Quantitative characterization method and system for glutenite impurity group content based on CT digital core

Abstract

The invention belongs to the technical field of sandy conglomerate oil-gas exploration and development, and discloses a sandy conglomerate impurity group content quantitative characterization method and system based on a CT digital core. According to the method, a three-dimensional digital rock core image is obtained through CT scanning and preprocessed, an intelligent large model and a deep learning standard model are constructed, intelligent segmentation and volume content calculation of a miscellaneous base area are achieved, and the geological sweet spot section is judged with the miscellaneous base content, porosity and permeability as parameters. Compared with a traditional slice method, the method has the advantages that the impurity base segmentation IoU is improved to 0.89, the sweet spot / non-sweet spot identification accuracy rate reaches 88.9%, the single sample analysis time is shortened from 72 hours to 8 hours, the problems of low three-dimensional representation precision and poor efficiency are solved, accurate parameters are provided for oil and gas exploration and development, the efficiency is remarkably improved, and the cost is reduced.

Description

Technical Field

[0001] This invention belongs to the field of sandstone and conglomerate oil and gas exploration and development technology, specifically relating to a quantitative characterization method and system for the matrix content of sandstone and conglomerate based on CT digital cores. Background Technology

[0002] In conglomerate reservoirs, the content of matrix (muddy and fine silt interstitial material with a particle size <0.0625mm) directly affects pore connectivity and permeability. Quantitative characterization of matrix content in conglomerate has long been a world-class technical challenge that has remained largely unresolved in oil and gas exploration and development engineering.

[0003] Currently, characterization techniques for matrix content in sandstone and conglomerate mainly rely on qualitative analysis of large thin sections. However, the limited field of view of thin sections can only depict matrix development characteristics within a small area of ​​a two-dimensional slice, failing to adequately portray the lateral variations of the sandstone and conglomerate matrix. Current methods only obtain random probability values ​​and cannot effectively characterize the true three-dimensional spatial state of the sandstone and conglomerate matrix content. This results in low accuracy in obtaining information about the development of the sandstone and conglomerate matrix, consequently leading to a technical problem of low accuracy in oil and gas exploration and development of sandstone and conglomerate reservoirs.

[0004] Therefore, there is an urgent need to provide a three-dimensional quantitative characterization method for matrix content development in sandstone and conglomerate to improve the accuracy and efficiency of oil and gas exploration and development in sandstone and conglomerate reservoirs. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the above-mentioned background technology and provide a quantitative characterization method for the matrix content of sandstone and conglomerate based on CT digital cores. This method aims to solve the technical problem that inaccurate measurements in the prior art lead to low accuracy in obtaining the matrix development status of sandstone and conglomerate, which in turn leads to low accuracy in predicting oil and gas exploration and development in sandstone and conglomerate.

[0006] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution: In a first aspect, the present invention provides a quantitative characterization method for the matrix content of sandstone and conglomerate based on CT digital cores, comprising: S1. Perform CT scanning on the sandstone core to obtain three-dimensional digital core image data, and preprocess the three-dimensional digital core image data; S2. Based on sandstone and conglomerate image datasets with different geological conditions and genetic types, an intelligent large model of sandstone and conglomerate matrix development and distribution images is constructed through pre-segmentation, feature extraction, and deep learning training. S3. Based on the intelligent large model, determine the appropriate image segmentation method, perform intelligent segmentation on the three-dimensional digital core image data, and construct a segmentation digital model of sandstone gravel and matrix; S4. Based on the aforementioned segmentation digital model, a convolutional neural network architecture is adopted, and three-dimensional digital core image data is used as training data. Through joint optimization of Dice loss and cross-entropy loss, a deep learning standard model for segmentation of gravel and matrix in sandstone and conglomerate is constructed. S5. Based on the aforementioned deep learning standard model for segmentation, matrix region segmentation is performed on the preprocessed three-dimensional digital core image data, and three-dimensional binary segmentation results identifying matrix and non-matrix voxels are output. The matrix volume content is calculated by statistically analyzing the volume of matrix voxels and the total volume of the core. S6. Based on the matrix volume content, key parameters are provided for determining the sweet spot of the oil and gas reservoir.

[0007] As a further optimization of the present invention, in step S5, the formula for calculating the heterobase volume content is: .

[0008] As a further optimization of the present invention, in step S6, the key parameters also include porosity and permeability. When V matrix ≤ 15%, porosity ≥ 8%, and permeability ≥ 1mD, it is determined to be a geological sweet spot.

[0009] As a further optimization of the present invention, in step S1, the CT scanning parameters include: resolution 5-20μm, voltage 80-140kV, current 50-200μA, and slice thickness ≤0.625mm.

[0010] As a further optimization of the present invention, in step S1, the three-dimensional digital core image data is preprocessed, including grayscale normalization and data enhancement processing of the CT image, wherein the data enhancement includes rotation, scaling and Gaussian blur.

[0011] As a further optimization of the present invention, in step S2, the pre-segmentation includes threshold pre-segmentation and region growing pre-segmentation, and the feature extraction includes multispectral fusion and multi-scale feature fusion; the feature extraction process is optimized by using the convolutional block attention module CBAM and the receptive field module RFB.

[0012] As a further optimization of the present invention, in step S2, a dataset of sandstone and conglomerate images from more than 10 sedimentary basins and more than 5 genetic types is collected, with a sample size of no less than 2000 sets, and the dataset is divided into training set, validation set and test set in an 8:1:1 ratio.

[0013] As a further optimization of the present invention, in step S5, the Otsu thresholding method is used to binarize the segmentation result.

[0014] Secondly, the present invention provides a quantitative characterization method for the matrix content of sandstone and conglomerate based on CT digital cores, for implementing the aforementioned quantitative characterization method for the matrix content of sandstone and conglomerate based on CT digital cores, the system comprising: The image acquisition and preprocessing module is used to acquire three-dimensional digital core image data of sandstone and conglomerate cores through CT three-dimensional scanning and to preprocess the images. The intelligent large model construction module is used to construct intelligent large models of sandstone and conglomerate matrix development and distribution images based on sandstone and conglomerate image datasets with different geological conditions and different genetic types through pre-segmentation, feature extraction and deep learning training. The image segmentation module is used to determine the appropriate image segmentation method based on the intelligent large model, intelligently segment the three-dimensional digital core image data, and construct a segmentation digital model of gravel and matrix in sandstone and conglomerate. The deep learning model training module is used to construct a standard deep learning model for the segmentation of gravel and matrix of sandstone and conglomerate based on the segmentation digital model, using a convolutional neural network architecture and three-dimensional digital core image data as training data, and through joint optimization of Dice loss and cross-entropy loss. The results analysis module is used to extract the digital distribution results of matrix content using the segmentation deep learning standard model, and to determine the key parameters of the sweet spot segment of the oil and gas reservoir by combining geological data.

[0015] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program, the computer program being stored in the memory and configured to be executed by the processor to implement the quantitative characterization method for matrix content of sandstone and conglomerate based on CT digital cores.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which is executed by a processor to implement the quantitative characterization method for matrix content of sandstone and conglomerate based on CT digital cores.

[0017] Compared with existing technologies, the quantitative characterization method and system for matrix content in sandstone and conglomerate based on CT digital cores proposed in this invention exhibits the following significant advantages and beneficial effects: The quantitative characterization method for the matrix content of conglomerate based on CT digital cores provided in this invention first uses digital core images of conglomerate matrix content. Then, based on an intelligent large-scale image model of conglomerate matrix development and distribution, an image segmentation method is used to intelligently segment the conglomerate matrix, obtaining a segmented digital model of the conglomerate gravel and matrix. This improves the accuracy of the constructed three-dimensional digital model of the conglomerate matrix distribution characteristics. Based on the segmented digital model of the conglomerate gravel and matrix, a convolutional neural network is used for deep learning to create a standard deep learning model, which can accurately identify the distribution characteristics of the conglomerate matrix in the image. This model is then applied to extract the distribution range and digitization results of the conglomerate matrix, significantly improving the accuracy of digitization result extraction compared to traditional methods. This invention solves the problem of inaccurate existing measurements, thereby improving the accuracy and efficiency of oil and gas exploration and development prediction for conglomerate oil and gas reservoirs, increasing the efficiency of oil and gas exploration and development in conglomerate oil and gas reservoirs, and reducing oil and gas exploration and development costs. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a three-dimensional image of the conglomerate development obtained by CT scanning of a conglomerate core.

[0020] Figure 2 This is a digital quantitative model diagram of matrix development in sandstone and conglomerate obtained by matrix region segmentation using a standard deep learning model.

[0021] Figure 3 This is a graph showing the heterobase volume content results obtained through statistical calculation. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0023] The flowcharts used in this invention illustrate operations implemented according to some embodiments of the invention. It should be understood that the operations in the flowcharts may be implemented out of order, and steps without logical contextual relationships may be reversed or performed simultaneously. Those skilled in the art, guided by the content of this invention, may add one or more additional operations to the flowcharts, or remove one or more operations from the flowcharts. These functional entities may be implemented in software.

[0024] In this invention, the term "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0025] Example 1 This invention provides a quantitative characterization method for matrix content in sandstone and conglomerate based on CT digital cores. The method includes: S1. Perform CT scans on the sandstone and conglomerate cores to obtain three-dimensional digital core image data, and preprocess the three-dimensional digital core image data.

[0026] Specifically, a micron-level CT scanner was used to perform a comprehensive 360° scan of the sandstone and conglomerate core samples. Based on the core characteristics, scanning parameters such as voltage, current, and resolution were meticulously optimized to ensure the acquisition of high-quality, high-precision three-dimensional digital core images, clearly revealing the fine internal structure of the core, including pores, gravel, and matrix. The resolution was controlled at 5-20 μm to ensure the identification of the boundaries between matrix and gravel. The voltage was controlled at 80-140 kV, and the current at 50-200 μA to balance penetration and image contrast. For high-density conglomerate, the voltage could be appropriately increased to 120 kV to enhance X-ray penetration. The layer thickness was controlled to ≤0.625 mm to ensure the accuracy of the three-dimensional reconstruction and avoid the loss of matrix distribution information due to excessive layer thickness. The scanning range needed to cover the entire core sample, typically with a diameter of 25-50 mm and a length of 50-100 mm, to obtain complete three-dimensional matrix distribution information.

[0027] After scanning, further data preprocessing is required. Image enhancement employs methods such as rotation (±15°), scaling (0.8-1.2 times), and Gaussian blur (σ=1-2) to improve the model's generalization ability. Gray-level normalization normalizes the gray-level values ​​of the CT image to the [0,1] interval, eliminating the influence of device differences on segmentation.

[0028] S2. Based on sandstone and conglomerate image datasets with different geological conditions and genetic types, an intelligent large model of sandstone and conglomerate matrix development and distribution images is constructed through pre-segmentation, feature extraction, and deep learning training.

[0029] Specifically, a comprehensive and rich dataset of sandstone and conglomerate images from different geological regions with different formation mechanisms (such as alluvial fan formation and braided river formation) will be constructed. When building the model, the collection of image information should ensure a balanced proportion of images under different geological conditions, and that images of different formation types under each geological condition are sufficiently representative.

[0030] In this embodiment, CT images of sandstone and conglomerate from more than 10 sedimentary basins and more than 5 genetic types (such as alluvial fans, braided rivers, etc.) were collected, with a sample size of no less than 2000 sets. The images were denoised (Gaussian filtering), grayscale normalized (0-1 normalization), and labeled (artificially delineating the boundaries between matrix and gravel), and divided into training set, validation set, and test set in an 8:1:1 ratio.

[0031] The pre-segmentation process can employ a combination of methods, such as threshold pre-segmentation followed by simple region growing pre-segmentation, to improve the accuracy of pre-segmentation. This initially separates the main components in the core image, such as gravel and matrix, laying the foundation for subsequent precise feature extraction. For the pre-segmented image, a series of parameters effectively characterizing matrix development are extracted, such as grayscale features, texture features (obtained through methods like gray-level co-occurrence matrix), and morphological features (area, perimeter, shape factor, etc.). Feature extraction employs multispectral fusion and multi-scale feature fusion. Multispectral fusion combines single-polarized and multi-angle orthogonal polarized microscopic images to extract spectral features such as interference colors and extinction, improving the distinguishability between matrix and gravel. Multi-scale feature fusion uses a pyramid pooling module (PPM) to fuse feature maps at different scales, enhancing the ability to identify matrix particles of different sizes. The model architecture uses UNet. 3+ The network embeds a receptive field module (RFB) and a convolutional block attention module (CBAM) in the encoding layer to expand the receptive field and focus on the target region. Dense skip connections are introduced in the decoder to fuse semantic information from different levels, improving segmentation accuracy. The model performance is evaluated using five-fold cross-validation to ensure a mean intersection-over-union (mIoU) ≥ 85% and an F1 score ≥ 90%. By continuously adjusting the network parameters, the model can accurately learn the intrinsic correlation between different features and matrix development distribution, constructing a high-performance intelligent large-scale model for sandstone and conglomerate matrix development distribution images.

[0032] S3. Based on the intelligent large model, determine the appropriate image segmentation method, perform intelligent segmentation on the three-dimensional digital core image data, and construct a segmentation digital model of sandstone gravel and matrix.

[0033] Specifically, the preprocessed 3D digital core image data is input into a pre-constructed intelligent large-scale model. Based on the learned features and patterns, the model accurately determines the boundaries between different components, thereby identifying the most suitable image segmentation algorithm and parameter combination for the current core image. For example, if the model determines that a certain part of the image features highly match the matrix features, and the boundary features conform to a specific pattern, then a segmentation method based on edge detection and region growing is selected to segment that part. Following the determined segmentation method, the gravel and matrix in the 3D digital core image data are precisely segmented pixel by pixel, constructing a segmentation digital model that clearly identifies the spatial distribution of gravel and matrix. In the model, each voxel is clearly labeled as belonging to either gravel or matrix.

[0034] S4. Based on the aforementioned segmentation digital model, a convolutional neural network architecture is adopted, and three-dimensional digital core image data is used as training data. Through joint optimization of Dice loss and cross-entropy loss, a deep learning standard model for segmentation of gravel and matrix in sandstone and conglomerate is constructed.

[0035] Specifically, the following training parameters need to be clarified: Loss function: The Dice loss and cross-entropy loss are jointly optimized, and the formula is as follows:

[0036] Among them, the Dice coefficient measures the overlap between the segmentation result and the true value, CE is the cross-entropy loss, and α is set to 0.5 to balance the two losses.

[0037] The optimizer uses the AdamW optimizer, with a learning rate initialized to 1e-4 and dynamically adjusted using cosine annealing. Batch size: 16-32, adjusted based on GPU memory. Epochs: 100-200, terminating when the validation set loss stabilizes. Data partitioning: Divided into training, validation, and test sets in an 8:1:1 ratio, using five-fold cross-validation to evaluate model performance. Evaluation metrics: Mean Intersection over Union (mIoU) ≥ 85%, F1 score ≥ 90%, to ensure segmentation accuracy.

[0038] S5. Based on the aforementioned deep learning standard model for segmentation, the preprocessed three-dimensional digital core image data is segmented into matrix regions, and the three-dimensional binary segmentation results that identify matrix and non-matrix voxels are output. The matrix volume content is calculated by statistically analyzing the volume of matrix voxels and the total volume of the core.

[0039] Specifically, the following image processing steps are needed to achieve quantitative characterization of heterozygosity: The Otsu thresholding method was used to binarize the segmentation results, marking the matrix region as 1 and the gravel and pore regions as 0. The number of voxels with a value of 1 in the binarized image was counted, and multiplied by the volume of a single voxel (the cube of the voxel's side length) to obtain the total volume of the matrix.

[0040] The formula for calculating the volume content of heterologous groups is: .

[0041] S6. Based on the matrix volume content, key parameters are provided for determining the sweet spot of the oil and gas reservoir.

[0042] Specifically, after obtaining the matrix volume content, key geological parameters such as porosity and permeability are comprehensively considered. Porosity can be calculated based on the number of pore space voxels using existing methods in digital core analysis; permeability can be calculated using numerical simulation methods based on digital cores, such as the lattice Boltzmann method. For the target oil and gas reservoir, appropriate threshold ranges for porosity, permeability, and matrix volume content are pre-defined. If the matrix volume content of a certain core segment is within a specific reasonable range, and the porosity and permeability also meet the corresponding threshold requirements, then this core segment is determined to be a sweet spot in the oil and gas reservoir. These sweet spots have important guiding significance for oil and gas exploration and development and are key areas for subsequent focus and development.

[0043] In this embodiment, a reservoir is classified as a geological sweet spot if V matrix content is ≤15%, porosity is ≥8%, and permeability is ≥1 mD; otherwise, it is classified as a non-geological sweet spot. The upper limit for the effective reservoir's clay matrix content is 15%; exceeding this value significantly reduces reservoir permeability. The effective reservoir's permeability must be ≥1 mD to ensure effective oil and gas flow. Porosity must be ≥8% to ensure sufficient storage space and flow channels.

[0044] Based on the above design concept, taking a sandstone-conglomerate body in an oilfield as an example, the matrix content of the sandstone-conglomerate body was characterized using the characterization method of this invention and the traditional thin section method, respectively. The results are shown in [Figure number missing]. Figure 1-3 See Table 1.

[0045] Table 1

[0046] Figure 1 This is a three-dimensional image of the conglomerate development obtained by CT scanning of a conglomerate core. Figure 2 This is a digital quantitative model diagram of matrix development in sandstone and conglomerate obtained by matrix region segmentation using a standard deep learning model. Figure 3The graph shows the matrix volume content obtained through statistical calculation. The matrix content of the 2355.15-2356.05m well section was characterized using the characterization method of this invention. The results showed a matrix content of 40-60%, with a matrix content greater than 15%, indicating that this well section is not a geological sweet spot.

[0047] As shown in Table 1, the method of this invention improves the accuracy of matrix content characterization in conglomerate bodies, resulting in higher efficiency, lower cost, and higher accuracy in sweet spot / non-sweet spot localization. Specifically, the matrix segmentation IoU reaches 0.89±0.03 (compared to only 0.62±0.05 in the traditional method); the single-sample analysis time is reduced from 72 hours to 8 hours; and the sweet spot / non-sweet spot localization accuracy reaches 92% (compared to 68% in the traditional method).

[0048] Example 2 Based on a general inventive concept, embodiments of this application provide a quantitative characterization system for the matrix content of sandstone and conglomerate based on CT digital cores, the system comprising: The image acquisition and preprocessing module is used to acquire three-dimensional digital core image data of sandstone and conglomerate cores through CT three-dimensional scanning and to preprocess the images. The intelligent large model construction module is used to construct intelligent large models of sandstone and conglomerate matrix development and distribution images based on sandstone and conglomerate image datasets with different geological conditions and different genetic types through pre-segmentation, feature extraction and deep learning training. The image segmentation module is used to determine the appropriate image segmentation method based on the intelligent large model, intelligently segment the three-dimensional digital core image data, and construct a segmentation digital model of gravel and matrix in sandstone and conglomerate. The deep learning model training module is used to construct a standard deep learning model for the segmentation of gravel and matrix of sandstone and conglomerate based on the segmentation digital model, using a convolutional neural network architecture and three-dimensional digital core image data as training data, and through joint optimization of Dice loss and cross-entropy loss. The results analysis module is used to extract the digital distribution results of matrix content using the segmentation deep learning standard model, and to determine the key parameters of the sweet spot segment of the oil and gas reservoir by combining geological data.

[0049] Example 3 Based on a general inventive concept, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program, wherein the computer program is stored in the memory and configured to be executed by the processor to implement the quantitative characterization method for matrix content of sandstone and conglomerate based on CT digital cores.

[0050] Example 4 Based on a general inventive concept, embodiments of this application provide a computer-readable storage medium storing a computer program that is executed by a processor to implement the quantitative characterization method for matrix content of sandstone and conglomerate based on CT digital cores.

[0051] In summary, the quantitative characterization method and system for matrix content in sandstone and conglomerate based on CT digital cores provided in this invention have the following significant advantages: (1) By controlling the CT scan resolution to 5-20μm and the voltage to 80-140kV, the boundary details of matrix and gravel can be accurately captured. Combined with multispectral fusion (such as single polarized light and cross polarized light microscopic images) and multi-scale feature fusion (pyramid pooling module PPM), it can effectively distinguish matrix and mineral components of different particle sizes and maintain segmentation accuracy even under complex geological conditions such as high-density conglomerate.

[0052] (2) In the preprocessing stage, data enhancement methods such as rotation, scaling, and Gaussian blur are used to make the model more robust to core scanning angle deviation and noise interference. The training set covers more than 10 sedimentary basins and more than 5 genetic types of sandstone and conglomerate data to ensure that the model can output reliable results stably under different geological scenarios.

[0053] (3) The time required for the traditional thin-film method has been shortened from 72 hours / sample to 8 hours / sample. Through the cascade processing of intelligent large model and deep learning standard model, the entire process from image acquisition to matrix content calculation is automated, which reduces human intervention error and greatly improves work efficiency.

[0054] (4) Compared with traditional two-dimensional thin sections which can only depict local features, this method directly outputs the spatial distribution of matrix through a three-dimensional digital core model, avoiding misjudgment caused by sample randomness, reducing the exploration cost of a single well by about 30%, and is especially suitable for the efficient evaluation of complex oil reservoirs such as deep sea and sandstone gas.

[0055] (5) In addition to matrix content, a comprehensive evaluation system is constructed by combining key parameters such as porosity (≥8%) and permeability (≥1mD), which improves the accuracy of sweet spot / non-sweet spot identification from 68% of the traditional method to 92%, providing more accurate geological basis for horizontal well trajectory design and fracturing scheme optimization.

[0056] (6) The characterization system adopts a modular design including image acquisition, model building, and segmentation analysis, which can be seamlessly connected to the existing oilfield digital platform. Its deep learning model supports API interface expansion, which is convenient for integration with well logging and seismic data, and promotes the transformation of oil and gas exploration from "experience-driven" to "data-driven".

[0057] (7) By optimizing feature extraction through the convolutional block attention module (CBAM) and receptive field module (RFB), the model achieves voxel-level accuracy in recognizing micro-nano matrix, providing AI technology support for the entire chain of "sweet spot prediction - reserve assessment - development plan" for unconventional oil and gas reservoirs, and promoting the leap from qualitative description to quantitative prediction in geological research.

[0058] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0059] The contents not described in detail in this specification are existing technologies known to those skilled in the art.

Claims

1. A quantitative characterization method for matrix content in sandstone and conglomerate based on CT digital cores, characterized in that, include: S1. Perform CT scanning on the sandstone core to obtain three-dimensional digital core image data, and preprocess the three-dimensional digital core image data; S2. Based on sandstone and conglomerate image datasets with different geological conditions and genetic types, an intelligent large model of sandstone and conglomerate matrix development and distribution images is constructed through pre-segmentation, feature extraction, and deep learning training. S3. Based on the intelligent large model, determine the appropriate image segmentation method, perform intelligent segmentation on the three-dimensional digital core image data, and construct a segmentation digital model of sandstone gravel and matrix; S4. Based on the aforementioned segmentation digital model, a convolutional neural network architecture is adopted, and three-dimensional digital core image data is used as training data. Through joint optimization of Dice loss and cross-entropy loss, a deep learning standard model for segmentation of gravel and matrix in sandstone and conglomerate is constructed. S5. Based on the aforementioned deep learning standard model for segmentation, matrix region segmentation is performed on the preprocessed three-dimensional digital core image data, and three-dimensional binary segmentation results identifying matrix and non-matrix voxels are output. The matrix volume content is calculated by statistically analyzing the volume of matrix voxels and the total volume of the core. S6. Based on the matrix volume content, key parameters are provided for determining the sweet spot of the oil and gas reservoir.

2. The quantitative characterization method for matrix content in sandstone and conglomerate based on CT digital cores according to claim 1, characterized in that, In step S5, the formula for calculating the heterobase volume content is: 。 3. The quantitative characterization method for matrix content in sandstone and conglomerate based on CT digital cores according to claim 1, characterized in that, In step S5, the Otsu thresholding method is used to binarize the segmentation results.

4. The quantitative characterization method for matrix content in sandstone and conglomerate based on CT digital cores according to claim 1, characterized in that, In step S6, the key parameters also include porosity and permeability. When V matrix ≤ 15%, porosity ≥ 8%, and permeability ≥ 1mD, it is determined to be a geological sweet spot.

5. The quantitative characterization method for matrix content in sandstone and conglomerate based on CT digital cores according to claim 1, characterized in that, In step S1, the CT scan parameters include: resolution 5-20μm, voltage 80-140kV, current 50-200μA, and slice thickness ≤0.625mm.

6. The quantitative characterization method for matrix content in sandstone and conglomerate based on CT digital cores according to claim 1, characterized in that, In step S1, the three-dimensional digital core image data is preprocessed, including grayscale normalization and data enhancement processing of the CT image. The data enhancement includes rotation, scaling and Gaussian blur.

7. The quantitative characterization method for matrix content in sandstone and conglomerate based on CT digital cores according to claim 1, characterized in that, In step S2, the pre-segmentation includes threshold pre-segmentation and region growing pre-segmentation, and the feature extraction includes multispectral fusion and multi-scale feature fusion; the feature extraction process is optimized using the convolutional block attention module CBAM and the receptive field module RFB.

8. A quantitative characterization system for matrix content in sandstone and conglomerate based on CT digital cores, characterized in that, The system is used to implement the quantitative characterization method for matrix content of sandstone and conglomerate based on CT digital cores as described in any one of claims 1 to 7, the system comprising: The image acquisition and preprocessing module is used to acquire three-dimensional digital core image data of sandstone and conglomerate cores through CT three-dimensional scanning and to preprocess the images. The intelligent large model construction module is used to construct intelligent large models of sandstone and conglomerate matrix development and distribution images based on sandstone and conglomerate image datasets with different geological conditions and different genetic types through pre-segmentation, feature extraction and deep learning training. The image segmentation module is used to determine the appropriate image segmentation method based on the intelligent large model, intelligently segment the three-dimensional digital core image data, and construct a segmentation digital model of gravel and matrix in sandstone and conglomerate. The deep learning model training module is used to construct a standard deep learning model for the segmentation of gravel and matrix of sandstone and conglomerate based on the segmentation digital model, using a convolutional neural network architecture and three-dimensional digital core image data as training data, and through joint optimization of Dice loss and cross-entropy loss. The results analysis module is used to extract the digital distribution results of matrix content using the segmentation deep learning standard model, and to determine the key parameters of the sweet spot segment of the oil and gas reservoir by combining geological data.

9. An electronic device comprising a memory, a processor, and a computer program, characterized in that: The computer program is stored in the memory and configured to be executed by the processor to implement the quantitative characterization method for matrix content of sandstone and conglomerate based on CT digital cores as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which is executed by a processor to implement the quantitative characterization method for matrix content of sandstone and conglomerate based on CT digital cores as described in any one of claims 1 to 7.

Images