Geologic image-based magmatic rock mineral analysis method and system

Abstract

The invention discloses a magmatic rock mineral analysis method and system based on geological images, and relates to the technical field of mineral analysis. A magmatic rock mineral analysis system based on geological images comprises a magmatic rock feature analysis module and a magmatic rock comprehensive analysis module. According to the method, the magmatic rock mineral image and the magmatic rock geological auxiliary data are fused, so that the mineralization evolution history and the future mineralization potential of the magmatic rock can be disclosed; high-precision automatic recognition and classification of minerals in magmatic rocks are achieved through the mineral classification model, manual intervention is reduced, and the recognition efficiency and accuracy are improved; the extracted mineral crystallization characteristics are integrated with the oxygen fugacity characteristics, so that the conjoint analysis of the physical structure and the chemical property of the mineral is realized, and more comprehensive data support is provided for subsequent geological evolution research and mineral product prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of mineral analysis, and particularly relates to a method and system for analyzing magmatic rock minerals based on geological images. Background Art

[0002] The existing methods for analyzing magmatic rock minerals mainly rely on manual observation of rock thin sections and elemental probe analysis. The process is complex, inefficient, and easily affected by subjective factors, making it difficult to achieve the automation and systematization of mineral identification and evolution analysis. In addition, the traditional methods lack the in-depth utilization of geological image information and are difficult to reveal the relationship between mineral microstructure and geological evolution. Therefore, it is necessary to conduct a fusion analysis of the multi-dimensional information of magmatic rock minerals, perform a spatio-temporal evolution analysis of magmatic rock minerals, reveal the evolution process of magmatic rocks under different geological periods and tectonic backgrounds, clarify the formation sequence, crystallization environment and variation law of minerals, so as to improve the understanding level of magmatic activity mechanisms and deep geological processes. Summary of the Invention

[0003] The present invention aims to provide a method and system for analyzing magmatic rock minerals based on geological images, and conduct a fusion analysis of the multi-dimensional information of magmatic rock minerals.

[0004] A method for analyzing magmatic rock minerals based on geological images includes the following steps: Based on the magmatic rock to be analyzed, obtain the magmatic rock mineral image; extract the surrounding geological components of the magmatic rock to be analyzed to obtain the auxiliary geological data of the magmatic rock; Extract mineral characteristics based on the magmatic rock mineral image and the magmatic rock geological data to obtain the magmatic rock mineral characteristics; the magmatic rock mineral characteristics include the magmatic rock mineral crystallization characteristics and the magmatic rock mineral coexisting oxygen fugacity characteristics; Conduct retrospective mineral analysis on the magmatic rock to be analyzed based on the magmatic rock mineral characteristics to obtain the magmatic rock retrospective characteristics; Conduct spatio-temporal evolution analysis on the magmatic rock to be analyzed based on the magmatic rock mineral characteristics to obtain the magmatic rock evolution characteristics; Conduct comprehensive magmatic rock mineral analysis based on the magmatic rock mineral characteristics, the magmatic rock retrospective characteristics and the magmatic rock evolution characteristics to obtain the full-process magmatic rock mineral analysis results.

[0005] As a preferred technical solution of the present invention, the specific steps of extracting mineral characteristics based on the magmatic rock mineral image and the magmatic rock geological data include: Use a pre-trained mineral classification model to classify the magmatic rock mineral image to obtain the mineral classification image G n , n = 1, 2,..., N; N is the total number of mineral categories after mineral identification of the magmatic rock mineral image, and each mineral classification image G nCorresponding to a classified or unclassified mineral classification; denote the mineral classification image G n The corresponding part in the magmatic rock to be analyzed is the magmatic rock block F n ; For the unclassified mineral classification image G n Perform mineral classification, and the specific steps are as follows: Based on the unclassified mineral classification image G n , obtain the special BSE image B n of the magmatic rock block F n and the EPMA mineral characteristics E n ; Based on the special BSE image B n and the EPMA mineral characteristics E n perform mineral classification annotation on the unclassified mineral classification image G n , and convert the unclassified mineral classification image G n into a mineral classification image G n with classified minerals.

[0006] As a preferred technical solution of the present invention, the specific steps for further analyzing the mineral classification image G n with classified minerals include: Perform grayscale processing on the mineral classification image G n to obtain the mineral classification grayscale image H n ; Use guided filtering to segment mineral grains and adjust the CLAHE window output for the mineral classification grayscale image H n to obtain the enhanced mineral classification grayscale image H n '; Calculate the gray-level co-occurrence matrix based on the enhanced mineral classification grayscale image H n ' to obtain the mineral crystallization matrix Z n ; Take the centers of adjacent grains in the enhanced mineral classification grayscale image H n ' as nodes and the grain boundary contact length as the edge weight, calculate the network topology parameters to obtain the mineral crystallization topology parameter C n ; Combine the mineral crystallization matrix Z n , the mineral crystallization topology parameter C n and output the magmatic rock mineral crystallization characteristics J n ; The specific steps for calculating the oxygen fugacity based on the labeled mineral classification labels of the mineral classification image G n include: Obtain the EPMA mineral characteristics E n of the mineral classification image G n , and calculate the oxygen fugacity based on the EPMA mineral characteristics E n to obtain the mineral classification oxygen fugacity O n ; Based on the mineral classification oxygen fugacity On Output the co-oxygen fugacity characteristics D of magmatic rock minerals n ; Combine all the crystallization characteristics J of magmatic rock minerals n and the co-oxygen fugacity characteristics D of magmatic rock minerals n to obtain the characteristics of magmatic rock minerals.

[0007] As a preferred technical solution of the present invention, the specific steps of performing retrospective mineral analysis and spatio-temporal evolution analysis on the magmatic rock to be analyzed based on the characteristics of magmatic rock minerals include: Construct a dual-channel and two-way mineral analysis network model, which includes a retrospective mineral analysis layer and a spatio-temporal evolution analysis layer; In the retrospective mineral analysis layer, based on the adversarial generative network, extract the auxiliary geological data of the magmatic rock for inverse mapping to obtain the retrospective geological conditions of the magmatic rock; based on the retrospective geological conditions of the magmatic rock and the characteristics of magmatic rock minerals, perform inverse time-step evolution to obtain the retrospective characteristics of the magmatic rock; In the spatio-temporal evolution analysis layer, based on the spatio-temporal Transformer mechanism, perform predictive analysis on the auxiliary geological data of the magmatic rock to obtain the predicted geological conditions of the magmatic rock; based on the predicted geological conditions of the magmatic rock and the characteristics of magmatic rock minerals, perform predictive time-step evolution to obtain the evolution characteristics of the magmatic rock.

[0008] As a preferred technical solution of the present invention, the specific steps of performing comprehensive magmatic rock mineral analysis based on the characteristics of magmatic rock minerals, the retrospective characteristics of magmatic rock, and the evolution characteristics of magmatic rock include: Construct a digital twin model of the magmatic rock based on the magmatic rock to be analyzed; For the digital twin model of the magmatic rock, introduce the characteristics of magmatic rock minerals, the retrospective characteristics of magmatic rock, and the evolution characteristics of magmatic rock for feature embedding to obtain an updated digital twin model of the magmatic rock; Generate the full-process magmatic rock mineral analysis results based on the updated digital twin model of the magmatic rock.

[0009] As a preferred technical solution of the present invention, the specific steps of training the mineral classification model include: Collect several groups of mineral classification training samples; each group of mineral classification training samples contains mineral type images and corresponding annotation labels; combine several groups of mineral classification training samples to obtain a mineral classification training set; Input the mineral classification training set into the CNN model for model training to obtain an initial mineral classification model; perform model evaluation on the initial mineral classification model. If the recognition accuracy of the initial mineral classification model reaches the preset standard value, use the initial mineral classification model as the mineral classification model; otherwise, continue to perform model training using the mineral classification training set.

[0010] A magma rock mineral analysis system based on geological images, comprising: A magma rock feature analysis module, including a block feature analysis unit and a spatio-temporal feature analysis unit; the block feature analysis unit is used to obtain magma rock mineral images based on the magma rock to be analyzed; extract the surrounding geological components of the magma rock to be analyzed to obtain magma rock auxiliary geological data; extract mineral features based on the magma rock mineral images and magma rock geological data to obtain magma rock mineral features; the magma rock mineral features include magma rock mineral crystallization features and magma rock mineral coexisting oxygen fugacity features; the spatio-temporal feature analysis unit is used to perform retrospective mineral analysis on the magma rock to be analyzed based on the magma rock mineral features to obtain magma rock retrospective features; perform spatio-temporal evolution analysis on the magma rock to be analyzed based on the magma rock mineral features to obtain magma rock evolution features; A magma rock comprehensive analysis module, including a comprehensive feature analysis unit, which is used to perform comprehensive magma rock mineral analysis based on the magma rock mineral features, magma rock retrospective features and magma rock evolution features to obtain the full-process magma rock mineral analysis results.

[0011] The present invention has the following advantages: 1. By integrating magma rock mineral images and magma rock geological auxiliary data, the present invention constructs a full-process magma rock mineral analysis method integrating mineral identification, feature extraction, spatio-temporal evolution analysis and digital twin modeling, which helps to reveal the metallogenic evolution history and future metallogenic potential of magma rocks; by using a mineral classification model combined with EPMA and BSE image features, high-precision automatic identification and classification of minerals in magma rocks can be realized, reducing manual intervention and improving the identification efficiency and accuracy; the extracted mineral crystallization features and the integrated oxygen fugacity features are used to realize the joint analysis of the physical structure and chemical properties of minerals, providing more comprehensive data support for subsequent geological evolution research and mineral prediction.

[0012] 2. By constructing a dual-channel and bidirectional mineral analysis network model and introducing an adversarial generation network and a spatio-temporal Transformer mechanism, the present invention helps to realize the bidirectional dynamic modeling of the geological process of magma rocks, which can not only trace back the early genetic conditions of magma rocks, but also predict the future evolution trend of magma rocks, breaking through the limitations of traditional methods that can only perform static analysis or one-way reasoning, and providing an innovative technical path for the analysis of rock genetic mechanisms and the restoration of deep geological processes. Description of the Drawings

[0013] Figure 1 It is a schematic structural diagram of a magma rock mineral analysis system based on geological images adopted in an embodiment of the present invention. Detailed Embodiment

[0014] To enable those skilled in the art to better understand the technical solutions in the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention.

[0015] Embodiment 1, a method for analyzing magmatic rock minerals based on geological images, comprising the following steps: Based on the magmatic rock to be analyzed, obtain magmatic rock mineral images; extract the surrounding geological components of the magmatic rock to be analyzed to obtain magmatic rock auxiliary geological data; The method for obtaining magmatic rock mineral images is to select representative samples from the magmatic rock to be analyzed, make standard magmatic rock thin sections, and use a scanning electron microscope (SEM) or a backscattered electron imaging (BSE) system to scan the thin sections to obtain high-resolution mineral images as magmatic rock mineral images; For the existing geological data around the magmatic rock to be analyzed, such as basic information such as strata, structures, lithology, age, etc., and use an electron probe microanalyzer (EPMA) to test the major, trace and rare earth elements of the magmatic rock sample to be analyzed to obtain chemical composition data as magmatic rock auxiliary geological data; Extract mineral characteristics based on the magmatic rock mineral images and magmatic rock geological data to obtain magmatic rock mineral characteristics; the magmatic rock mineral characteristics include magmatic rock mineral crystallization characteristics and magmatic rock mineral coexisting oxygen fugacity characteristics; The specific steps for extracting mineral characteristics based on the magmatic rock mineral images and magmatic rock geological data include: Use a pre-trained mineral classification model to classify the magmatic rock mineral images to obtain a mineral classification image G n , n = 1, 2,..., N; N is the total number of mineral categories after mineral recognition of the magmatic rock mineral images, and each mineral classification image G n corresponds to a labeled or unlabeled mineral classification; denote the corresponding part of the mineral classification image G n in the magmatic rock to be analyzed as the magmatic rock block F n ; When using the pre-trained mineral classification model to classify the magmatic rock mineral images, when the predicted result of the mineral category output by the mineral classification model for a certain image area has a high confidence level, for example, the prediction probability of the image reaches the set threshold, the set threshold is set manually and can be modified based on the actual situation, generally can be set to 0.9, and this category already exists in the mineral knowledge base, then it is determined as the labeled mineral classification image G n ; otherwise, when the classification confidence level output by the model is lower than the set threshold, or the recognition result fails to effectively match the existing mineral categories, it is regarded as an unlabeled mineral classification image G n , and it is necessary to further combine special BSE images and EPMA mineral characteristics for auxiliary classification and manual verification; The specific steps for training a mineral classification model include: Collect several groups of mineral classification training samples; each group of mineral classification training samples contains mineral type images and corresponding annotation labels; combine several groups of mineral classification training samples to obtain a mineral classification training set; Input the mineral classification training set into the CNN model for model training to obtain an initial mineral classification model; perform model evaluation on the initial mineral classification model. If the recognition accuracy of the initial mineral classification model reaches the preset standard value, then use the initial mineral classification model as the mineral classification model; otherwise, continue model training using the mineral classification training set; The collection method of the mineral classification training set includes: collecting existing rock slice databases, including optical images, BSE images, EPMA images, etc. under the microscope. After the field samples are prepared, unified image acquisition is carried out through a standardized scanning device to ensure consistent image quality and resolution; geological experts manually segment the mineral areas in the images and label the mineral types; use the initial classification model or image segmentation algorithm to assist in dividing the mineral areas, and manually confirm and correct them; combine the BSE image and EPMA test results to determine the mineral composition, and then map it back to the corresponding area of the image to improve the annotation accuracy; The collected mineral classification training set covers various types of magmatic rocks, such as granite, gabbro, diorite, etc., and at the same time includes common mineral species, such as quartz, plagioclase, biotite, olivine, pyroxene, etc.; the acquisition of mineral images covers different scales, different degrees of crystallization and variable tectonic areas to enhance the generalization ability of the model; When the average recognition accuracy of the mineral classification training set reaches the preset standard value, terminate the training. The preset standard value is determined by humans. For example, the preset standard value is 90%; if the recognition accuracy of the initial mineral classification model does not improve within the set number of rounds, automatically stop the training to prevent overfitting; When using the pre-trained mineral classification model to classify magmatic rock mineral images, the magmatic rock mineral images are slid and cut into several image blocks of a fixed size based on the CNN-based image classification and localization structure. Each block retains its position information in the original image and is input into the pre-trained CNN model for classification and recognition; the CNN end outputs the mineral category labels of each block through the softmax layer, aggregates the image blocks by category to reconstruct the binary mask map of each type of mineral, and obtains the corresponding mineral classification image G n ; each mineral classification image G n The corresponding original image area is the magmatic rock block F n , indicating the specific distribution position of this type of mineral in the image.

[0016] For the unannotated mineral classification image G n Perform mineral classification. The specific steps are as follows: Based on the unlabeled mineral classification image G n , obtain the magmatic rock block F n 's special BSE image B n and EPMA mineral characteristics E n ; Based on the special BSE image B n and EPMA mineral characteristics E n perform mineral classification annotation on the unlabeled mineral classification image G n , and convert the unlabeled mineral classification image G n into a mineral classification image G with labeled mineral classification n ; Locate the corresponding area of the unlabeled mineral classification image G n in the original magmatic rock mineral image as the magmatic rock block F n ; Use a scanning electron microscope (SEM) to perform backscattered electron imaging on the magmatic rock block F n to obtain a special BSE image B with high resolution and reflecting mineral density differences n ; At the same time, perform image enhancement (such as histogram equalization, denoising filtering) processing to improve the clarity of mineral particle boundaries; Select several representative mineral areas in the magmatic rock block F n and use electron probe microanalysis (EPMA) technology for point analysis or area scanning to extract major elements (such as Si, Al, Fe, Mg, etc.), trace elements and their distribution characteristics to form EPMA mineral characteristics E n ; Based on the particle morphology characteristics and gray level distribution provided by the special BSE image B n , combined with the chemical composition information revealed by the EPMA mineral characteristics E n , use these two types of characteristics as input feature vectors, and use a pre-trained special mineral recognition model to determine the mineral type of the magmatic rock block F n . On the basis of a clear determination result, perform mineral name annotation on the unlabeled mineral classification image G n , and convert it into a mineral classification image G with labeled mineral classification n ; The pre-trained special mineral recognition model is trained by fusing image features and compositional features. Its sample sources include BSE images of standard rock thin sections and EPMA test data of corresponding regions. Each group of sample content includes mineral image patches, corresponding chemical element compositions, and mineral category labels confirmed by experts. The training objective is to enable the model to accurately identify different mineral types and output classification results with high confidence. For training the basic model, support vector machine (SVM), random forest, decision tree, etc. can be selected for training, and the input is the combination of image features and EPMA features. During the training process, cross-validation is used to evaluate the model performance. When the recognition accuracy of the validation set reaches the preset training threshold (such as 90%) and there is no significant improvement in accuracy for several consecutive rounds, the training is terminated and the optimal model is saved, which is the special mineral recognition model. For the mineral classification image G with labeled mineral classification n The specific steps for further analysis include: For the mineral classification image G n Perform grayscale processing to obtain the mineral classification grayscale image H n ; Convert the mineral classification image G n from a color or multi-channel form to a single-channel grayscale image, which can be achieved through weighted average or color channel conversion, and can provide a unified grayscale basis for subsequent particle recognition and texture analysis. The input of this step is the mineral classification image G n , and the output is the mineral classification grayscale image H n , which can simplify the image structure, enhance the contrast of mineral particle boundaries, and contribute to image segmentation processing; For the mineral classification grayscale image H n Use guided filtering to segment mineral particles and adjust the CLAHE window output to obtain the enhanced mineral classification grayscale image H n '; First, use the guided filter to perform edge-preserving filtering on the mineral classification grayscale image H n with the mineral classification image G n as a reference; Then apply contrast-limited adaptive histogram equalization (CLAHE) to enhance the local contrast, which helps to strengthen the particle contour, improve the local detail information of the image, improve the problems of particle adhesion and blurring, and enhance the accuracy of subsequent particle boundary detection and grain recognition; Based on the enhanced mineral classification grayscale image H n ', calculate the gray-level co-occurrence matrix to obtain the mineral crystallization matrix Z n ; Calculate the gray-level co-occurrence matrix (GLCM) of the enhanced mineral classification grayscale image H n ' at different directions and distances, and extract texture features (such as contrast, entropy, energy, etc.), which can reflect the crystal arrangement characteristics and surface texture rules of mineral particles, reveal the internal crystal microstructure and crystallization degree of minerals, and provide a basis for quantitative analysis; Based on the enhanced mineral classification grayscale image H n ', take the centers of adjacent grains as nodes and the grain boundary contact length as the edge weight, calculate the network topology parameters, and obtain the mineral crystallization topology parameter C n ; take the grain centroids identified in the enhanced mineral classification grayscale image H n ' as the graph nodes and the grain boundary contact length between adjacent grains as the edge weight to construct a grain spatial topology graph; calculate topology parameters such as the average degree, clustering coefficient, and shortest path, etc.; it can quantify the spatial connection pattern and crystallization network structure of mineral grains, characterize the organization and complexity of the mineral crystallization process, and assist in identifying genetic characteristics such as cooling rate or metamorphism; Combine the mineral crystallization matrix Z n , the mineral crystallization topology parameter C n to output the mineral crystallization characteristics J of magmatic rocks n ; combine the various eigenvectors extracted from the mineral crystallization matrix Z n and the mineral crystallization topology parameter C n to form a unified mineral crystallization characteristic J of magmatic rocks n , which can comprehensively reflect the geometric morphology, arrangement pattern, and texture structure of mineral crystallization in magmatic rocks, provide a quantitative crystallization characteristic index, and provide support for subsequent spatio-temporal evolution and genetic analysis; The specific steps for calculating the oxygen fugacity based on the labeled mineral classification labels of the mineral classification image G n include: Obtain the EPMA mineral characteristics E n of the mineral classification image G n , calculate the oxygen fugacity based on the EPMA mineral characteristics E n to obtain the mineral classification oxygen fugacity O n ; in the corresponding area of the mineral classification image G n , extract the EPMA mineral characteristics E n (including element concentration, valence state, etc.), and calculate the mineral classification oxygen fugacity O 3+ / Fe 2+ ratio) based on the mineral chemical model (such as the ratio of Fe n ; it helps to reflect the redox conditions of the magmatic crystallization environment and reveal deep geological processes such as the ore-forming environment and electrochemical state; Based on the mineral classification oxygen fugacity O n to output the mineral cooperative oxygen fugacity characteristics D of magmatic rocks n ; extract the characteristics of the mineral classification oxygen fugacity On and output the mineral cooperative oxygen fugacity characteristics D of magmatic rocks n ; Combine all the mineral crystallization characteristics J of magmatic rocks n and the mineral cooperative oxygen fugacity characteristics D of magmatic rocks nCombined to obtain the mineral characteristics of magmatic rocks; divide the same magmatic rock into blocks F n Correspondingly, the crystal characteristics J of magmatic rock minerals of two eigenvectors n And the coexisting oxygen fugacity characteristics D of magmatic rock minerals n Merged in a unified format to form a set of magmatic rock mineral characteristics, which serves as the input for the subsequent backtracking and evolution analysis model; it can completely describe the three-dimensional characteristics of the structure-composition-environment of minerals in magmatic rocks, enhance the adaptability and discriminability of the subsequent analysis model to the rock formation process, and realize the in-depth association from images to geological processes; Based on the magmatic rock mineral characteristics, perform backtracking mineral analysis on the magmatic rock to be analyzed to obtain the backtracking characteristics of the magmatic rock; based on the magmatic rock mineral characteristics, perform spatio-temporal evolution analysis on the magmatic rock to be analyzed to obtain the evolution characteristics of the magmatic rock; among them, the specific steps include: Construct a dual-channel bidirectional mineral analysis network model, which includes a backtracking mineral analysis layer and a spatio-temporal evolution analysis layer; Construct a deep learning architecture composed of two functional channels, including a backtracking mineral analysis layer and a spatio-temporal evolution analysis layer; the input is the magmatic rock mineral characteristics and the magmatic rock auxiliary geological data, which respectively guide the time analysis tasks in two directions; used to realize the bidirectional simulation analysis of the magmatic rock geological process, support the genetic restoration and evolution trend judgment; different from the traditional unidirectional geological analysis model, this structure realizes the bidirectional time series modeling of geological information and improves the integrity and reliability of the magmatic rock evolution modeling; In the backtracking mineral analysis layer, based on the adversarial generative network, extract the magmatic rock auxiliary geological data for reverse mapping to obtain the backtracking magmatic rock geological conditions; based on the backtracking magmatic rock geological conditions and the magmatic rock mineral characteristics, perform reverse time step evolution to obtain the backtracking characteristics of the magmatic rock; Adopt the adversarial generative network structure, with the magmatic rock mineral characteristics as the conditional input, the discriminator compares the actual and generated auxiliary geological data, and trains the generator to reverse output the geological conditions in the "historical state", simulating the "reverse" evolution process of the magmatic rock from the current state to the geological environment in the early crystallization or formation period; through the reverse mapping mechanism of GAN, break through the limitation of traditional geological evolution that "can only predict the future", realize the reconstruction and feature reproduction of the magmatic rock formation initial environment, and provide an intelligent tool for paleogeological reconstruction; In the training of the retrospective mineral analysis layer, the sample sources include the auxiliary geological data of magmatic rocks from the typical sample database of magmatic rocks at different known times and the corresponding magmatic rock mineral characteristics. The sample content is composed of the pairing of the current magmatic rock mineral characteristics and historical geological conditions. The training objective is to enable the generator to accurately generate the corresponding retrospective magmatic rock geological conditions under the condition of given magmatic rock mineral characteristics. The discriminator differentiates between real geological data and generated data to improve the authenticity and geological consistency of the generated results. The conditional adversarial network (Conditional GAN) is used as the training basic model, and the magmatic rock mineral characteristics are introduced as the conditional vector. The training termination condition is set as when the geological conditions output by the generator reach the preset similarity index (such as the structural similarity SSIM≥0.9) in the validation set, and the discriminant accuracy fluctuation of the discriminator tends to be stable (the change rate is lower than 1%), it is considered that the model converges, the training is terminated, and the final model is output. The values 0.9 and 1% here are set manually. In the spatio-temporal evolution analysis layer, the spatio-temporal Transformer mechanism is used to predict and analyze the auxiliary geological data of magmatic rocks to obtain the predicted magmatic rock geological conditions. Based on the predicted magmatic rock geological conditions and magmatic rock mineral characteristics, the evolution of the prediction time step is carried out to obtain the magmatic rock evolution characteristics. The spatio-temporal Transformer mechanism is used to process the time-series auxiliary geological data of magmatic rocks, which is used as the time-space joint input to extract the dynamic evolution pattern therein and output the possible future geological states. The spatio-temporal Transformer introduces the attention mechanism to process the long-distance dependence relationship, making the model more accurate and adaptable in capturing the regional geological change trend and tectonic evolution rhythm, and breaking through the performance bottleneck of traditional time-series models in geological applications. In the spatio-temporal evolution analysis layer, for the model training based on the spatio-temporal Transformer mechanism, the sample sources are multi-temporal magmatic rock profile data and regional geological evolution records. The sample content includes the sequence of auxiliary geological data of magmatic rocks at consecutive time steps (such as element content, tectonic parameters, temperature and pressure conditions, etc.) and the magmatic rock mineral characteristics at the corresponding time points. The training objective is to accurately predict the future magmatic rock geological conditions by learning the joint dependence pattern of time and space. The improved spatio-temporal Transformer structure is used as the training basic model, and the multi-head attention mechanism is combined to process the dynamic correlation of long time series and multi-variable geological characteristics. The training termination condition is set as when the mean square error (MSE) of the geological condition prediction of the model in the validation set is lower than the threshold (such as 0.01), and there is no significant improvement in accuracy (the improvement amplitude is lower than 0.5%) in multiple consecutive epochs, the training is terminated to ensure that the model converges and has good generalization ability. The values 0.01 and 0.5% here are set manually.

[0017] Based on the mineral characteristics of magmatic rocks, the traceability characteristics of magmatic rocks and the evolution characteristics of magmatic rocks, a comprehensive magmatic rock mineral analysis is carried out to obtain the full-process magmatic rock mineral analysis results; the specific steps include: Construct a digital twin model of magmatic rocks based on the magmatic rocks to be analyzed; use 3D modeling technology and deep learning framework, combined with the actual spatial distribution, structural structure, geological properties and mineral image information of the magmatic rocks to be analyzed, to construct a digital rock model with time-series evolution capability as a digital twin model of magmatic rocks; the digital twin model of magmatic rocks is implemented in the form of graph structure or 3D grid, supporting dynamic feature update and prediction and deduction; it helps to reproduce real samples through virtual environment, open up the channel between geological samples and data-driven modeling, support intelligent simulation and multi-scenario evolution experiments, and improve model visualization and interpretability; For the magmatic rock digital twin model, the mineral characteristics, magmatic rock traceability characteristics and magmatic rock evolution characteristics of the magmatic rock are introduced for feature embedding to obtain an updated magmatic rock digital twin model; The mineral characteristics, traceability characteristics and evolution characteristics of igneous rocks are taken as high-dimensional feature vectors, and dynamically integrated with the igneous rock digital twin model through graph neural networks, embedded mapping networks or Transformer encoder modules to update the attributes of each spatial unit or node; the igneous rock digital twin model can learn the nonlinear relationship between features and optimize the simulation accuracy; this step maps multidimensional mineral characteristic information such as time, structure, and chemistry into the igneous rock digital twin model to achieve deep coupling of structure-process-genesis, innovatively integrate static mineral images with dynamic evolution processes, and provide a unified modeling platform for the genetic mechanism and evolution path of the entire geological behavior from formation to the current state; Generate the full-process magmatic rock mineral analysis results based on the updated magmatic rock digital twin model; use the updated magmatic rock digital twin model to perform simulation analysis, including: mineral spatial distribution prediction, genetic process deduction and mineralization potential or tectonic impact assessment, and finally output comprehensive full-process magmatic rock mineral analysis results; The prediction of mineral spatial distribution uses the mineral features of magmatic rocks embedded in the updated magmatic rock digital twin model to analyze the three-dimensional spatial units or grid nodes of the model point by point, and construct a spatial prediction module, such as a neural network or a three-dimensional convolutional neural network, to learn the distribution law between mineral types and spatial positions. At the same time, considering the spatial synergy between minerals, an adjacency weight graph based on node relationships is established to improve the predicted structural continuity and boundary clarity. Based on the updated magmatic rock digital twin model and mineral feature mapping, a three-dimensional mineral distribution map is output; The genetic process deduction constructs a time series simulation model based on the backtracking characteristics and evolution characteristics of magmatic rocks, and performs simulation calculations on the geological states at each time step in the model, including the magma cooling rate, the mineral evolution path, and the change in oxygen fugacity; the simulation path forms a mineral evolution trajectory through the superposition of time steps, which can be regarded as the geological life cycle chain of magmatic rocks, and outputs a genetic evolution sequence diagram based on the backtracking characteristics and evolution characteristics of magmatic rocks; The assessment of ore-forming potential or tectonic influence uses mineral types, spatial distribution, oxygen fugacity characteristics, etc. as input features, and combines regional geological structure parameters to construct a classification regression network to predict the ore-forming potential level of the area where the magmatic rock to be analyzed is located, and generate an ore-forming potential heat map; the above three types of analysis results are fused to output a structured and visualized full-process magmatic rock mineral analysis result; By introducing the digital twin model and deep learning technology, the magmatic rock mineral images, historical formation conditions and future evolution trends are organically integrated, realizing a closed-loop analysis path of magmatic rocks from "image recognition - feature extraction - process modeling - global analysis". This method not only has strong visualization and high model interpretability, but also has high scalability, and is particularly suitable for deployment in geological big data platforms and intelligent mineral analysis systems, which is a key technological breakthrough in realizing intelligent geological analysis.

[0018] Example 2, a magmatic rock mineral analysis system based on geological images, see Figure 1 as shown, including: The magmatic rock feature analysis module includes a block feature analysis unit and a spatio-temporal feature analysis unit; the block feature analysis unit is used to obtain magmatic rock mineral images based on the magmatic rock to be analyzed; extract the surrounding geological components of the magmatic rock to be analyzed to obtain magmatic rock auxiliary geological data; perform mineral feature extraction based on the magmatic rock mineral images and magmatic rock geological data to obtain magmatic rock mineral features; the magmatic rock mineral features include magmatic rock mineral crystallization features and magmatic rock mineral cooperative oxygen fugacity features; the spatio-temporal feature analysis unit is used to perform backtracking mineral analysis on the magmatic rock to be analyzed based on the magmatic rock mineral features to obtain magmatic rock backtracking features; perform spatio-temporal evolution analysis on the magmatic rock to be analyzed based on the magmatic rock mineral features to obtain magmatic rock evolution features; The magmatic rock comprehensive analysis module includes a comprehensive feature analysis unit, which is used to perform comprehensive magmatic rock mineral analysis based on the magmatic rock mineral features, magmatic rock backtracking features and magmatic rock evolution features to obtain a full-process magmatic rock mineral analysis result.

[0019] It should be understood that for those of ordinary skill in the art, improvements or transformations can be made according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention. The parts not described in detail in this specification belong to the prior art well-known to those of ordinary skill in the art.

Claims

1. A method for analyzing magmatic rock minerals based on geological images, characterized in that, Including the following steps: Based on the magmatic rock to be analyzed, obtain the magmatic rock mineral image; extract the surrounding geological components of the magmatic rock to be analyzed to obtain the magmatic rock auxiliary geological data; Extract mineral characteristics based on the magmatic rock mineral image and the magmatic rock geological data to obtain the magmatic rock mineral characteristics; the magmatic rock mineral characteristics include the magmatic rock mineral crystallization characteristics and the magmatic rock mineral coexisting oxygen fugacity characteristics; Conduct retrospective mineral analysis on the magmatic rock to be analyzed based on the magmatic rock mineral characteristics to obtain the magmatic rock retrospective characteristics; Conduct spatio-temporal evolution analysis on the magmatic rock to be analyzed based on the magmatic rock mineral characteristics to obtain the magmatic rock evolution characteristics; Conduct comprehensive magmatic rock mineral analysis based on the magmatic rock mineral characteristics, the magmatic rock retrospective characteristics and the magmatic rock evolution characteristics to obtain the full-process magmatic rock mineral analysis result.

2. The method for analyzing magmatic rock minerals based on geological images according to claim 1, wherein The specific steps of extracting mineral characteristics based on the magmatic rock mineral image and the magmatic rock geological data include: Use a pre-trained mineral classification model to classify the minerals in the magmatic rock mineral images, and obtain the mineral classification image G n , n = 1, 2,..., N; N is the total number of mineral categories after mineral identification of the magmatic rock mineral images. Each mineral classification image G n corresponds to a labeled or unlabeled mineral classification; Denote the corresponding part of the mineral classification image G n in the magmatic rock to be analyzed as the magmatic rock block F n ; For the unlabeled mineral classification image G n Perform mineral classification. The specific steps are as follows: Based on the unlabeled mineral classification image G n , obtain the segmented igneous rock F n and its special BSE image B n and EPMA mineral characteristics E n ; Based on the special BSE image B n and EPMA mineral characteristics E n , perform mineral classification annotation on the unlabeled mineral classification image G n , and convert the unlabeled mineral classification image G n into a mineral classification image G with labeled minerals n .

3. The method for analyzing magmatic rock minerals based on geological images according to claim 2, wherein, For the mineral classification image G with labeled mineral classification n Specific steps for further analysis include: Perform grayscale processing on the mineral classification image G n to obtain the grayscale mineral classification image H n ; Use guided filtering to segment mineral particles and adjust the CLAHE window output for the grayscale mineral classification image H n to obtain the enhanced grayscale mineral classification image H n ’; Calculate the gray-level co-occurrence matrix based on the enhanced grayscale mineral classification image H n ’ to obtain the mineral crystallization matrix Z n ; Take the centers of adjacent grains in the enhanced grayscale mineral classification image H n ’ as nodes and the grain boundary contact length as the edge weight, calculate the network topology parameters to obtain the mineral crystallization topology parameter C n ; Combine the mineral crystallization matrix Z n and the mineral crystallization topology parameter C n to output the mineral crystallization characteristics J of igneous rocks n ; Based on the labeled mineral classification labels of mineral classification image G n The specific steps for calculating the oxygen fugacity based on the labeled mineral classification labels include: Obtain the mineral classification image G n EPMA mineral characteristics E of n , based on the EPMA mineral characteristics E n Perform oxygen fugacity calculation to obtain the oxygen fugacity O of mineral classification n ; Based on the oxygen fugacity O of mineral classification n Output the co - oxygen fugacity characteristics D of magmatic rock minerals n ; Combine all the crystallization characteristics J of magmatic rock minerals n and the combined oxygen fugacity characteristics D of magmatic rock minerals n to obtain the characteristics of magmatic rock minerals.

4. The method for analyzing magmatic rock minerals based on geological images according to claim 3, characterized in that, The specific steps of conducting retrospective mineral analysis and spatio-temporal evolution analysis on the magmatic rock to be analyzed based on the magmatic rock mineral characteristics include: Construct a dual-channel and bidirectional mineral analysis network model, and the dual-channel and bidirectional mineral analysis network model includes a retrospective mineral analysis layer and a spatio-temporal evolution analysis layer; In the retrospective mineral analysis layer, based on the adversarial generative network, extract the magmatic rock auxiliary geological data for reverse mapping to obtain the retrospective magmatic rock geological conditions; based on the retrospective magmatic rock geological conditions and the magmatic rock mineral characteristics, conduct reverse time-step evolution to obtain the magmatic rock retrospective characteristics; In the spatio-temporal evolution analysis layer, based on the spatio-temporal Transformer mechanism, conduct predictive analysis on the magmatic rock auxiliary geological data to obtain the predicted magmatic rock geological conditions; based on the predicted magmatic rock geological conditions and the magmatic rock mineral characteristics, conduct predictive time-step evolution to obtain the magmatic rock evolution characteristics.

5. A method for analyzing magmatic rock minerals based on geological images according to claim 4, characterized in that, The specific steps of conducting comprehensive magmatic rock mineral analysis based on the magmatic rock mineral characteristics, the magmatic rock retrospective characteristics and the magmatic rock evolution characteristics include: Construct a digital twin model of the magmatic rock based on the magmatic rock to be analyzed; For the digital twin model of the magmatic rock, introduce the magmatic rock mineral characteristics, the magmatic rock retrospective characteristics and the magmatic rock evolution characteristics for feature embedding to obtain an updated digital twin model of the magmatic rock; Generate the full-process magmatic rock mineral analysis result based on the updated digital twin model of the magmatic rock.

6. The method for analyzing magmatic rock minerals based on geological images according to claim 5, characterized in that, The specific steps of training the mineral classification model include: Collect several groups of mineral classification training samples; each group of mineral classification training samples includes a mineral type image and a corresponding annotation label; combine several groups of mineral classification training samples to obtain a mineral classification training set; Input the mineral classification training set into the CNN model for model training to obtain an initial mineral classification model; conduct model evaluation on the initial mineral classification model. If the recognition accuracy of the initial mineral classification model reaches the preset standard value, use the initial mineral classification model as the mineral classification model; otherwise, continue to conduct model training using the mineral classification training set.

7. A magma rock mineral analysis system based on geological images, characterized in that, The system, a method for analyzing magmatic rock minerals based on geological images as described in any one of the above claims 1-6, includes: The magmatic rock feature analysis module includes a block feature analysis unit and a spatio-temporal feature analysis unit; the block feature analysis unit is used to obtain magmatic rock mineral images based on the magmatic rock to be analyzed; extract the surrounding geological components of the magmatic rock to be analyzed to obtain magmatic rock auxiliary geological data; extract mineral features based on the magmatic rock mineral images and magmatic rock geological data to obtain magmatic rock mineral features; the magmatic rock mineral features include magmatic rock mineral crystallization features and magmatic rock mineral coexisting oxygen fugacity features; the spatio-temporal feature analysis unit is used to perform retrospective mineral analysis on the magmatic rock to be analyzed based on the magmatic rock mineral features to obtain magmatic rock retrospective features; perform spatio-temporal evolution analysis on the magmatic rock to be analyzed based on the magmatic rock mineral features to obtain magmatic rock evolution features. The magmatic rock comprehensive analysis module includes a comprehensive feature analysis unit, which is used to perform comprehensive magmatic rock mineral analysis based on the magmatic rock mineral features, magmatic rock retrospective features and magmatic rock evolution features to obtain the full-process magmatic rock mineral analysis results.

Images