and edge texture features; S23: dividing the hyperspectral high spatial resolution fused image data according to the spatial-spectral joint distance metric to obtain multiple connected regions. By considering the differences in principal component pseudo-color features, spectral features, and edge texture features of the image data, the divided connected regions can adaptively conform to the mineral grain boundaries and fracture morphology in the core, thereby reducing property contamination caused by crossing physical boundaries and fragmentation misjudgment in subsequent identification.

[0022] Figure 1 This is a schematic diagram of the distribution of mineral categories identified by the identification method provided in the embodiments of this application, such as... Figure 1 As shown, different colored patches represent different mineral categories: yellow patches represent short-wave illite, dark blue patches represent long-wave illite, red patches represent kaolinite and dickite, purple patches represent carbonates, and green patches represent chlorite. Figure 1This reflects the spatial distribution of various hydrothermal alteration minerals on the surface of long core sections. Because the identification method provided in this application extracts principal component false-color features, spectral difference features, and edge texture features from hyperspectral, high spatial resolution fused image data, it enables… Figure 1 The identification results of the edges of each color block in the image do not present the mosaic or random fragmentation commonly seen in traditional pixel-by-pixel classification. Instead, they present a form that is highly consistent with the real physical texture of the core, such as vein-like occurrence and fissure filling. This indicates that by extracting edge texture and spectral difference features, the image was successfully divided into connected regions that fit the physical boundaries of the minerals, thus overcoming the problem of blurred mineral boundaries caused by insufficient spatial resolution from the source.

[0023] In some embodiments, in step S21, the principal component pseudo-color features of the hyperspectral high spatial resolution fused image data are extracted. This enables principal component analysis (PCA) to be performed on the fused image data cube along the spectral dimension. The first three principal components can be extracted and mapped to a pseudo-color image. The obtained pseudo-color image can compressively represent the main spatial variations and edge information in the fused image.

[0024] In some embodiments, in step S21, the spectral difference features of the hyperspectral high spatial resolution fused image data are used to describe the degree of difference between pixels in the continuous spectrum, and spectral information divergence (SID) can be used as a measure of difference. For the spectral vector of a pixel and the spectral vector of the cluster center, their spectral information divergence can be calculated to characterize the difference in their spectral probability distributions. By introducing spectral difference features, pixels within the same connected region can have higher consistency in material composition.

[0025] In some embodiments, in step S21, the edge texture features of the hyperspectral high spatial resolution fused image data are used to characterize the gradient change intensity at locations such as cracks, mineral boundaries, and vein edges in the image. An edge detection operator can be applied to the high-frequency components of the pseudo-color image or fused image obtained by mapping to extract edge texture features, thereby suppressing cell clustering that crosses obvious physical boundaries during superpixel segmentation.

[0026] In some embodiments, in step S22, the joint spatial-spectral distance metric can be determined as follows: ;in: is the Euclidean distance between the pixel spatial coordinates and the cluster center coordinates; The spectral information divergence between the pixel spectral vector and the cluster center spectral vector; This is a term representing the difference in edge texture gradients or a boundary penalty. This is a joint spatial-spectral distance metric; α, β, and γ are non-negative weighting parameters, which can be set based on empirical values, validation set optimization results, or different core scenarios. By simultaneously considering spatial compactness, spectral homogeneity, and edge constraints, the final connected region boundary can more closely resemble the actual grain boundary.

[0027] In some embodiments, in step S23, an improved Simple Linear Iterative Clustering (SLIC) algorithm can be used to perform superpixel clustering segmentation on the fused image data: First, the cluster centers are initialized. Then, the spatial-spectral joint distance between the pixels of the fused image data and the cluster centers is calculated within a local search window, and the pixels are assigned to the cluster centers with the smallest distance. Subsequently, the spatial location and feature values ​​of the cluster centers are updated, and the above iterative process is repeated until the model converges or the preset number of iterations is reached, so as to achieve superpixel clustering segmentation.

[0028] In some embodiments, after superpixel segmentation is completed, the connected regions can be further modified by merging excessively small connected regions into neighboring connected regions or splitting non-connected regions, thereby effectively ensuring the spatial connectivity of the connected regions. Through the above modifications, millions of pixels can be compressed into a set of thousands of physically meaningful connected regions.

[0029] In some embodiments, in step S23, multiple connected regions are made to correspond as closely as possible to mineral grains, mineral aggregates, veinlets, fracture boundaries, or other physically significant micro-geological entities in the rock core.

[0030] In some embodiments, in step S30, the spatial adjacency, boundary relationships, and spectral similarity relationships between connected regions are organized into a graph structure, which facilitates subsequent aggregation of contextual information through graph neural networks.

[0031] In some embodiments, step S30 further includes the following steps: S31: determining that any two connected regions share a spatial boundary; S32: establishing a connecting edge between corresponding nodes of the shared boundary; determining the length of the shared boundary based on the connecting edge; S33: determining the edge weight of the connecting edge based on at least one of the length of the shared boundary, the spectral similarity between connected regions, and the boundary gradient intensity; S34: determining the graph structure based on the edge weight of the connecting edge. Utilizing the contact geometric features and physical property differences between connected regions to construct topological relationships can transform discrete identification units into a structured whole with geological relevance, thereby providing a topological carrier for simulating the spatial co-existence and associated occurrence of minerals in real geological bodies.

[0032] In some embodiments, in step S32, if two connected regions have a common boundary in the image plane, they are considered to be adjacent, and an undirected edge is established between the corresponding nodes; directed edges can also be established, but undirected edges are preferred to represent the bidirectional relationship between adjacent connected regions.

[0033] In some embodiments, in step S33, a longer shared boundary indicates a higher degree of contact between two connected regions; more similar spectra among the connected regions indicate closer material properties; and a stronger boundary gradient can reduce the information propagation intensity between two connected regions to avoid excessive propagation across obvious physical boundaries. The graph structure determined in the above manner can incorporate contextual information about mineral coexistence, association, and zoning in subsequent identification.

[0034] In some embodiments, in step S40, the node features are vector representations used to characterize the attributes of each connected region, including at least the spectral information and spatial morphological information of the connected region.

[0035] In some embodiments, step S40 further includes the following steps: S41: determining the representative spectrum of the pixel spectrum within each connected region; S42: determining the spectral features of the connected region based on the representative spectrum; S43: determining the spatial texture and morphological information of the image block corresponding to the connected region, and determining the spatial features of the connected region based on the spatial texture and morphological information; S44: determining the node features corresponding to the connected region based on the spectral features and spatial features. The above settings can improve the separability of complex mineral assemblages.

[0036] In some embodiments, in step S41, statistical aggregation can be performed on the spectra of all pixels contained within each connected region to generate a representative spectrum for that connected region. By statistically aggregating the spectra within connected regions to suppress local noise interference, the sensitivity to component identification can be improved.

[0037] Preferably, a robust median spectrum can be used as the representative spectrum, that is, the median value of all pixels in each band within the connected region is taken to obtain a one-dimensional vector of length B (B represents the number of bands in the hyperspectral image data): ;in, This represents the representative value of the i-th connected region in the B-th band. This represents the spectral vector of the i-th connected region.

[0038] Using a robust median spectrum can reduce the impact of outliers, shadowed pixels, and local noise on the representative spectrum. In addition to the median spectrum, robust mean spectrum, truncated mean spectrum, or principal component representative spectrum can also be used.

[0039] In some embodiments, in step S42, a representative spectrum can be input into a spectral feature extraction network to obtain the spectral features of the connected regions.

[0040] In some embodiments, in step S42, a one-dimensional real vector of length B of the connected region can be used as the representative spectrum input to the spectral feature extraction network to output a spectral feature vector. This spectral feature extraction network can be constructed using a one-dimensional convolutional neural network (1D-CNN) combined with a channel attention mechanism. Since both are known model structures in the art, those skilled in the art can implement them according to the input, output, and network functions disclosed in the specification.

[0041] In a specific feasible structure, the spectral feature extraction network sequentially includes: a first one-dimensional convolutional layer with a kernel size of 5 and 32 output channels; a second one-dimensional convolutional layer with a kernel size of 5 and 64 output channels; and a third one-dimensional convolutional layer with a kernel size of 3 and 128 output channels. Each convolutional layer is followed by a batch normalization layer and a ReLU (Modified Linear Unit) activation function. This is then connected to a channel attention module for assigning different weights to band channels, a global average pooling layer, and finally a fully connected layer for outputting the spectral feature vector. Through this network structure, 1D-CNN can effectively extract local peak and valley morphological features from spectral curves, such as the absorption bands in mineral diagnostics, including the location of absorption peaks, the depth of absorption valleys, and the variation trends of adjacent bands. Furthermore, it utilizes a channel attention mechanism to enhance the role of key bands in feature representation.

[0042] In some embodiments, in step S43, spatial texture and morphological information of the image blocks corresponding to the connected regions are extracted and input into the spatial feature extraction network to obtain the spatial features of the connected regions. First, the smallest bounding rectangle image block of each connected region is extracted as the network input. This two-dimensional image block can be derived from a primary component pseudo-color image, a high-order principal component composite image of a fused image, or other image representations that can reflect spatial texture and boundary morphology. At the same time, the non-connected regions within the bounding rectangle are masked using a zero-filling method to preserve the true shape of the connected regions. After receiving the input, the spatial feature extraction network outputs a spatial feature vector.

[0043] In some embodiments, the spatial feature extraction network may employ a lightweight two-dimensional convolutional neural network (2D-CNN) known in the art. Such models have low computational cost and are suitable for efficient processing of a large number of connected image blocks. Of course, other known lightweight 2D-CNN structures capable of extracting spatial texture and morphological features may also be used. Through this network, spatial information such as the granularity, texture coarseness, cleavage direction, boundary morphology, and vein features of the spatial feature extraction network can be effectively extracted, thereby compensating for the shortcomings of relying solely on spectral information.

[0044] In some embodiments, in step S44, the spectral feature vector and the spatial feature vector can be concatenated and input into a multilayer perceptron (MLP) for dimensionality reduction mapping to obtain node feature vectors of uniform dimension. ,in: The spectral characteristics of the i-th connected region; The spatial characteristics of the i-th connected region; Let be the fused feature of the i-th node; MLP indicates that a multilayer perceptron is used for dimensionality reduction mapping.

[0045] In some embodiments, the MLP can consist of one or more fully connected layers and can be used in conjunction with the ReLU activation function and Dropout layers to output node features of a uniform dimension, such as 128-dimensional node features.

[0046] In some embodiments, in step S50, the context-enhanced representation may also incorporate node features from its neighboring nodes and even a larger neighborhood.

[0047] In some embodiments, in step S50, the graph structure and node features can be input into a graph neural network for information aggregation. The graph neural network includes at least one of a graph convolutional neural network, a graph attention network, and a graph sampling aggregation network. Preferably, GraphSAGE (graph sampling aggregation method) or GAT (graph attention network) models can be used, where GraphSAGE can improve processing efficiency in large graph scenarios by sampling and aggregating neighboring nodes; GAT can adaptively learn the importance of different neighboring nodes through an attention mechanism. Since the above models are all known in the art, those skilled in the art can select and implement them as needed.

[0048] like Figure 1 As shown, Figure 1 This reflects the associated relationships between long-wave illite, short-wave illite, chlorite, and carbonates, indicating that this application utilizes graph neural networks to aggregate mineral symbiotic context information. This results in mineral assemblages exhibiting obvious zonal or stratified distribution patterns in regions with different alteration intensities, achieving continuous identification that conforms to geological laws. Furthermore, it demonstrates that graph structures can effectively utilize the adjacency relationships between connected regions to correct noise interference and isolated misjudgments that are prone to occur in single-pixel identification, making the identification results more geologically significant.

[0049] In some embodiments, a two-layer GraphSAGE network can be used for context aggregation. After two layers of aggregation, a context-enhanced representation of each node can be obtained. Through information propagation in the graph neural network, each connected region, when determining its own mineral category, utilizes not only its own representative spectrum and morphological information, but also the category tendencies and co-occurrence characteristics of its adjacent connected regions. For example, if the spectrum of a connected region is similar to that of multiple minerals, but the surrounding connected regions generally exhibit a certain specific alteration combination, the reliability of the connected region's determination of the corresponding mineral category can be improved.

[0050] In some embodiments, step S60 further includes the following steps: S61: determining the similarity between the context-enhanced representation and the standard features of each mineral in the standard mineral feature library, and determining the candidate mineral category based on the similarity; S62: determining the absorption features in the original spectrum of the corresponding connected region; S63: performing consistency verification based on the physical spectral rules corresponding to the absorption features and the candidate mineral category; S64: determining the mineral category of the connected region based on the verification result. By cross-validating the context-enhanced representation with the absorption features in the original spectrum, the prediction results can be verified using the inherent absorption features of the minerals, thereby improving the reliability and interpretability of the recognition results.

[0051] Figure 2 This is a schematic diagram illustrating the mineral categories and corresponding typical reflectance spectral curves of core samples obtained at different depths using the method provided in the embodiments of this application. Figure 2 As shown, in the continuous core section from 402.37 meters to 410.41 meters, different colored patches represent different mineral categories: yellow patches represent medium-aluminum sericite, dark blue patches represent low-aluminum sericite, light blue patches represent kaolinite, purple patches represent carbonates, and green patches represent chlorite. In the typical reflectance spectrum curve 1 of light blue kaolinite, the spectrum exhibits a clear double-peak absorption characteristic near 2200 nm. The typical reflectance spectrum curve 2 of purple carbonate shows a strong single absorption valley near 2330 nm, indicating that each identified mineral category has significant and standard physical diagnostic absorption characteristics. This one-to-one correspondence proves the necessity and scientific validity of the physical spectral rules constrained in the S60 step.

[0052] In some embodiments, in step S61, the cosine similarity between the context-enhanced representation of the connected region to be identified and the standard feature vectors of each mineral in the standard mineral feature library can be calculated: ;in, This represents the standard eigenvector of mineral class c. This represents a context-enhanced representation, where Sim represents cosine similarity. Based on the cosine similarity score, it outputs the top few (e.g., the top 3) candidate mineral categories and their similarity scores.

[0053] In some embodiments, in step S62, the original spectrum is preferably a representative spectrum of the corresponding connected region, or it can be the average value of the spectrum of typical pixels within the connected region. The absorption features include at least one of the following: the center wavelength position of the absorption band, the band depth of the absorption band, the bandwidth of the absorption band, the symmetry of the absorption band, and the band depth ratio between different absorption bands. The original spectrum can first undergo continuum removal, and then local minima can be detected within a preset wavelength window to calculate the parameters included in the absorption features.

[0054] like Figure 2 As shown, by extracting key parameters such as the center wavelength, band depth, and band depth ratio of the representative spectrum, a rigorous consistency check was performed on the candidate categories output by the graph neural network. This ensures that even in complex boundary regions, the final output mineral category conforms to its spectroscopic physical properties, improving the interpretability of the identification results and ensuring high reliability of the identification results in engineering applications.

[0055] In some embodiments, in step S63, the physical spectral rules may include: whether the center wavelength of the characteristic absorption band is within a preset range; whether the band depth of the characteristic absorption band is greater than a threshold; whether the ratio of the characteristic absorption bands meets the threshold requirement; whether multiple characteristic absorption bands exist simultaneously; and whether the symmetry of the characteristic absorption bands meets the requirements. For example, for Al-OH minerals (aluminum hydroxyl minerals), it may be required that they have a characteristic absorption band around approximately 2200 nm; for carbonate minerals, it may be required that they have a diagnostic absorption band around approximately 2330 nm; and for minerals such as chlorite, it may be required that the band depth ratio of a specific band meets a preset threshold.

[0056] In some embodiments, in step S64, if the first candidate mineral category satisfies the corresponding physical spectral rules, it is determined as the final mineral category of the connected region; if the first candidate category does not satisfy the rules, the second and third candidate categories are checked sequentially; if none of them satisfy the rules, the connected region can be marked as an abnormal category, an unknown mineral, or a category requiring manual verification. This approach avoids making judgments that contradict physical spectral rules based solely on feature similarity, thus improving the reliability of the identification results.

[0057] Another embodiment of this application provides a mineral classification system for deep borehole cores, comprising: a sensor configured to acquire hyperspectral high spatial resolution fused image data of deep borehole cores; and a processor configured to: divide the hyperspectral high spatial resolution fused image data to obtain multiple connected regions composed of several spatially adjacent pixels with similar spatial and spectral features; determine a graph structure, the graph structure using connected regions as nodes and including spatial adjacency relationships, boundary relationships, and spectral similarity relationships between connected regions; determine node features corresponding to the connected regions, the node features including spectral features and spatial features; aggregate the graph structure and node features to obtain a context-enhanced representation that retains its own spatial and spectral features while incorporating the features of its adjacent nodes; determine the mineral classification of each connected region based on the similarity between the context-enhanced representation and a standard mineral feature library, and in conjunction with physical spectral rule constraints; and determine the mineral classification of the deep borehole core based on the mineral classification of each connected region.

[0058] The mineral classification identification system for deep borehole cores provided in this application acquires high-spectral, high-spatial-resolution fused image data of deep borehole cores through sensors. This system simultaneously preserves the spatial structure information and continuous spectral information of the core, fundamentally improving the problems of blurred mineral boundaries and spectral mixing caused by insufficient spatial resolution. By dividing the fused image data into connected regions that better fit the physical boundaries of the minerals through a processor, the system reduces boundary fragmentation and computational redundancy caused by mechanical slicing or pixel-by-pixel processing. By aggregating information from the graph structure and the node features corresponding to the connected regions, the system can utilize the spatial adjacency relationships between connected regions and the mineral co-occurrence context information to improve the ability to distinguish easily confused minerals in complex geological scenes. Furthermore, by combining a standard mineral feature library and physical spectral rules to constrain and verify the identification results, the system can improve the reliability and interpretability of the mineral identification results, thereby achieving automated and high-precision identification of mineral categories in deep borehole cores.

[0059] In some embodiments, the processor is further configured to: extract principal component pseudo-color features, spectral difference features, and edge texture features from the hyperspectral high spatial resolution fused image data; determine a spatial-spectral joint distance metric based on the principal component pseudo-color features, spectral difference features, and edge texture features; and divide the hyperspectral high spatial resolution fused hyperspectral image data according to the spatial-spectral joint distance metric to obtain multiple connected regions. By considering the differences in principal component pseudo-color features, spectral features, and edge texture features of the image data, the divided connected regions can adaptively conform to the mineral grain boundaries and fracture morphologies in the core, thereby reducing property contamination caused by crossing physical boundaries and fragmentation misjudgments in subsequent identification.

[0060] In some embodiments, the processor is further configured to: determine that any two connected regions share a spatial boundary; establish a connecting edge between corresponding nodes of the shared boundary; determine the length of the shared boundary based on the connecting edge; determine the edge weight of the connecting edge based on at least one of the length of the shared boundary, the spectral similarity between connected regions, and the boundary gradient intensity; and determine the graph structure based on the edge weight of the connecting edge. Utilizing the contact geometry features and physical property differences between connected regions to construct topological relationships can transform discrete identification units into a structured whole with geological relevance, thereby providing a topological carrier for simulating the spatial co-existence and associated occurrence of minerals in real geological bodies.

[0061] In some embodiments, the processor is further configured to: determine a representative spectrum of the pixel spectrum within each connected region; determine the spectral features of the connected region based on the representative spectrum; determine the spatial texture and morphological information of the image block corresponding to the connected region; determine the spatial features of the connected region based on the spatial texture and morphological information; and determine the node features corresponding to the connected region based on the spectral features and spatial features. The above configuration can improve the separability of complex mineral assemblages.

[0062] In some embodiments, the processor is further configured to: determine the similarity between the context-enhanced representation and the standard features of each mineral in the standard mineral feature library; determine candidate mineral categories based on the similarity; determine the absorption features in the original spectrum of the corresponding connected region; perform consistency verification based on the physical spectral rules corresponding to the absorption features and the candidate mineral categories; and determine the mineral category of the connected region based on the verification result. By cross-validating the context-enhanced representation with the absorption features in the original spectrum, the prediction results can be verified using the inherent absorption features of the minerals, thereby improving the reliability and interpretability of the recognition results.

[0063] Embodiments of this application also provide a computer storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, perform the aforementioned method.

[0064] Regarding the embodiments of this application, it should also be noted that, without conflict, the embodiments of this application and the features in the embodiments can be combined with each other to obtain new embodiments.

[0065] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. The scope of protection of this application shall be determined by the scope of the claims.

Claims

1. A method for identifying mineral types in deep borehole core samples, characterized in that, It includes the following steps: S10: Acquire the hyperspectral and high spatial resolution fused image data of the deep borehole core; S20: Divide the hyperspectral high spatial resolution fused image data to obtain multiple connected regions composed of several spatially adjacent pixels with similar spatial and spectral features; S30: Determine the graph structure, wherein the graph structure uses the connected regions as nodes and includes the spatial adjacency relationship, boundary relationship and spectral similarity relationship between the connected regions; S40: Determine the node features corresponding to the connected region, wherein the node features include spectral features and spatial features; S50: Aggregate the information of the graph structure and the node features to obtain a context-enhanced representation that retains its own spatial spectral features and has the features of each connected region after fusing the features of its neighboring nodes; S60: Based on the similarity between the context-enhanced representation and the standard mineral feature library, and combined with physical spectral rule constraints, determine the mineral category of each connected region; S70: Determine the mineral category of the deep borehole core according to the mineral category of each of the connected regions.

2. The identification method according to claim 1, characterized in that, In step S10, the hyperspectral high spatial resolution fused image data is the data obtained by fusing hyperspectral image data and high spatial resolution image data. The data includes the spatial structure information and continuous spectral information of the deep borehole core.

3. The identification method according to claim 1, characterized in that, Step S20 also includes the following steps: S21: Extract the principal component pseudo-color features, spectral difference features, and edge texture features from the hyperspectral high spatial resolution fused image data; S22: Determine the spatial-spectral joint distance metric based on the principal component pseudo-color features, the spectral difference features, and the edge texture features; S23: Based on the spatial-spectral joint distance metric, the hyperspectral high spatial resolution fused image data is divided to obtain multiple connected regions.

4. The identification method according to claim 1, characterized in that, Step S30 also includes the following steps: S31: Determine that any two of the connected regions share a spatial boundary; S32: Establish connecting edges between corresponding nodes of the shared boundary; determine the length of the shared boundary based on the connecting edges; S33: Determine the edge weight of the connecting edge based on at least one of the length of the shared boundary, the spectral similarity between the connected regions, and the boundary gradient strength; S34: Determine the graph structure based on the edge weights of the connecting edges.

5. The identification method according to claim 1, characterized in that, Step S40 also includes the following steps: S41: Determine the representative spectrum of the pixel spectrum within each of the connected regions; S42: Determine the spectral characteristics of the connected region based on the representative spectrum; S43: Determine the spatial texture and morphological information of the image block corresponding to the connected region, and determine the spatial features of the connected region based on the spatial texture and the morphological information; S44: Determine the node features corresponding to the connected region based on the spectral features and the spatial features.

6. The identification method according to claim 1, characterized in that, Step S60 also includes the following steps: S61: Determine the similarity between the context-enhanced representation and the standard features of each mineral in the standard mineral feature library, and determine the candidate mineral category based on the similarity; S62: Determine the absorption characteristics in the original spectrum of the corresponding connected region; S63: Perform consistency verification based on the absorption characteristics and the physical spectral rules corresponding to the candidate mineral categories; S64: Determine the mineral category of the connected region based on the verification results.

7. A mineral classification system for deep borehole core samples, characterized in that, It includes: The sensor is configured to acquire hyperspectral, high spatial resolution fused image data of deep borehole cores; The processor is configured to: The hyperspectral high spatial resolution fused image data is divided to obtain multiple connected regions composed of several spatially adjacent pixels with similar spatial and spectral features; A graph structure is defined, wherein the connected regions are used as nodes, and the graph structure includes the spatial adjacency relationship, boundary relationship and spectral similarity relationship between the connected regions; Determine the node features corresponding to the connected region, wherein the node features include spectral features and spatial features; By aggregating the graph structure and the node features, a context-enhanced representation is obtained that retains its own spatial spectral features and incorporates the features of its neighboring nodes for each connected region. Based on the similarity between the context-enhanced representation and the standard mineral feature library, and in conjunction with physical spectral rule constraints, the mineral category of each connected region is determined; The mineral category of the deep borehole core is determined based on the mineral category of each of the connected regions.

8. The identification system according to claim 7, characterized in that, The processor is also configured to: Principal component pseudo-color features, spectral difference features, and edge texture features are extracted from the hyperspectral high spatial resolution fused image data. Based on the principal component pseudo-color features, the spectral difference features, and the edge texture features, a spatial-spectral joint distance metric is determined. The hyperspectral high spatial resolution fused hyperspectral image data is then divided according to the spatial-spectral joint distance metric to obtain multiple connected regions.

9. The identification system according to claim 7, characterized in that, The processor is also configured to: Determine that any two of the connected regions share a spatial boundary; Establish connecting edges between corresponding nodes of the shared boundary; determine the length of the shared boundary based on the connecting edges; The edge weight of the connecting edge is determined based on at least one of the length of the shared boundary, the spectral similarity between the connected regions, and the boundary gradient strength. The graph structure is determined based on the edge weights of the connecting edges.

10. The identification system according to claim 7, characterized in that, The processor is also configured to: Determine the representative spectrum of the pixel spectrum within each connected region, determine the spectral features of the connected region based on the representative spectrum, determine the spatial texture and morphological information of the image block corresponding to the connected region, and determine the spatial features of the connected region based on the spatial texture and morphological information; Based on the spectral features and the spatial features, the node features corresponding to the connected region are determined.

11. The identification system according to claim 7, characterized in that, The processor is also configured to: The similarity between the context-enhanced representation and the standard features of each mineral in the standard mineral feature library is determined, and candidate mineral categories are determined based on the similarity. The absorption features in the original spectrum of the corresponding connected region are determined, and the consistency between the absorption features and the physical spectral rules corresponding to the candidate mineral categories is checked. Based on the check results, the final mineral category of the connected region is determined.

12. A computer storage medium, characterized in that, The computer storage medium stores a computer program, the computer program including program instructions, which, when executed by a processor, perform the method as described in any one of claims 1-6.

Images