A deep learning-based igneous rock automatic identification method and system

Abstract

The present application relates to a kind of automatic identification method and system of igneous rock based on deep learning, comprising: obtaining target seismic data;Using the three-dimensional convolutional neural network U-net model based on deep learning established in advance is identified to target seismic data, and the igneous rock identification result is obtained;Wherein, the three-dimensional convolutional neural network U-net model based on deep learning established in advance is trained using the igneous rock label data training set obtained by actual seismic data processing.The present application proposes to use actual data to make label training set, determine the range by seismic attribute, and form label data by fine interpretation and attribute modeling for target area;Label data set is input into the optimized convolutional neural network U-net and is trained, and the identification result of the analysis prediction block is obtained, and it is concluded that the present application has higher accuracy compared with traditional geologic body sculpture method.The present application can be widely applied in seismic data interpretation, lithofacies identification field.

Description

Technical Field

[0001] This invention belongs to the field of seismic data interpretation and lithofacies identification, and involves related content such as seismic attributes, deep learning, and convolutional neural networks. Specifically, it relates to an automatic identification method and system for igneous rocks based on deep learning. Background Technology

[0002] Seismic data contains a wealth of geological information, such as stratigraphic boundaries, faults, fractures, and karst caves. Accurately locating this information requires geophysicists to process and interpret the seismic data. Seismic interpreters use geological information from well logging and other sources to perform structural and lithological interpretations, where experience is crucial. Interpretation requires interpreters to be proficient in specialized software and to use their expertise and experience to find and identify relevant information from large amounts of seismic data. However, this approach still has some limitations. For example, many similar geological phenomena exhibit strong regularity and characteristics in seismic data, which can be identified using application-based and intelligent technologies; seismic interpretation relies on the experience of professionals, but errors are inevitable; and manual interpretation by professionals is time-consuming and labor-intensive. To reduce costs and improve efficiency, the application of intelligent technologies is the main solution. With advancements in seismic data acquisition and processing technologies, wide-azimuth and high-density seismic data now possess high signal-to-noise ratios, fidelity, and resolution. Given that algorithms, computing power, and data availability are all sufficient, automated interpretation technology and data mining represent the future direction of development.

[0003] The research revealed that igneous rock intrusions and eruptions caused by volcanic activity are of significant research value for tectonic movement analysis and geological resource exploration. In petroleum geophysical exploration, accurate identification of igneous rock boundaries can precisely analyze whether oil and gas traps have been damaged, reducing exploration risks. Furthermore, accurate identification of igneous rocks can aid in locating migration pathways and traps. Many igneous rock intrusions have formed favorable migration pathways, connecting mature source rocks and potentially forming oil and gas reservoirs. Igneous rock intrusions can also create structural fractures, with dome-shaped anticline and fault-bounded anticline traps providing accumulation sites for oil and gas.

[0004] Traditional igneous rock identification primarily relies on core samples, thin sections, well logging interpretation results, and various curve intersection charts to comprehensively analyze and determine the type and characteristics of igneous rocks. This involves interpreting and sculpting surrounding igneous rocks using a combination of well and seismic data. This method achieves the desired results with high accuracy when sufficient drilling data, such as core and thin section data, is available. However, in actual exploration, to reduce risk, drilling locations often avoid volcanic areas, resulting in lower exploration levels. In areas with fewer wells, identifying igneous rocks based solely on core and well logging data becomes impractical. Research into these methods reveals that previous approaches mainly relied on seismic facies analysis, using forward modeling and other methods to determine the location and lithofacies type of igneous rocks. These techniques incorporate seismic attributes, waveform classification, and multi-attribute fusion to identify igneous lithofacies and characterize igneous bodies. Because igneous rocks exhibit significant differences from surrounding rocks, energy-based and continuity-based detection attributes can effectively identify and characterize most igneous lithofacies. However, the engraving results could not accurately depict the volcanic conduit and the rock mass at the stratigraphic interface. Subsequently, the selection of a threshold for engraving further affected the accuracy of the results. Summary of the Invention

[0005] To address the aforementioned problems, the purpose of this invention is to provide a method and system for automatic identification of igneous rocks based on deep learning, which can accurately and quickly identify igneous rock bodies in seismic data.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides an automatic identification method for igneous rocks based on deep learning, comprising the following steps:

[0008] Acquire target earthquake data;

[0009] The target seismic data was identified using a pre-established deep learning-based three-dimensional convolutional neural network U-net model to obtain igneous rock identification results. The pre-established deep learning-based three-dimensional convolutional neural network U-net model was trained using a training set of igneous rock label data obtained from the processing of actual seismic data.

[0010] Furthermore, the method for establishing a labeled dataset based on actual earthquake data includes:

[0011] Based on actual seismic data and drilling data, the igneous facies characteristics, distribution range, and development stages corresponding to the actual seismic data are determined by using three-dimensional visualization functions.

[0012] Lithology is identified based on seismic attributes, and a refined seismic interpretation method is used as a constraint. Seed point automatic picking technology is used to carve the identified igneous facies to obtain the original igneous label dataset.

[0013] By moving the cutting position, the original igneous rock label dataset is augmented to obtain the igneous rock label data training set.

[0014] Furthermore, the analysis using three-dimensional visualization function refers to determining the distribution and development scale and development stages of igneous rocks corresponding to the actual seismic data based on the seismic frequency, amplitude characteristics, and development location patterns of each igneous facies.

[0015] Furthermore, when identifying rock masses based on seismic attributes, the root mean square amplitude attribute is used for intrusive and effusive igneous rocks, while the variance attribute is used for volcanic conduit igneous rocks.

[0016] Furthermore, the training of the U-net model, a three-dimensional convolutional neural network based on deep learning, using a training set of igneous rock label data obtained from actual seismic data processing, includes:

[0017] Normalize the training set of igneous rock label data;

[0018] The pre-constructed deep learning-based 3D convolutional neural network U-net model was trained using a normalized igneous rock label data training set.

[0019] Furthermore, the normalization process for the igneous rock label data training set refers to dividing the dataset into cubic data with 256 sampling points vertically and 256 main survey lines and connecting survey lines.

[0020] Furthermore, the deep learning-based 3D convolutional neural network U-net model includes:

[0021] The encoding module is used to extract features from the normalized igneous rock label data training set to obtain feature data.

[0022] The decoding module is used to restore the learned features to the original image size and obtain the recognition result.

[0023] Secondly, the present invention provides an automatic identification system for igneous rocks based on deep learning, comprising:

[0024] The dataset acquisition module is used to acquire target seismic data;

[0025] The automatic identification module is used to identify target seismic data using a pre-established deep learning-based three-dimensional convolutional neural network U-net model to obtain igneous rock identification results; wherein, the pre-established deep learning-based three-dimensional convolutional neural network U-net model is trained using a training set of igneous rock label data obtained from the processing of actual seismic data.

[0026] Thirdly, the present invention provides a computer-readable storage medium for storing one or more programs, said one or more programs including instructions that, when executed by a computing device, cause the computing device to perform any method.

[0027] Fourthly, the present invention provides a computing device comprising: one or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing any method.

[0028] The present invention has the following advantages due to the adoption of the above technical solutions:

[0029] 1. This invention improves upon the traditional U-net network for automatic identification of igneous rocks, achieving significantly higher efficiency than traditional 3D geological body carving methods while maintaining high accuracy. It has been proven that the U-net network trained for 25 epochs achieves an accuracy of 98.2% in its final model, enabling efficient and accurate identification of predicted blocks.

[0030] 2. This invention uses a change in the starting position of the cut to quickly augment data without altering its geological significance, thereby improving the robustness of the network model and the accuracy of the results.

[0031] 3. The training results of this invention can be incorporated into the project work area to assist in related research work.

[0032] Therefore, this invention can be widely applied in the fields of seismic data interpretation and lithofacies identification. Attached Figure Description

[0033] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. In the drawings:

[0034] Figure 1 A flowchart of automatic identification of igneous rock masses based on deep learning provided for embodiments of the present invention;

[0035] Figure 2 This is a diagram showing the overall structure of the optimized U-net network.

[0036] Figure 3 A comparison diagram of volcanic rock facies and seismic facies characteristics in the example area;

[0037] Figure 4a and Figure 4bThis is a typical profile of a depression-fissure volcanic conduit in the example area, where, Figure 4a This is a typical seismic profile of a depression-fissure volcanic conduit in the example area; Figure 4b This is a model diagram of a volcano along a seismic profile.

[0038] Figure 5 This is a seismic profile of a typical volcanic system in the example area;

[0039] Figure 6 This is a schematic diagram of igneous rocks from the original seismic profile.

[0040] Figure 7 This is a schematic diagram for identifying igneous rocks based on root-mean-square amplitude property profiles.

[0041] Figure 8 A schematic diagram for identifying igneous rocks based on variance attribute profiles;

[0042] Figure 9 The results of sculpting intrusive facies rock bodies using the geological body sculpting module;

[0043] Figure 10 This is a distribution diagram of the training and test sets;

[0044] Figure 11a and Figure 11b Create a visualization of the sculpted boundary constraints for the labeled dataset, where... Figure 11a Automatic tracking of the upper and lower interfaces of intrusive rocks; Figure 11b Automatic 3D spatial display of intrusive rock interface tracking;

[0045] Figure 12 To create a 3D rendering of an igneous geological body sculpted using geological body sculpting techniques and boundary constraints;

[0046] Figures 13a-13d We created a demonstration of the results for a portion of the training set, including... Figure 13a This is the inline seismic profile of igneous rock No. 16; Figure 13b Threshold segmentation label data for igneous rock No. 16; Figure 13c Seismic data profiles of igneous rocks 7 and 8 along the inline direction; Figure 13d Threshold segmentation label data for igneous rocks No. 7 and No. 8;

[0047] Figures 14a-14d For 3D visualization of earthquake data and label data, Figure 14a A 3D representation of earthquake body No. 4; Figure 14b A 3D representation of the igneous rock label for seismic body No. 4; Figure 14c A 3D representation of earthquake body No. 100; Figure 14d A 3D representation of the igneous rock label for earthquake body No. 100;

[0048] Figure 15This is an accuracy curve for the model based on the U-net network.

[0049] Figure 16 The graph shows the loss function of the model based on the U-net network.

[0050] Figure 17 This is a typical volcanic structure in the example area;

[0051] Figure 18 A 3D visualization of intrusive igneous rock identification based on U-net network. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the described embodiments of the present invention are within the scope of protection of the present invention.

[0053] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0054] In recent years, convolutional neural networks (CNNs) have achieved remarkable results in fields such as computer vision and object detection, and have been widely applied in practice. Compared with shallow neural networks, CNNs have significant advantages in feature extraction and modeling. Deep learning excels at extracting abstract feature representations from raw input data and has good generalization ability. With the significant increase in the amount of training datasets and the dramatic increase in chip processing power, it has achieved remarkable results in object detection, computer vision, semantic analysis, and other fields.

[0055] To reduce the error rate of existing geological body carving data results, this invention provides a deep learning-based automatic igneous rock identification method. This method, for the first time, proposes using actual seismic data to create labeled data, which is then used to train a deep learning convolutional neural network (U-net) model for automatic igneous rock identification. Results show that this method is robust, numerically stable, computationally efficient, and easy to apply. This method can save significant manpower, improve efficiency, and reduce costs.

[0056] Correspondingly, some other embodiments of the present invention provide an automatic identification system for igneous rocks based on deep learning.

[0057] Example 1

[0058] like Figure 1 As shown, this embodiment provides an automatic identification method for igneous rocks based on deep learning, including the following steps:

[0059] 1) Establish a training set of igneous rock label data based on actual seismic data;

[0060] 2) The pre-constructed deep learning-based 3D convolutional neural network U-net model was trained using the established igneous rock label data training set;

[0061] 3) The trained U-net model, a three-dimensional convolutional neural network based on deep learning, is used to identify the target seismic data and obtain the igneous rock identification results.

[0062] Preferably, in step 1) above, the method for establishing a labeled dataset based on actual seismic data includes the following steps:

[0063] 1.1) Based on actual seismic data and drilling data, use three-dimensional visualization function to analyze and determine the igneous facies characteristics, distribution range and development stages corresponding to the actual seismic data;

[0064] 1.2) Lithology is identified based on seismic attributes, and a refined seismic interpretation method is used as a constraint. Seed point automatic picking technology is used to refine the identified igneous facies to obtain the original igneous label dataset.

[0065] 1.3) By moving the cutting position, the original igneous rock label dataset is augmented to obtain the igneous rock label data training set.

[0066] Preferably, in step 1.1) above, the analysis using three-dimensional visualization function refers to: determining the geological background information such as the distribution and development scale and development stage of igneous rocks corresponding to the actual seismic data based on the seismic frequency, amplitude characteristics and development location patterns of each igneous facies.

[0067] Preferably, in step 1.2) above, when identifying rock masses based on seismic attributes, the root mean square amplitude attribute is used for intrusive and overflow igneous rocks, and the variance attribute is used for volcanic conduit igneous rocks.

[0068] Preferably, in step 2) above, the model training process includes the following steps:

[0069] 2.1) Normalize the training set of igneous rock label data obtained in step 1);

[0070] 2.2) The pre-constructed three-dimensional convolutional neural network U-net model based on deep learning was trained using the normalized igneous rock label data training set.

[0071] like Figure 2 As shown, in this embodiment, the U-net model based on deep learning three-dimensional convolutional neural network includes: an encoding module for extracting features from the normalized label dataset to obtain feature data; and a decoding module for restoring the learned features to the original image size.

[0072] Preferably, the encoding module includes four feature extraction layers, each of which contains two convolutional layers, one ReLU activation function, and one pooling layer. The convolutional kernel size of the convolutional layers is 3×3, and the stride of the pooling layers is 2×2. With each convolutional operation, the number of feature channels doubles, and the image size changes accordingly.

[0073] Preferably, the decoding module includes four upsampling layers. Each upsampling layer upsamples the feature map using a 2×2 deconvolution. The upsampled result is concatenated with the feature map of the encoding path at the same location. The concatenated result is then subjected to two more 3×3 convolutions, with a ReLU activation function applied after each convolution. Finally, a 1×1 convolution maps each of the 64-component feature vectors to the desired category. The original U-net network had 23 convolutional layers, while the simplified network has only 15. This achieves the desired results in both training efficiency and recognition accuracy after training.

[0074] Preferably, in step 2.1) above, normalizing the igneous rock label data training set means cutting the dataset into cubic data with 256 sampling points in the vertical direction and 256 main survey lines and connecting survey lines.

[0075] Example 2

[0076] This embodiment further illustrates the deep learning-based automatic identification method for igneous rocks proposed in this invention through a specific case.

[0077] (1) Analysis of the characteristics of igneous rock types

[0078] Igneous rocks include intrusive and extrusive rocks, characterized by complex structures, rapid lateral variations, and significant differences in physical properties among volcanic rocks of different facies zones, exhibiting distinct seismic reflection characteristics on seismic profiles. Surveys of igneous core samples drilled in the East China Sea and Bohai Sea regions indicate that Cenozoic igneous rocks are primarily composed of basalt and tuff of the effusive and explosive facies, along with intermediate extrusive andesite and volcanic breccia, as well as intrusive diabase and gabbro. Based on the modes of magmatic action and environmental differences, different scholars have proposed various classification schemes for igneous rock facies. Building upon previous classification schemes and actual survey results, and considering the developmental characteristics of igneous rocks in the study area, this invention primarily considers the following six facies: explosive facies, effusive facies, intrusive facies, volcanic conduit facies, secondary igneous facies, and eruptive sedimentary facies.

[0079] like Figure 3 As shown, through analogy with the seismic data of the example area, it was found that the example area mainly develops three typical igneous facies: overflow facies, intrusive facies, and volcanic conduit facies, and these are clearly distinguishable from the sedimentary host rocks.

[0080] The lava within the effusive facies exhibits good continuity and stable distribution. On seismic profiles, it manifests as medium-to-strong amplitude, medium-to-low frequency, continuous, parallel-to-subparallel seismic reflection structures, often with layered or lenticular shapes. Effluent facies are mostly developed at the top of volcanic conduits and distributed around the crater perimeter. Because the eruptions of effusive facies are mostly instantaneous bursts and overflows, their formation time is consistent with that of the surrounding strata.

[0081] Volcanic conduit facies often have a nearly upright, cone-shaped shape, with a relatively mixed composition of igneous rocks inside. On seismic profiles, they typically exhibit weak to medium amplitude, high frequency, and poorly continuous chaotic reflection characteristics, with a nearly upright columnar seismic reflection shape or features along fault development. The establishment of volcanic eruption models can effectively predict the distribution patterns and spatial distribution of volcanic rock facies. Based on their eruption conduits, they can be divided into three main categories:

[0082] ① Central eruption: Magma surges upward along tubular channels and overflows from the crater. This type of eruption is caused by tectonic activity that propels magma out through tubular channels, with volcanic debris splashing onto the crater rim to form a volcanic cone protruding from the surface. This is a typical characteristic of central volcanic eruptions.

[0083] ② Fissure eruption (linear eruption): such as Figure 4a and Figure 4b As shown, magma erupting to the surface through deep, large fractures and connected fractures is called a fissure eruption. Volcanic conduits appear as narrow, long lines on the surface, with a wall-like vertical characteristic. Most of the volcanic conduits in the example depression exhibit this pattern. Analysis reveals that the fissure conduits and sedimentary fault systems in this region are separate.

[0084] ③ Penetrating eruption (planar eruption): When magma rises, the high temperature causes the overlying rocks to melt and break open, forming volcanic conduits with large diameters and irregular shapes. The example area mainly develops volcanic rocks with central and fissure eruptions, and a few volcanic conduits with penetrating eruption patterns exist.

[0085] Intrusive facies are mainly located near deep faults or in the lower part of volcanic conduits, often exhibiting piercing and dish-shaped intrusive structures with strong, continuous mid-frequency reflections. Dishe-shaped intrusive rocks refer to dish-shaped structures formed when magma, during its ascent, intrudes into the surrounding rocks primarily due to its own buoyancy. For weakly deformed overlying sedimentary strata, relatively complete dish-shaped intrusive bodies are often formed, while in tectonically active basins, irregular, dish-like intrusive rocks are more likely to form, sometimes referred to as sills, cisterns, or caps. Since drilling mostly avoids sills, the development mechanism of dish-shaped intrusions is controversial. Sorenssen et al. proposed a model with a central source point, while Francis and Chevallier proposed a model of downward magma flow. The study of Area 2 suggests that the magma source of the dish-shaped sill is located at the lowest point of the sill. The higher sill can form concurrently with or later than the lower sill. In the latter case, the early-cooled magma conduit is melted and becomes an intrusive conduit again, while the higher sill cools to form the later-stage sill. The formation time of the intrusive phase can be defined as the latest time when magma penetrates the rock strata, indicating that magma continues to intrude upwards.

[0086] like Figure 5 As shown, dish-shaped intrusive rocks are well-developed in Area 2, while they are poorly developed in Area 1. Intrusive rocks often have a strong mechanical destructive effect on sedimentary surrounding strata. Dome-shaped structures are often formed above the intrusive body, and magma supply channels are often developed below or beside the intrusive body. Extrusive igneous rocks may also develop in the shallow layers, forming a complete volcanic structure or system.

[0087] Based on the above research summary, effusive facies exhibit medium-to-strong amplitude, medium-to-low frequency, and relatively continuous seismic reflection characteristics on seismic profiles. Intrusive facies mostly show piercing and dish-shaped intrusive structures with medium-frequency continuous strong reflections, while volcanic conduit facies mostly show disordered reflection characteristics with poor continuity on seismic profiles. Volcanic conduit facies differ significantly from the former two on seismic profiles, and all three are clearly distinguishable from the surrounding strata. Therefore, seismic attributes are used to describe and sculpt them. Research shows that amplitude and energy attributes are widely used to identify effusive and intrusive igneous rocks, while continuity detection attributes have a good identification effect on disordered and discontinuous volcanic conduit facies. After testing various attributes, this invention preferentially uses root-mean-square amplitude and variance attributes for targeted identification of igneous rocks. Figures 6 to 7It can be seen that the root mean square amplitude attribute is effective in identifying intrusive rocks and effusive facies, but shows no indication for volcanic conduits. Conversely, variance-type attributes ( Figure 8 It can accurately depict the chaotic and discontinuous reflections of volcanic channels, but it does not show the continuous and strongly reflective intrusive and overflow phases.

[0088] Based on the above attribute identification methods, this invention establishes a flowchart for rapid three-dimensional carving of igneous rocks. Figure 1 Starting from seismic data and drilling information, the characteristics of igneous facies are analyzed by analogy, seismic attributes are optimized and extracted, and the optimal variance attribute and root mean square amplitude attribute threshold are determined. The geological body sculpting technology of the software is used to sculpt various igneous rock bodies, and the igneous phases are determined based on the contact relationship between different igneous facies and surrounding strata.

[0089] Using the aforementioned attribute recognition methods, carving results can be obtained quickly. However, it is unavoidable that threshold segmentation using geological body carving techniques will produce many erroneous data. The main reason for this is that the continuous strong reflections at medium and low frequencies not only include overflow and intrusive igneous rock masses, but also contain many stratigraphic strong axes (…). Figure 9 The sculpting still requires meticulous trimming, adding, and combining to achieve the desired result. However, trimming can easily cut out accurately depicted geological features. The adding function can increase pixel data, but adding pixels requires clicking on existing pixels to extend them; it cannot be done arbitrarily. Selecting pixels that are too small will result in very detailed sculpting, but it will be slow and inefficient. Enlarging the pixel units will result in inaccurate depiction of the geological features and severe blocky edges. Therefore, it is recommended to use this function for small-scale geological feature sculpting. Large-scale geological feature sculpting is time-consuming and laborious, and the accuracy of the result cannot be guaranteed. Therefore, this invention, based on the above research results and combined with artificial intelligence deep learning technology, achieves automatic and accurate identification and sculpting of igneous rocks.

[0090] (2) Dataset creation

[0091] The creation of deep learning label datasets directly affects the accuracy of the final prediction results. Research indicates that current geological and geophysical datasets mainly fall into three categories: ① Using open-source data, such as the SEAM simulation dataset and the North Sea F3 dataset; these datasets contain stratigraphic, fault, fracture, and geological body interpretation data, serving as training and validation sets for identifying such geological phenomena. ② Using geophysical simulation data, including geophysical simulation platforms and computer software for forward modeling earthquake data. This method is indeed feasible, but for 3D forward modeling, ensuring both the quantity and diversity of the training set requires significant human and material resources. ③ Simplifying the operation based on the second category, directly using geological models and convolution operations can yield a large amount of training data in a short time. However, it is undeniable that such simulation data differs from actual data; the seismic response of many complex geological phenomena cannot be simply expressed using convolution simulation.

[0092] In summary, due to the lack of open-source datasets for igneous rock identification, the creation of simulated data is insufficient to fully represent the differences and diversity of seismic lithofacies such as volcanic conduit facies and intrusive facies. This invention uses actual seismic data combined with manual evaluation of relevant attributes to create a seismic label dataset for igneous rocks. Since the two wells encountered thin layers of Cenozoic igneous rocks in the southern part of the example depression, the southern part of the example depression's 3D data volume was used as the training set, and the northern part as the prediction set for identification and prediction. Figure 10 ).

[0093] The aforementioned method of geological body sculpting based on attributes can lead to the creation of numerous erroneous data points. To address this issue, this invention employs a refined seismic interpretation method for constraint. For effusive and intrusive igneous rocks, strong acoustic impedance interfaces form between the upper and lower parts of the rock mass. This invention utilizes automatic seed point picking technology to automatically interpret the boundaries of the igneous rock mass. Figure 11a and Figure 11b As shown), and using this as the boundary to constrain the carving of geological bodies, a precise carving result of igneous geological bodies can ultimately be obtained. Figure 12 This method is significantly superior to the results obtained by using only seismic attributes to depict geological bodies.

[0094] Based on the above carving results, the geological body carving results are added to the model using attribute modeling. For binary classification problems, only the identified content needs to be marked as 1 in the geological body model of attribute modeling, and other parts need to be marked as 0. If it is a multi-class classification problem that includes volcanic conduit facies, intrusive facies, and effusive facies, then the labels can be marked as 0, 1, and 2 respectively. Figures 13a-13d This is a partial display of the tag data, in which... Figure 13a This is the seismic profile in the training set, section 16, along the inline direction of the intrusive igneous phase. Figure 13b Yes Figure 13a The label data obtained after attribute extraction and threshold segmentation Figure 13c These are seismic data profiles in the inline direction from intrusive igneous phases 7 and 8 in the training set. Figure 13d Yes Figure 13c Label data obtained after attribute extraction and threshold segmentation.

[0095] Because the encoding and decoding processes of deep learning convolutional neural networks generate multi-dimensional information during training, which is difficult for computer memory to handle, and to allow the network model to learn from more sample data, this invention divides the aforementioned large-scale earthquake data and labeled training set into cubic data with 256 sampling points vertically and 256 main survey lines and connecting survey lines. This data structure facilitates computer reading and computation. After detailed interpretation and data augmentation, a total of 600 training datasets were obtained. Figures 14a-14d Two of them will be selected for 3D display.

[0096] (3) Automatic identification of igneous rocks based on U-net network

[0097] Since the U-net network was initially used to recognize complex medical images, its deep network structure resulted in long training times, affecting the actual efficiency of the method. To improve the efficiency of the network training model while ensuring the accuracy of the network recognition results, this invention, based on the problem of igneous rock mass recognition, reduces the number of network layers while maintaining prediction accuracy. After testing, a 20-layer network structure was ultimately determined to be able to accurately obtain multi-dimensional feature information of igneous rock masses in both aspects. Figure 2 The training time for the network was reduced to one-fifth of the original time.

[0098] like Figure 15 The figure shows the accuracy curve of the U-net network model. It can be seen that the final model accuracy reached over 95%. Here, `train` represents the degree of agreement between the training results and the training set labels, and `test` represents the validation set, used to verify the accuracy of the model results. After verification, the accuracy of the trained model and the label data reached over 95%. Therefore, provided the label data is completely accurate, the higher the model accuracy, the higher the accuracy of the final recognition result. Figure 16 The smaller the value of the loss function curve, the better the model convergence. It can be seen that the model loss function drops to 0.002. From the model training curve, it can be seen that the U-net network model was trained for a total of 25 epochs. When it reached 20 epochs, the model basically converged and the accuracy reached 97.23%.

[0099] The network model that ultimately converges in the iterations is selected and applied to igneous rock identification in the test data. Due to the limitations of computer GPU memory, it is not possible to input the entire 3D seismic data set for prediction. Therefore, representative seismic bodies of volcanic facies were selected and segmented for Area 2 of Example 2. Figure 17 As shown, this is a very typical central-emission volcanic structure in Example 2. Magma surges up along tubular channels, intrudes into the surrounding rocks in the lower part of the volcanic conduit facies to form intrusive igneous rocks, and erupts to the surface to form overflow igneous rocks.

[0100] This invention automatically identifies the three-dimensional seismic data and predicts based on the trained U-net network model to obtain... Figure 18 The three-dimensional prediction results are shown. The entire automatic identification process takes only tens of seconds, making this method far more efficient than traditional geological body carving processes. The accuracy of igneous rock mass identification is based on the accuracy of the labeled dataset, and the accuracy of the results can be guaranteed through validation on the validation set. This invention greatly improves work efficiency. The prediction results can be incorporated into interpretation software to assist in further igneous rock research.

[0101] Example 3

[0102] The above-described embodiment 1 provides an automatic identification method for igneous rocks based on deep learning. Correspondingly, this embodiment provides an automatic identification system for igneous rocks based on deep learning. The system provided in this embodiment can implement the automatic identification method for igneous rocks based on deep learning in embodiment 1. The system can be implemented through software, hardware, or a combination of both. For example, the system may include integrated or separate functional modules or units to execute the corresponding steps in the methods of embodiment 1. Since the system in this embodiment is basically similar to the method embodiment, the description process in this embodiment is relatively simple. Relevant details can be found in the description of embodiment 1. The system embodiment provided in this embodiment is merely illustrative.

[0103] The deep learning-based automatic identification system for igneous rocks provided in this embodiment includes:

[0104] The dataset acquisition module is used to build a training set of igneous rock label data based on actual seismic data.

[0105] The model training module is used to train a pre-built deep learning-based 3D convolutional neural network U-net model using the established igneous rock label data training set.

[0106] The automatic identification module is used to identify target seismic data using a trained deep learning-based 3D convolutional neural network U-net model to obtain igneous rock identification results.

[0107] Example 4

[0108] This embodiment provides a processing device corresponding to the deep learning-based automatic identification method for igneous rocks provided in Embodiment 1. The processing device can be a client-side processing device, such as a laptop, tablet, or desktop computer, to execute the method of Embodiment 1.

[0109] The processing device includes a processor, a memory, a communication interface, and a bus. The processor, memory, and communication interface are connected via the bus to enable communication between them. The memory stores a computer program that can run on the processor. When the processor runs the computer program, it executes the deep learning-based automatic identification method for igneous rocks provided in Embodiment 1.

[0110] In some embodiments, the memory may be high-speed random access memory (RAM), and may also include non-volatile memory, such as at least one disk storage device.

[0111] In other embodiments, the processor can be a general-purpose processor of various types, such as a central processing unit (CPU) or a digital signal processor (DSP), and is not limited thereto.

[0112] Example 5

[0113] The deep learning-based automatic identification method for igneous rocks in Embodiment 1 can be specifically implemented as a computer program product. The computer program product may include a computer-readable storage medium on which computer-readable program instructions for executing the deep learning-based automatic identification method for igneous rocks described in Embodiment 1 are loaded.

[0114] A computer-readable storage medium can be a tangible device that holds and stores instructions for use by an instruction execution device. A computer-readable storage medium can be, for example, but not limited to, an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any combination thereof.

[0115] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method for automatic identification of igneous rocks based on deep learning, characterized in that, Includes the following steps: Acquire target earthquake data; The target seismic data was identified using a pre-established deep learning-based three-dimensional convolutional neural network U-net model to obtain igneous rock identification results; the pre-established deep learning-based three-dimensional convolutional neural network U-net model was trained using a training set of igneous rock label data obtained from the processing of actual seismic data. The method for obtaining a training set of igneous rock label data from actual seismic data processing includes the following steps: Based on actual seismic data and drilling data, the igneous facies characteristics, distribution range, and development stages corresponding to the actual seismic data are determined by using three-dimensional visualization functions. Lithology is identified based on seismic attributes, and a refined seismic interpretation method is used as a constraint. Seed point automatic picking technology is used to carve the identified igneous facies to obtain the original igneous label dataset. By moving the cutting positions, the original igneous rock label dataset is augmented to obtain the igneous rock label training set.

2. The method for automatic identification of igneous rocks based on deep learning as described in claim 1, characterized in that, The aforementioned analysis using three-dimensional visualization refers to determining the distribution and development scale and development stages of igneous rocks corresponding to actual seismic data based on the seismic frequency, amplitude characteristics, and development location patterns of each igneous facies.

3. The method for automatic identification of igneous rocks based on deep learning as described in claim 1, characterized in that, When identifying lithology based on seismic attributes, the root mean square amplitude attribute is used for intrusive and effusive igneous rocks, while the variance attribute is used for volcanic conduit igneous rocks.

4. The method for automatic identification of igneous rocks based on deep learning as described in claim 1, characterized in that, The training of the U-net model, a three-dimensional convolutional neural network based on deep learning, using a training set of igneous rock label data obtained from actual seismic data processing, includes: Normalize the training set of igneous rock label data; The pre-constructed deep learning-based 3D convolutional neural network U-net model was trained using a normalized igneous rock label data training set.

5. The method for automatic identification of igneous rocks based on deep learning as described in claim 4, characterized in that, The normalization process for the igneous rock label data training set refers to dividing the dataset into cubic data with 256 sampling points vertically and 256 main survey lines and connecting survey lines.

6. The method for automatic identification of igneous rocks based on deep learning as described in claim 4, characterized in that, The deep learning-based 3D convolutional neural network U-net model includes: The encoding module is used to extract features from the normalized igneous rock label data training set to obtain feature data. The decoding module is used to restore the learned features to the original image size and obtain the recognition result.

7. An automatic identification system for igneous rocks based on deep learning, characterized in that, include: The dataset acquisition module is used to acquire target seismic data; An automatic identification module is used to identify target seismic data using a pre-established deep learning-based three-dimensional convolutional neural network U-net model to obtain igneous rock identification results; wherein, the pre-established deep learning-based three-dimensional convolutional neural network U-net model is trained using a training set of igneous rock label data obtained from the processing of actual seismic data; The method for obtaining a training set of igneous rock label data from actual seismic data processing includes: Based on actual seismic data and drilling data, the igneous facies characteristics, distribution range, and development stages corresponding to the actual seismic data are determined by using three-dimensional visualization functions. Lithology is identified based on seismic attributes, and a refined seismic interpretation method is used as a constraint. Seed point automatic picking technology is used to carve the identified igneous facies to obtain the original igneous label dataset. By moving the cutting positions, the original igneous rock label dataset is augmented to obtain the igneous rock label training set.

8. A computer-readable storage medium for storing one or more programs, characterized in that, The one or more programs include instructions that, when executed by a computing device, cause the computing device to perform any of the methods described in claims 1 to 6.

9. A computing device, characterized in that, include: One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs including instructions for performing any of the methods described in claims 1 to 6.

Images