Carbonate Rock Image Tagging System Based on Deep Learning Technology

Abstract

The present invention relates to the technical field of data processing, and specifically to a carbonate rock image tagging system based on deep learning technology. It includes a multi-source feature extraction unit, a spectrum classification unit, and a type recognition unit. The present invention introduces technologies such as deep learning, multi-source data fusion, and spectrum intelligent recognition into the identification of carbonate rock thin sections. By constructing spectrum tags, establishing an identification method system, summarizing the correlations between typical carbonate rock thin section types and rock material compositions, textures, and physical property parameters, establishing spectrum recognition marks for typical carbonate rock thin section types, and conducting comprehensive analysis from the perspectives of diagrams, spectra, and multi-source physical property parameter information, the technical route is feasible, which is conducive to improving the accuracy of type recognition, optimizing and enhancing the work of traditional visual identification of thin sections, and can provide technical support for the intelligent identification, non-destructive testing, and in-situ parameter analysis of carbonate rock thin sections.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and more specifically, to a carbonate rock image tagging system based on deep learning technology. Background Art

[0002] Geology is a course that involves the composition, structure, and tectonics of the earth's materials, the principles of geological processes, and the earth's evolutionary history. Mineral rocks (ores) are important record carriers of the geological evolutionary history and are the "natural books" of geology. Observing rocks and conducting thin section identification of rocks are important contents and important skills in geological work, and are also important bases for further analyzing geological processes and geological processes, and for applying special topics such as resource environment.

[0003] Currently, most thin section identifications mainly rely on a large number of geologists using microscopes for visual identification. Rock and mineral identification personnel need to undergo a large amount of training and experience accumulation to be competent for the work, and face a large amount of heavy repetitive labor in the work. There are problems such as strong subjectivity of the identification results, long judgment cycles, high judgment costs, and inability to balance the energy and efficiency of the identification personnel. At the same time, due to differences in the geological characteristics and development mechanisms of rocks in different regions, the identification levels of participating petrologists are often uneven, which can lead to inconsistent analysis, thus affecting subsequent work such as cross-source identification and comparative analysis.

[0004] With the development of graphics and image processing technologies, information extraction using thin section images has become an important research direction. Early thin section image information extraction mainly relied on information such as image texture, color, and grayscale, and used methods such as image enhancement processing, unsupervised classification, and image density segmentation to combine color information with texture descriptions, assist expert knowledge decision-making, and extract obvious trace feature information in thin sections of rocks, and proposed a man-machine combined system analysis method for rock microscopic structure images.

[0005] However, due to certain differences in the material composition and particle size in the thin section, different types and scales of thin section graphics require different image processing methods for information extraction. The image processing is computationally complex, and the accuracy and efficiency of information extraction are not high, and stable calculation results cannot be obtained. In recent years, the application of image intelligent processing technology has been continuously deepened in various fields, and the classification and information extraction of rock images have developed greatly. Neural network models and various numerical analysis methods have been gradually introduced into the research work of image processing and intelligent recognition of thin sections of rocks, and the automated output of analysis results has been realized to a certain extent. However, the accuracy of the analysis results is greatly affected by sample screening and the extraction effect of image parameters. Especially in the identification of carbonate rock thin sections, single graphic parameters limit the identification accuracy. In view of this, we propose a carbonate rock image tagging system based on deep learning technology. Summary of the Invention

[0006] The present invention aims to solve the problems that the image processing in the prior art is computationally complex and the recognition accuracy is limited by a single graphic parameter. While extracting information from microscopic images of rock thin sections at different spatial scales, the present invention comprehensively utilizes cathodoluminescence images and elemental spectral plane scan images to extract trace fabric and material composition information, constructs a three-dimensional spatial framework for various types of rock thin sections, realizes the analysis of multi-source and multi-level data under the background of the three-dimensional spatial framework, and conducts correlation analysis and type classification to achieve the identification of rock thin section types, avoiding the accuracy and migration problems faced by a single data source.

[0007] To achieve the above object, the present invention provides a carbonate rock image tagging system based on deep learning technology, including a multi-source feature extraction unit, a spectrum classification unit, and a type recognition unit;

[0008] The multi-source feature extraction unit is used to establish a carbonate rock particle image data set, extract spatial trace information from microscopic images of rock thin sections at different spatial scales by using remote sensing image processing algorithms, and comprehensively utilize cathodoluminescence images and elemental spectral plane scan images to extract trace fabric and material composition information, model the structural characteristics of carbonate rocks in a three-dimensional space model, and construct a three-dimensional spatial framework for various types of rock thin sections through transfer learning;

[0009] The spectrum classification unit is used to screen the fabric trace and material composition feature spectrum information of carbonate rocks, initially construct a spectrum classifier, reflect the correlation between the material composition and the fabric trace through the Pearson correlation coefficient, summarize the content and spatial distribution law of its material composition, and explore its influencing factors;

[0010] The type recognition unit is used to fuse the trace fabric and material composition feature spectrum information in a unified three-dimensional spatial framework, establish a spectrum fingerprint of the carbonate rock thin section, use the BP neural network model with the spectrum fingerprint as the input, and conduct the identification of the rock thin section type by comprehensively considering the coupling law of the carbonate rock thin section spectrum information and the results of transfer learning, and the output result is the predicted type of the carbonate rock thin section.

[0011] As a further improvement of this technical solution, when the multi-source feature extraction unit extracts spatial trace information, it includes the following steps:

[0012] Scan the two-dimensional digital images of various types of rock thin sections, classify and establish a graphic library to form a carbonate rock particle image data set;

[0013] Using the morphological, textural, color, brightness, and roughness characteristic parameters in a two-dimensional digital image, the structural framework is described graphically within a two-dimensional grid plane, including the extraction of spatial trace information by overlapping the position and orientation changes to obtain the graphical average morphology.

[0014] As a further improvement of this technical solution, the multi-source feature extraction unit comprehensively utilizes the cathodoluminescence image and the elemental spectral plane scan image to extract the trace fabric and material composition information, including the following steps:

[0015] Adjust the electron beam parameters of the electron microscope to focus the electron beam on the surface of the rock thin section to excite the luminescence centers in the sample. The energy spectrometer is installed on the electron microscope to record the X-ray signals emitted by the sample under the excitation of the electron beam, and convert the recorded X-ray signals into digital images, using gray values to represent the luminescence intensity. First, perform image denoising and image enhancement preprocessing on the thin section microscopic image to keep the area where carbonate rock particles are located with good edges and textures, and enhance the gray difference between the particle area and the background area in the image to form a cathodoluminescence image and an elemental spectral plane scan image;

[0016] Align the cathodoluminescence image and the elemental spectral plane scan image to fuse the information of the two images to generate a composite image;

[0017] Use a clustering algorithm to classify the features, identify different mineral phases, and visually generate a trace fabric map and a material composition map.

[0018] As a further improvement of this technical solution, the information fusion of the two images to generate a composite image adopts a multi-scale fusion algorithm, including the following steps:

[0019] Read the cathodoluminescence image and the elemental spectral plane scan image, and normalize the image data to the same dynamic range;

[0020] Use a feature point detection method to extract the feature points of the images, match the feature points of the two images, calculate the transformation matrix, align one image to the other image, use wavelet transform to decompose the aligned image into sub-bands of different scales, and fuse the sub-bands of different scales to form a composite image.

[0021] As a further improvement of this technical solution, the structural features include spatial trace information, trace fabric, and material composition information.

[0022] As a further improvement of this technical solution, through transfer learning, a three-dimensional spatial framework of each type of rock thin section is constructed, including the following steps:

[0023] Read the composite image, convert the image data into a format suitable for three-dimensional modeling, create a three-dimensional grid, and interpolate the composite image data into the three-dimensional grid;

[0024] Trace-based transfer learning: Based on trace features, construct a trace model for transfer learning to adapt to the classification task of trace features;

[0025] Substance-based transfer learning: Based on substance composition information, construct a transfer model for transfer learning to adapt to the classification task of substance composition;

[0026] Coupling-based transfer learning: Combine trace features and substance composition information, extract coupling features, and construct a coupling model for transfer learning to adapt to the classification task of coupling features;

[0027] Model-based transfer learning: Combine trace features, substance composition information, and coupling features, construct a multi-modal model for transfer learning to adapt to the classification task of multi-modal features;

[0028] Apply the classification results to a three-dimensional spatial lattice to generate a three-dimensional model of carbonate rock thin sections.

[0029] As a further improvement of this technical solution, in the atlas classification unit, initially construct an atlas classifier, including the following steps:

[0030] In a unified spatial scale, from the three-dimensional model, mark according to the different optical characteristics of internal biological frameworks, grains, and cement substances, screen the fabric trace and substance composition feature map information of the main rock thin sections to form multiple single vector layers;

[0031] Analyze the content and spatial distribution law of the substance composition in the single vector layer in the three-dimensional model, and correspond to the marked types of carbonate rock thin sections.

[0032] As a further improvement of this technical solution, in the type recognition unit, the output result of the BP neural network model is the predicted type of carbonate rock thin section, including the following steps:

[0033] Use Keras to construct a BP neural network model, compile the model, specify the loss function, optimizer, and evaluation metrics, use the atlas fingerprints of multiple carbonate rock thin sections as the training set data to train the model, use the test set data to evaluate the performance of the model, and use the trained model to predict new carbonate rock thin sections. Input the atlas fingerprints and output the type of rock thin section.

[0034] Compared with the prior art, the beneficial effects of the present invention:

[0035] In the carbonate rock image tagging system based on deep learning technology, technologies such as deep learning, multi-source data fusion, and atlas intelligent recognition are introduced into the identification of carbonate rock thin sections. By constructing atlas tags, establishing an identification method system, summarizing the correlations between typical carbonate rock thin section types and rock material compositions, textures, and physical property parameters, and establishing atlas recognition markers for typical carbonate rock thin section types, comprehensive analysis is carried out from the perspectives of diagrams, spectra, and multi-source physical property parameter information. The technical route is feasible, which is conducive to improving the accuracy of type recognition;

[0036] When the "atlas fingerprint" of the carbonate rock thin section is input into the BP neural network model of the type recognition unit, the predicted carbonate rock thin section type is output, optimizing and enhancing the work of traditional visual identification of thin sections, and providing technical support for the intelligent recognition, non-destructive testing, and in-situ parameter analysis of carbonate rock thin sections;

[0037] Moreover, in modern geological teaching and work, the virtual simulation platform formed by the carbonate rock image tagging system based on deep learning technology plays a crucial role. Through the system, geological phenomena can be more intuitively understood, providing technical support for the teaching of petrology and the construction of virtual simulation platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 is the overall structural principle block diagram of the present invention;

[0039] Figure 2 is the demonstration schematic diagram of the present invention.

[0040] The meanings of the various reference numerals in the figure are as follows:

[0041] 100, multi-source feature extraction unit; 200, atlas classification unit; 300, type recognition unit. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention. Embodiment

[0043] Please refer to Figure 1 - Figure 2 as shown, this embodiment provides a carbonate rock image tagging system based on deep learning technology, including a multi-source feature extraction unit 100, an atlas classification unit 200, and a type recognition unit 300;

[0044] The multi-source feature extraction unit 100 is used to establish a carbonate rock particle image dataset, extract the spatial trace information of the microscopic images of rock thin sections at different spatial scales using remote sensing image processing algorithms, and comprehensively utilize the cathodoluminescence images and elemental spectral plane scan images to extract the trace fabric and material composition information. It three-dimensionally models the structural characteristics of carbonate rocks (the structural characteristics include spatial trace information, trace fabric, and material composition information), and through transfer learning, constructs the three-dimensional spatial framework of various types of rock thin sections, realizing the analysis of multi-source and multi-level data in the context of the three-dimensional spatial framework, laying a spatial framework foundation for the subsequent fusion of carbonate rock structure data, spectral databases, and auxiliary attribute data;

[0045] The atlas classification unit 200 is used to screen the fabric trace and material composition characteristic atlas information of carbonate rocks, initially construct an atlas classifier, reflect the correlation between the material composition and the fabric trace through the Pearson correlation coefficient, summarize the content and spatial distribution law of its material composition, and explore its influencing factors;

[0046] The type recognition unit 300 is used to fuse the trace fabric and material composition characteristic atlas information in a unified three-dimensional spatial framework, establish the atlas fingerprint of the carbonate rock thin section, use the BP neural network model with the atlas fingerprint as the input, and comprehensively judge the type of the rock thin section by combining the coupling law of the carbonate rock thin section atlas information and the results of transfer learning. The output result is the predicted type of the carbonate rock thin section, avoiding the accuracy and migration problems faced by a single data source.

[0047] On the above basis, specifically:

[0048] When the multi-source feature extraction unit 100 extracts the spatial trace information, it includes the following steps:

[0049] Scan the two-dimensional digital images of various types of rock thin sections, classify and establish a graphic library to form a carbonate rock particle image dataset. The sources of the rock thin sections can be collected teaching materials, scientific research materials, and materials accumulated by predecessors. Preprocess the collected images, including operations such as denoising, enhancing contrast, and cropping, to ensure the image quality. According to the standard rock types of the obtained image materials, classify and establish a graphic library according to the annotation content to realize the classification of graphic materials;

[0050] Use the morphological, texture, color, brightness, and roughness characteristic parameters in the two-dimensional digital images to graphically describe the structural framework in the two-dimensional grid plane, including the positional and orientation change overlays to obtain the graphically averaged morphology, and extract the spatial trace information;

[0051] Among them, when obtaining the characteristic parameters: extract the morphological characteristics such as the shape, size, and boundary of the thin rock slice, extract the texture characteristics of the thin rock slice, such as the gray-level co-occurrence matrix GLCM, extract the color characteristics of the thin rock slice, such as RGB values or HSV values, extract the brightness characteristics of the thin rock slice, such as gray-level values, and extract the surface roughness characteristics of the thin rock slice;

[0052] When making an image description: determine the position of each feature in the two-dimensional grid, describe the orientation change of the feature, superimpose the image descriptions of multiple features together to form an average morphology, and enrich the image parameters through statistical methods, such as mean, variance, correlation, etc.

[0053] Furthermore, the multi-source feature extraction unit 100 comprehensively utilizes the cathodoluminescence image and the element spectral plane scan image to extract the trace fabric and material composition information, including the following steps:

[0054] Adjust the electron beam parameters of the electron microscope to focus the electron beam on the surface of the thin rock slice to excite the luminescence center in the sample. The energy spectrometer is installed on the electron microscope to record the X-ray signal emitted by the sample under the excitation of the electron beam, and convert the recorded X-ray signal into a digital image. Use the gray-level value to represent the luminescence intensity, and first perform preprocessing of image denoising and image enhancement on the thin slice microscopic image to keep the area where the carbonate rock particles are located with good edges and textures, and enhance the gray-level difference between the particle area and the background area in the image to form a cathodoluminescence image and an element spectral plane scan image;

[0055] Align the cathodoluminescence image and the element spectral plane scan image to fuse the information of the two images to generate a comprehensive image;

[0056] Use the clustering algorithm to classify the features, identify different mineral phases, and visually generate a trace fabric map and a material composition map. When visualizing, generate a pseudo-color image according to the clustering result, and draw the distribution contour map of different elements or minerals;

[0057] Trace fabric: Interpret the trace features in the image, such as fractures, bedding, deformation bands, etc., and identify the fractures and bedding in the rock through the analysis of morphological and texture features;

[0058] Material composition: Interpret the material composition information in the image, such as the distribution of different elements, mineral phases, etc., and identify different mineral phases and their distributions through the analysis of spectral and luminescence features.

[0059] It should be noted that the information fusion of the two images to generate a comprehensive image adopts a multi-scale fusion algorithm, including the following steps:

[0060] Read the cathodoluminescence image and the element spectral plane scan image, normalize the image data to the same dynamic range, such as between 0-1, and perform denoising processing on the image to reduce the influence of noise on subsequent processing;

[0061] Use feature point detection methods such as SIFT and ORB to extract the feature points of the image, match the feature points of two images, calculate the transformation matrix, align one image to another image, use wavelet transform to decompose the aligned image into sub-bands of different scales, and fuse the sub-bands of different scales to form a comprehensive image; It changes the limitations such as "blind men feeling an elephant" in traditional visual identification and single graphic recognition, and conducts comprehensive analysis from the perspectives of images, spectra, and multi-source physical property parameter information. The technical route is feasible, and the research results are conducive to improving the accuracy of type recognition.

[0062] Through transfer learning, construct the three-dimensional spatial framework of each type of rock thin section, including the following steps:

[0063] Read the comprehensive image, convert the image data into a format suitable for 3D modeling, use software such as GOCAD to create a 3D grid, and interpolate the comprehensive image data into the 3D grid;

[0064] Trace-based transfer learning: According to the trace features, construct a trace model for transfer learning to adapt to the classification task of trace features;

[0065] Material-based transfer learning: According to the material composition information, construct a transfer model for transfer learning to adapt to the classification task of material composition;

[0066] Coupling-based transfer learning: Combine the trace features and material composition information, extract the coupling features, and construct a coupling model for transfer learning to adapt to the classification task of coupling features;

[0067] Model-based transfer learning: Combine the trace features, material composition information, and coupling features, and construct a multi-modal model for transfer learning to adapt to the classification task of multi-modal features;

[0068] Apply the classification results to the three-dimensional spatial framework to generate a three-dimensional model of the carbonate rock thin section.

[0069] Preliminarily construct a map classifier in the map classification unit 200, including the following steps:

[0070] In the unified spatial scale, mark according to the different optical characteristics of the internal biological framework, grains, and cement substances in the three-dimensional model, screen the fabric trace and material composition feature map information of the main rock thin sections to form multiple single vector layers;

[0071] Analyze the content and spatial distribution law of the material composition in the single vector layer in the three-dimensional model and correspond it to the marked types of carbonate rock thin sections;

[0072] Among them, the random forest model RF is an ensemble learning method that combines multiple decision trees and can be effectively applied to the linear regression analysis of big data in earth science, which is conducive to the quantitative analysis of material composition. The artificial neural network ANN is an operation model composed of many nodes, which can have various connection methods, has the ability to process adaptive information, has a wide range of applications, and has unique advantages in the knowledge mining of non-linear data in geoscience. The support vector machine SVM can reduce the complexity of training data according to limited sample information and obtain the optimal number of support vectors of the model. It is applicable to the non-linear classification of the material composition and fabric traces of atypical carbonate rocks. Select a suitable model according to the situation. The Pearson correlation coefficient is a statistic used to reflect the degree of linear correlation between two variables. The correlation coefficient is represented by r, where n is the sample size, which are the observed values and means of the two variables respectively. r describes the degree of linear correlation strength between the two variables. The larger the absolute value of r, the stronger the correlation. The Pearson correlation coefficient can be used to reflect the correlation between different variables. Select the Pearson correlation coefficient to analyze the correlation between the material composition and the fabric traces, summarize the content and spatial distribution law of its material composition, and explore its influencing factors.

[0073] In the type recognition unit 300, the output result of the BP neural network model is the predicted type of carbonate rock thin section, including the following steps:

[0074] Use Keras to build a BP neural network model, compile the model, specify the loss function, optimizer and evaluation metrics, use the spectral fingerprints of multiple carbonate rock thin sections as the training set data to train the model, use the test set data to evaluate the performance of the model, and use the trained model to predict new carbonate rock thin sections. Input the spectral fingerprint and output the type of rock thin section;

[0075] The BP neural network is a multi-layer feedforward neural network, similar to the signal transmission process of human neurons. The overall structure is 3 layers: input layer - hidden layer - output layer. This model has great processing ability for non-linear data. To build a BP neural network model, first determine the number of neuron nodes in the input layer , the number of neuron nodes in the hidden layer , and the number of neuron nodes in the output layer . Secondly, determine the threshold of the hidden layer , the threshold of the output layer , learning rate, number of training times, minimum error of training target, activation function, etc. Among them, the connection weight between the input layer and the hidden layer is , and the connection weight between the hidden layer and the output layer is , a BP neural network model with different structures is constructed in Matlab. The neuron nodes in the input layer are the "atlas fingerprints" of the established thin sections, and the output result is the predicted type of carbonate rock thin sections. By comprehensively considering the coupling law of the atlas information of carbonate rock thin sections and the achievements of transfer learning, the identification of rock thin section types is carried out to avoid the problems of accuracy and transferability faced by a single data source.

[0076] In summary, taking grain limestone as an example, grain limestone has three types of material components. The trace composition map a and the material composition map b are obtained through the multi-source feature extraction unit 100, realizing the visual display of spatial trace information, as well as trace texture and material composition information. The atlas classification unit 200 respectively screens out the single vector layers of the three material components, which is conducive to analyzing the correlation between the material components and the texture traces, summarizing the content and spatial distribution law of its material components, and exploring its influencing factors. When determining the type of carbonate rock, the "atlas fingerprint" of the carbonate rock thin section is obtained again through the multi-source feature extraction unit 100 and the atlas classification unit 200 and input into the BP neural network model of the type recognition unit 300, and the predicted type of carbonate rock thin section is output, optimizing and improving the work of traditional visual inspection of thin sections, and providing technical support for the intelligent identification, non-destructive testing and in-situ parameter analysis of carbonate rock thin sections;

[0077] Moreover, in modern geological teaching and work, the virtual simulation platform formed by the carbonate rock image tagging system based on deep learning technology plays a crucial role. Through the system, geological phenomena can be more intuitively understood, providing technical support for the teaching of petrology and the construction of virtual simulation platforms.

[0078] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A carbonate rock image tagging system based on deep learning technology, characterized in that: It includes a multi-source feature extraction unit (100), a spectrum classification unit (200), and a type recognition unit (300); The multi-source feature extraction unit (100) is used to establish a carbonate rock particle image data set, extract spatial trace information from microscopic images of rock thin sections at different spatial scales using remote sensing image processing algorithms, and comprehensively use cathodoluminescence images and elemental spectral surface scan images to extract trace fabric and material composition information, three-dimensionally model the structural characteristics of carbonate rocks, and construct a three-dimensional spatial framework for each type of rock thin section through transfer learning; Specifically, align the cathodoluminescence image and the elemental spectral surface scan image to fuse the information of the two images to generate a comprehensive image. The construction of the three-dimensional spatial framework for each type of rock thin section through transfer learning includes the following steps: Read the comprehensive image, convert the image data into a format suitable for three-dimensional modeling, create a three-dimensional grid, and interpolate the comprehensive image data into the three-dimensional grid; Trace-based transfer learning: According to trace characteristics, construct a trace model for transfer learning to adapt to the classification task of trace characteristics; Material-based transfer learning: According to the material composition information, construct a transfer model for transfer learning to adapt to the classification task of material composition; Coupling-based transfer learning: Combine trace characteristics and material composition information, extract coupling characteristics, and construct a coupling model for transfer learning to adapt to the classification task of coupling characteristics; Model-based transfer learning: Combine trace characteristics, material composition information, and coupling characteristics to construct a multi-modal model for transfer learning to adapt to the classification task of multi-modal characteristics; Apply the classification result to the three-dimensional spatial framework to generate a three-dimensional model of the carbonate rock thin section; The spectrum classification unit (200) is used to screen the fabric trace and material composition characteristic spectrum information of carbonate rocks, initially construct a spectrum classifier, reflect the correlation between the material composition and the fabric trace through the Pearson correlation coefficient, summarize the content and spatial distribution law of its material composition, and explore its influencing factors; The type recognition unit (300) is used to fuse the trace fabric and material composition characteristic spectrum information in a unified three-dimensional spatial framework, establish a spectrum fingerprint of the carbonate rock thin section, use the BP neural network model with the spectrum fingerprint as the input, and comprehensively judge the type of the rock thin section based on the coupling law of the carbonate rock thin section spectrum information and the results of transfer learning. The output result is the predicted type of the carbonate rock thin section.

2. The carbonate rock image tagging system based on deep learning technology according to claim 1, wherein: When the multi-source feature extraction unit (100) extracts spatial trace information, it includes the following steps: Scan the two-dimensional digital images of multiple types of rock thin sections, classify and establish a graphic library to form a carbonate rock particle image data set; Use the morphological, texture, color, brightness, and roughness characteristic parameters in the two-dimensional digital image to graphically describe the structural framework in the two-dimensional grid plane, including the positional and orientation change overprints to obtain the graphical average morphology, and extract the spatial trace information.

3. The carbonate rock image tagging system based on deep learning technology according to claim 2, wherein: The multi-source feature extraction unit (100) comprehensively uses cathodoluminescence images and elemental spectral surface scan images to extract trace fabric and material composition information, including the following steps: Adjust the electron beam parameters of the electron microscope to focus the electron beam on the surface of the rock thin section to excite the luminescence centers in the sample. An energy spectrometer is installed on the electron microscope to record the X-ray signals emitted by the sample under electron beam excitation, and convert the recorded X-ray signals into digital images. Use grayscale values to represent the luminescence intensity. First, perform image denoising and image enhancement preprocessing on the thin section microscopic images to keep the areas where carbonate rock particles are located with good edges and textures, and enhance the grayscale difference between the particle areas and the background areas in the image to form a cathodoluminescence image and an elemental spectral surface scan image; Align the cathodoluminescence image and the elemental spectral surface scan image to fuse the information of the two images to generate a composite image; Use a clustering algorithm to classify the features, identify different mineral phases, and visually generate a trace fabric map and a material composition map.

4. The carbonate rock image tagging system based on deep learning technology according to claim 3, characterized in that: The information fusion of the two images to generate a composite image adopts a multi-scale fusion algorithm, including the following steps: Read the cathodoluminescence image and the elemental spectral surface scan image, and normalize the image data to the same dynamic range; Use a feature point detection method to extract the feature points of the images, match the feature points of the two images, calculate the transformation matrix, align one image to the other image, use wavelet transform to decompose the aligned image into sub-bands of different scales, and fuse the sub-bands of different scales to form a composite image.

5. The carbonate rock image tagging system based on deep learning technology according to claim 4, characterized in that: The structural features include spatial trace information, trace fabric, and material composition information.

6. The carbonate rock image tagging system based on deep learning technology according to claim 1, wherein: In the map classification unit (200), a map classifier is preliminarily constructed, including the following steps: In a unified spatial scale, mark according to different optical characteristics of the internal biological framework, grains, and cement substances in the three-dimensional model, screen the fabric trace and material composition map information of the main rock thin sections to form multiple single vector layers; Analyze the content and spatial distribution law of the material composition in the three-dimensional model of the single vector layer, and correspond to the marked carbonate rock thin section types.

7. The carbonate rock image tagging system based on deep learning technology according to claim 6, characterized in that: In the type recognition unit (300), the output result of the BP neural network model is the predicted carbonate rock thin section type, including the following steps: Use Keras to build a BP neural network model, compile the model, specify the loss function, optimizer, and evaluation metrics, use the map fingerprints of multiple carbonate rock thin sections as the training set data to train the model, use the test set data to evaluate the performance of the model, and use the trained model to predict new carbonate rock thin sections. Input the map fingerprint and output the rock thin section type.

Images