Intelligent identification method, system and equipment for core deposition structure and medium
Through the intelligent identification method of core sedimentary structure based on the deep learning cloud platform, the problem of backward intelligent identification technology of core photos and field outcropping photos is solved, and efficient and accurate sedimentary structure recognition is achieved, which is suitable for the intelligent development of the petroleum exploration and development industry. .
Patent Information
- Application Number
- CN202310513220.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-08
- Publication Date
- 2024-11-08
AI Technical Summary
In the prior art, the intelligent identification technology of sedimentary structure images of core photos and field outcrop photos is still in its infancy, resulting in the failure of core data to be fully utilized, and the artificial recognition efficiency is low and there is personal experience deviation.
The intelligent identification method of core sedimentary structures based on the deep learning cloud platform is adopted, and the image classification algorithm is used to train the model to quickly identify images containing obvious sedimentary structures, improve the recognition efficiency and accuracy, and adapt to the intelligent development of the petroleum exploration and development industry.
It realizes intelligent identification of massive core photos, improves the accuracy and efficiency of sedimentary structure recognition, overcomes the inefficiency and deviation of manual identification, and is especially suitable for people with insufficient professional experience and non-professional personnel, and improves oil exploration and development The level of development of the industry.
Smart Images

Figure CN118918351A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of geological sedimentary phase research, and specifically to an artificial intelligence recognition method, system, electronic device and storage medium for sedimentary structures in core images. Background Art
[0002] Sedimentary facies research is a basic work in the petroleum industry. The important basis for determining sedimentary facies in petroleum geology includes drilling data, logging data, seismic data and field outcrops. However, only the core of the coring well is the most direct first-hand data. By observing the sedimentary structure on the core, the sedimentary environment and hydrodynamic conditions can be judged, and then the sedimentary facies can be judged. Sedimentary structure is one of the most common macroscopic characteristics of sedimentary rocks, and is the spatial distribution and arrangement of each component of sedimentary rocks. It is formed by physical, chemical, biological and other effects on the surface or inside of sediments during or after deposition. Among them, the structures formed during the sedimentation period are called primary structures, such as bedding, ripples and other flow-induced structures; structures formed after deposition, some are formed before sedimentation and consolidation, such as load structures, and some are formed after sedimentation and consolidation, such as sutures, stacked cones and other chemical-induced structures. In summary, sedimentary structures are mainly mechanical, chemical and biological. There are more than 50 types of sedimentary structures in sedimentary rocks, and there are more than a dozen common ones.
[0003] For a long time, the judgment of sedimentary structures on core photos mainly relied on the experience of geological experts. They observed the cores with the naked eye and took photos to describe them when they found cores with obvious sedimentary structures. In the past decade, with the accelerated advancement of digital oilfield construction, digitally scanned core images have gradually been shared, and "data islands" have been connected by the network. Originally, researchers could only see the cores of a few or dozens of wells, but now they can easily obtain core data of up to thousands of wells in the entire oil field or basin, or even a larger area. Traditional expert identification methods are difficult to normalize and standardize due to different personal experiences, and are extremely time-consuming. Therefore, there is a situation where precious core data is not fully utilized.
[0004] With the rapid development of artificial intelligence in recent years, image recognition based on computer vision has become a research hotspot in various fields. The main characteristics of artificial intelligence application research on rock images in the geological field are as follows: ① In the early stage of machine learning development, rock image recognition research mainly focused on the improvement of image feature extraction algorithms and classifier selection; ② In the deep learning model application stage, the application of deep learning models solves the cumbersome experimental process of traditional machine learning algorithms, realizes feature extraction automation, and can learn more types and quantities of image features. Transfer learning (TL) and convolutional neural networks (CNN) are used to identify lithofacies from core scanning images and their descriptive data, which has the advantages of high efficiency and objectivity. For simple classification models, the accuracy of lithofacies classification can reach 85-95%. Deep learning has been successfully applied to well logging interpretation for lithofacies identification, sedimentary facies prediction, and reservoir physical property interpretation of well logging curves.
[0005] However, in terms of the combination of rock image recognition and artificial intelligence, most of the deep learning research results are still focused on microscopic thin sections. For rock lithology, mineral composition, particle size, and porosity analysis, there is already a high recognition accuracy. The recognition of microscopic paleontology is relatively mature. The intelligent recognition of sedimentary structure images of core photos and field outcrop photos is still in the initial exploratory stage. There are few related research results at home and abroad, and there are no patent results in China. This patent takes core images as the research object and explores the implementation of intelligent sedimentary structure recognition methods to meet the needs of real-time sedimentary structure recognition and sedimentary phase research in the digital age. Summary of the invention
[0006] In view of this, in order to overcome the inefficiency of manual identification of sedimentary structures and the deviation caused by different personal experiences, combined with the significant technological progress of deep learning in image intelligent recognition in recent years, the present invention proposes a method, system, electronic device and storage medium for intelligent identification of core sedimentary structures based on a deep learning cloud platform.
[0007] The present invention can utilize an open deep learning cloud platform to train an artificial intelligence sedimentary structure recognition model based on a deep learning image classification algorithm, quickly and intelligently identify images containing obvious sedimentary structures, and a series of adjacent sedimentary structures with genetic connections point to specific sedimentary microfacies. The core judgment results have high credibility, and can confirm or correct the logging sedimentary microfacies recognition results, providing first-hand direct evidence for the determination of sedimentary microfacies. Intelligent identification of sedimentary structures is particularly helpful for people with insufficient professional experience and non-professionals to understand sedimentary phenomena. At the same time, when professionals are faced with a large number of core images, the efficiency of the recognition process is improved, which has positive significance for the oil exploration and development industry to adapt to the development of the intelligent era.
[0008] In a first aspect, a method for intelligently identifying core sedimentary structures comprises:
[0009] Select the sedimentary structure type and obtain the original image of the core;
[0010] Screening and processing the original core image to obtain a core image;
[0011] Classifying and labeling the core image according to sedimentary structure type;
[0012] Performing image enhancement and expansion on the core image to obtain a data set;
[0013] Build models based on deep learning image classification algorithms;
[0014] Conduct model training result analysis and model testing;
[0015] If the model fails the model test, the data set is optimized and the model is iteratively updated;
[0016] After the model passes the model test, it is published and applied online to intelligently identify the core sedimentary structure.
[0017] In a possible implementation, the sedimentary structure types include: grain-sequenced bedding, parallel bedding, wavy bedding, lens bedding and charcoal foliation.
[0018] In a possible implementation, screening and processing the original core image specifically includes:
[0019] Remove images that are too blurred, have unclear sedimentary structures, or have too much interference information, and select the original core images with obvious sedimentary structures;
[0020] Identify the sedimentary structure types of the screened core raw images;
[0021] Delete the interference information of the filtered core original image;
[0022] The length, width and size of the image are processed to obtain a single sedimentary structure image.
[0023] In a possible implementation, the image enhancement and expansion includes: adjusting image brightness, adjusting image color balance, adjusting image clarity, randomly blocking a portion of the image, flipping the image left and right, rotating the image, and adjusting the color of the image.
[0024] In one possible implementation, the model training result analysis and model testing are performed as follows:
[0025] Importing the data set into the model, randomly selecting a portion of images in the data set to form a training set, and the remaining images to form a test set, to perform the first model training;
[0026] When the sample quality is high, the classification feature discrimination is reasonable, and the model accuracy reaches the target range, the data set is randomly divided into a training set and a test set, and dozens of trainings are performed to obtain the model accuracy;
[0027] For any type of sedimentary structure, the F1-score index is further used to evaluate the model classification results to obtain the F1-score index of the model;
[0028] When both the model accuracy and F1-score indicators are qualified, the model is trained using a data set that has never participated in the model training set and test set. If the model accuracy reaches more than 85%, the model passes the test.
[0029] In a possible implementation, the data set is optimized and the model is iteratively updated, specifically:
[0030] If the model accuracy is not up to standard, check the F1-score index of each sedimentary structure type and determine the sedimentary structure type with the lowest score;
[0031] The datasets of the sedimentary structural types with the lowest scores were optimized by removing interfering images and supplementing them with more images with obvious features;
[0032] Use the optimized dataset to iteratively update the model.
[0033] In one possible implementation, the model can intelligently identify core sedimentary structures through a mobile phone.
[0034] Second, the core sedimentary structure intelligent identification device includes:
[0035] An image acquisition module is used to select the sedimentary structure type and obtain the original image of the core;
[0036] An image processing module is used to screen and process the original core image to obtain a core image;
[0037] An image annotation module, used to classify and annotate the core image according to sedimentary structure type;
[0038] An image expansion module, used to perform image enhancement and expansion on the core image to obtain a data set;
[0039] Model building module, used to build models based on deep learning image classification algorithms;
[0040] Model testing module, used for model training result analysis and model testing;
[0041] A model updating module, used for optimizing the data set and iteratively updating the model when the model fails the model test;
[0042] The model publishing module is used for model publishing and online application, and for intelligent identification of core sedimentary structures.
[0043] According to a third aspect, an electronic device includes:
[0044] processor;
[0045] Memory;
[0046] And a computer program, wherein the computer program is stored in the memory, and the computer program includes instructions, and when the instructions are executed by the processor, the electronic device executes any one of the methods described in the first aspect.
[0047] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute any one of the methods described in the first aspect.
[0048] Through the collection, processing and optimization process methods of the sedimentary structure image classification data set of the present invention, after improving and expanding the training set, the sedimentary structure recognition model can intelligently recognize a large number of core photos and scanned core images, quickly find sedimentary structures with sedimentary phase environment judgment marks, and select corresponding images. It overcomes the inefficiency of manual identification of sedimentary structures and the deviation caused by different personal experiences, not only improves the accuracy of sedimentary structure identification, but also improves the efficiency of the identification process, which is of great significance to the development of the oil exploration and development industry. The mobile phone test of the embodiment of this patent has an identification accuracy of 90% for five types of sedimentary structures. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0050] Figure 1 A schematic diagram of a process flow of a core sedimentary structure intelligent identification method provided in an embodiment of the present application;
[0051] Figure 2 A schematic diagram of a process for processing an original core image provided in an embodiment of the present application;
[0052] Figure 3A schematic diagram of image enhancement provided in an embodiment of the present application;
[0053] Figure 4 A pie chart showing the proportions of five types of core sedimentary structure images in the core images provided in the embodiment of the present application;
[0054] Figure 5 Typical sedimentary structure image examples and image quality statistics reports for training provided in the embodiments of the present application;
[0055] Figure 6 A schematic diagram of a process for building a model based on a deep learning image classification algorithm provided in an embodiment of the present application;
[0056] Figure 7 A schematic diagram of a confusion matrix for the ninth version of model training provided in an embodiment of the present application;
[0057] Figure 8 A schematic diagram of ten cumulative confusion matrices of partial classifications of the original model and the optimized model provided in the embodiment of the present application;
[0058] Fig. 9 The identification results and thermal diagram of the parallel bedding-wavy bedding pairs in the original model provided in the embodiment of the present application;
[0059] Fig.10 Typical core images and thermal images of three factors affecting model accuracy provided in the embodiments of the present application and their test results;
[0060] Fig.11 The mobile experience app interface and sedimentary structure picture intelligent recognition results provided in the embodiment of this application;
[0061] Fig.12 A structural block diagram of a core sedimentary structure intelligent identification method device provided in an embodiment of the present application;
[0062] Fig.13 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0063] In order to better understand the technical solution of the present application, the embodiments of the present application are described in detail below with reference to the accompanying drawings.
[0064] It should be clear that the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.
[0065] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.
[0066] It should be understood that the term "and / or" used in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.
[0067] See also Figure 1 , is a flow chart of a method for intelligent identification of core sedimentary structures provided in an embodiment of the present application. Figure 1 As shown, it mainly includes the following steps.
[0068] Step S1, select the sedimentary structure type and obtain the original image of the core.
[0069] First, determine which are the common sedimentary structures. Through a large amount of literature research, summarize the types of sedimentary structures, and give priority to common sedimentary structures as intelligent recognition objects, laying a good foundation for the collection of core images and model training. Sedimentary structure is one of the most common macroscopic features in sedimentary rocks, which is the spatial distribution and arrangement of the various components of sedimentary rocks. Table 1 summarizes the three types of sedimentary structures of origin, of which the most common are primary physical structures, of which there are about ten. Primary sedimentary structures can maintain their basic characteristics after the sedimentation period, and their observation and study do not require complex instruments. They are mainly carried out through outcrop research and core observation. Therefore, primary sedimentary structures and bedding sequences can be used as effective tools for environmental interpretation.
[0070] Table 1: Sedimentary structural types of three origins
[0071]
[0072]
[0073] In this embodiment, the core image data mainly comes from 38 coring wells in Dongying Depression of Jiyang Depression in China, where the oil field is a delta-fan delta sedimentary environment of the Paleogene, and the rest are 1 long coring well in Songliao Basin, 8 wells in Oriente Basin in South America and 12 wells in Kenya, Africa. The coring length varies from 10 to 300 meters. One part is the photos taken manually during the on-site core observation; the other part is the core scanning image, including the expanded image of the outer surface annular scan, the cylindrical core flat scan image and the half-sheet core section scan image.
[0074] Preferably, the deep learning method for image intelligent recognition training relies on a large amount of image data, and the number of samples in each category should be more than 100.
[0075] Step S2, screening and processing the original core image to obtain a core image. Specifically comprising:
[0076] Remove images that are too blurred, have unclear sedimentary structures, or have too much interference information, and select the original core images with obvious sedimentary structures;
[0077] Identify the sedimentary structure types of the screened core raw images;
[0078] Delete the background, labels, scale and other interfering information of the filtered core original image;
[0079] Process the length, width and size of the image.
[0080] Furthermore, on intelligent cloud platforms such as Baidu, Tencent, and Huawei, the requirements for the length, width, and size of a single image are: the size must be less than 4M, and the length-to-width ratio must not exceed 3 times.
[0081] In this embodiment, in order to meet the software's requirements for image length, width and size, the long scanned image is first automatically segmented into 2035 images according to the actual length of the core, 15 cm per segment. Together with the 1550 photographs, 3585 images are formed. The image size ranges from 320*300 to 1024*1550 pixels, and the image size ranges from 90kb to 1320kb. Among all the core images, due to the broken core, the severely uneven surface, or large sections of blocky sandstone or shale, most of the images cannot find sedimentary structures, and 792 original images are manually screened out. Figure 2 The flowchart for processing the original core image provided in the embodiment of the present application first screens out the original core photos with obvious sedimentary structures by naked eyes, then deletes the interference information such as background, labels, scale, etc., and finally crops a single sedimentary structure image and makes corresponding annotations.
[0082] Step S3: classify and label the core image according to the sedimentary structure type. Specifically:
[0083] The core images are classified and labeled with sedimentary structure types and stored in different folders.
[0084] Furthermore, the sedimentary structure types include: grain sequence bedding, parallel bedding, wavy bedding, lens bedding and charcoal foliation.
[0085] In this embodiment, according to the characteristics of the fan delta front phase mainly developed in the study area, five types of sedimentary structures commonly used in actual work are selected to establish a label database. They include grain sequence bedding, parallel bedding, wavy bedding, lens bedding and charcoal foliation. The first four types of bedding sedimentary structures reflect the process from strong to weak sedimentary water dynamics and from near to far from the source. The characteristics of grain sequence bedding are that the bottom scouring or grain size mutation can be seen in the core, and conglomerate is common. The grain size gradually becomes finer from bottom to top, reflecting the deposition of fan delta channels, and the hydrodynamic environment is the strongest; parallel bedding is composed of a group of parallel sandstone or silt laminae within the layer system, reflecting a relatively stable distributary channel flow; wavy bedding is a group of wavy laminae of fine silt and mudstone, mainly developed in the estuary bar microfacies at the front of the fan delta, an environment with relatively stable directional flow velocity; lenticular bedding is mainly mud, with sand or silt sandwiched in the form of streamers or lenses, formed in an environment where the water flow is turbid and the flow velocity is intermittently reduced or increased, and mainly developed in the far sand bar microfacies. Charcoal foliation is a sedimentary structure at the core level, with black carbonized plant debris visible. It is mainly developed on the floodplain from the end of the fan delta plain to the front edge, and is a landmark sedimentary structure that determines the sedimentary shoreline.
[0086] Step S4, performing image enhancement and expansion on the core image to meet the actual conditions of different scenes and obtain a data set.
[0087] For convolutional neural network models, the training effect is closely related to the number of sample data and the actual scene. Considering that the number of core images used for training is usually limited, and the actual core photography conditions vary, there may be problems such as different light intensity in the core library, different shooting distances or angles, and a single image environment, which may lead to problems such as weak model generalization ability. Therefore, data enhancement is used to increase the amount of sample data for training.
[0088] Image enhancement mainly uses digital image processing to perform geometric transformations on images, such as increasing the amount of relevant data in the data set by image flipping, random cropping, scale transformation, color brightness adjustment, and adding noise. Adjusting image brightness can better identify images with different brightness; adjusting image color balance can better identify images based on contours; adjusting image clarity can better identify images with different clarity; randomly blocking part of the image can better identify images with random occlusions; flipping the image left and right can better identify images that have been flipped left and right; rotating the image can better identify images that have been rotated; adjusting the color of the image can better identify images based on contours rather than colors. Through image enhancement, the number of core images of each type of sedimentary structure can be increased by more than 3 times, which is an effective way to expand samples, especially for small sample sedimentary structure types.
[0089] In this embodiment, the amount of data for grain sequence bedding and lens bedding is relatively small. In this study, the images of these two types of sedimentary structures were mainly flipped left and right, and expanded by image rotation, scale transformation, and color brightness adjustment. After enhancement and expansion, the total number of images is 1298, and the number of five types of core sedimentary structure images is 200-300, which is a balanced ratio. Figure 3 A schematic diagram of image enhancement provided in an embodiment of the present application; Figure 4 A pie chart showing the proportions of five types of core sedimentary structure images in the core images provided in the embodiment of the present application; Figure 5 Typical sedimentary structure image examples and image quality statistics reports for training provided in the embodiments of the present application, where A is a typical sedimentary structure image example (ac represents charcoal foliation; df represents wavy bedding; gi represents parallel bedding), and B is an image quality statistics report, which includes image size, resolution, aspect ratio, color deviation, etc. The model training platform is highly tolerant of input images and does not require uniform size, resolution and aspect ratio, which provides great convenience for the actual application of the model.
[0090] Step S5, building a model based on a deep learning image classification algorithm.
[0091] The model can be constructed based on the deep learning image classification algorithm of the open cloud platform. Deep convolutional neural networks have special advantages in image processing. In recent years, major Internet companies have developed very mature open source deep learning platforms, providing powerful computing support and personalized technical services. It is convenient for experts in various industries to build their own industry application models with almost zero code. Baidu and Tencent artificial intelligence cloud platforms have achieved 85-95% accuracy in intelligent identification of thin-section lithofacies under microscopes based on transfer learning and convolutional neural networks.
[0092] The core images obtained from the coring wells contain most of the characteristics of the sedimentary structures, including the structure and arrangement information of the sediments representing the formation process and environment of the sedimentary rocks. It is feasible to use TL (migration algorithm) and CNN (convolutional neural network algorithm) to intelligently identify the sedimentary structures of the core images.
[0093] This embodiment relies on the deep convolutional neural network training industry application model built into Baidu Open Platform. Figure 6A schematic diagram of the flow of building a model based on a deep learning image classification algorithm, TL (transfer learning) and CNN (convolutional neural network algorithm) provided in an embodiment of the present application. First, the core image is processed and labeled with classification labels to generate a training set. The convolutional neural network pre-trained model used as a feature extractor for image classification is trained from millions of classified graphic images, and then the TL algorithm is applied to the pre-trained model for target recognition. Finally, the CNN model for intelligent recognition of sedimentary structures in core images is trained to obtain ideal recognition results.
[0094] Step S6: perform model training result analysis and model testing.
[0095] Create a deep learning image classification model on the cloud platform. Import the organized data set. Randomly select 70% of the images in the data set to form the training set, and the remaining 30% to form the test set for the first model training. When the number of image samples is less than 2,000, the training time only takes 0.2 to 1 hour. When the sample quality is high, the classification feature discrimination is reasonable, and the model accuracy (the accuracy of the model refers to the ratio of the number of correctly classified samples in the test set to the total number of samples, the top1 accuracy refers to the ratio of the number of samples with the highest probability of being the correct result to the total number of samples when the model gives multiple classification results, and similarly, the top2 accuracy refers to the ratio of the number of samples with one correct result in the top 2 results to the total number of samples, and the top1 accuracy is often used to represent it in practice) reaches the target range, the same total data set is randomly divided into the training set and the test set, and dozens of trainings are performed. If the accuracy is above 80% and the distribution range is less than 10%, it means that the model has a relatively satisfactory accuracy and robustness. For the quality evaluation of the classification results of a certain sedimentary structure type, the F1-score indicator is used. For a certain category, F1-score is the harmonic mean of precision and recall. When the model's top1 accuracy and F1-score indicators are satisfactory, the model is tested with a dataset that has never been used in model training and testing. If the test accuracy reaches more than 85%, the model preliminarily passes the test.
[0096] In this embodiment, a deep learning image classification model named "Deposition Truth" is created on the cloud platform. 70% of the images in the data set are randomly selected to form the training set, and the remaining 30% are used to form the test set. The composition of the core images of the training set and the test set randomly divided by the model is shown in Table 2. The training group data is used for model training, and the test set data is used for verification. The first model training is performed. The first training result obtained the top1 accuracy of the original model of 89.82% (the accuracy of the model refers to the ratio of the number of correctly classified samples in the test set to the total number of samples. The top1 accuracy refers to the ratio of the number of samples with the highest probability that is exactly the correct result to the total number of samples when the model gives multiple classification results. Similarly, the top2 accuracy refers to the ratio of the number of samples with one correct result in the top 2 results to the total number of samples).
[0097] Table 2: The composition of core images in the training set and test set randomly divided by a model
[0098]
[0099] In this embodiment, 10 versions of results were obtained through 10 rounds of model training and verification of random training sets and test sets. The top1 accuracy of the original model is 88.00% to 92.73%, with an average of 90.01%. The top1 accuracy of each version of the original model and the optimized model is shown in Table 3. The top2 accuracy of the model is 93.37% to 98.54%. The average accuracy of the model's 30 verification results reached 90%, with a variation range of less than 6%, showing a high accuracy and good robustness.
[0100] For the quality evaluation of the classification results of a certain sedimentary structure type, the scoring indicator F1-score is used. For a certain category, F1-score is the harmonic mean of precision and recall. Precision refers to the ratio of "the number of samples of a certain category correctly predicted as this category" to "the total number of samples predicted as this category". Recall refers to the ratio of "the number of samples of a certain category correctly predicted as this category" to "the total number of samples labeled as this category". The overall evaluation of model quality can also use the F1-score indicator, which refers to the average value of the indicators of each category. In this example, the F1-score of a certain category in the ninth version of model training is 76% to 100% (parallel bedding 76%, charcoal foliation 84%, wavy bedding 87%, grain sequence bedding 98%, lenticular bedding 100%). Figure 7 This is a schematic diagram of the confusion matrix of the ninth version of the model training provided in the embodiment of the present application. In the test set, 14 parallel beddings were misidentified as wavy beddings, and 7 were misidentified as charcoal foliation. The overall F1-score of the model is 89.5%. Figure 8The ten cumulative confusion matrices of the partial classification of the original model and the optimized model provided in the embodiment of the present application, wherein Figure A is the original model and Figure B is the optimized model. Figure 8 As can be seen in Figure A, the original model has the lowest score for parallel bedding, which is most easily confused with wavy bedding. A total of 139 images were misidentified as wavy bedding in 10 simulations. Therefore, it is necessary to analyze the reasons for image misidentification and optimize the training set image data set.
[0101] Table 3: Top 1 accuracy of original model and optimized model versions
[0102]
[0103] Step S7: If the model fails the model test, the data set is optimized and the model is iteratively updated.
[0104] If the overall accuracy of the model is not up to standard during training and testing, it may be that the prediction result of a certain type of sedimentary structure is very poor. Check the F1-score indicators of various sedimentary structure types to determine which type of sedimentary structure has the lowest score. Use the confusion matrix (the confusion matrix of the ninth version of the original model is as follows Figure 7 As shown in the figure, the confusion between categories can be intuitively displayed, and images that are easily confused can be displayed to analyze the causes of confusion. Then, the data set of this category can be optimized, interfering images can be deleted in a targeted manner, and more images with obvious features can be added. Then, the optimized data set is used to iteratively update the model. After the model optimization training is iterated many times and the accuracy is improved to a certain extent, incremental training can be used to select the trained version of the model. When the data application scenario remains unchanged, the data can be expanded for incremental training to improve the efficiency and accuracy of model training.
[0105] Fig. 9 The recognition results and heat map of the parallel bedding-wavy bedding pair in the original model provided in the embodiment of this application. The highlighted area of the heat map is the main reference area for the model to obtain the recognition result. By analyzing the overall characteristics of the correctly identified and incorrectly identified parallel bedding samples, it is found that the three main image features that affect the recognition accuracy are: Fig.10 As shown in the figure: First, the imaging angle is non-orthogonal. When shooting from the bottom of the cylindrical core image, not all pixels in the image are orthogonal to the lens. Different depths of field make some straight line features in reality appear as curves in the two-dimensional image (such as Fig.10 The human brain can easily judge this trend and correct this distortion through global information, but the machine may need more training to eliminate the interference of depth of field. Secondly, the core surface is marked, including labels or scale marks on the core surface or regular marks left by the drill bit on the core surface during coring (such as Fig.10 As shown in Figure B, since both the traces and the sedimentary structures themselves have regular features, when identifying sedimentary structures, the machine is not sure which regular feature to use to make the final judgment, which leads to confusion. Finally, the rock surface is uneven. The uneven surface of the core in the two-dimensional image is shown as mottled shadows (such as Fig.10 (As shown in Figure C), the main reference feature for the machine to make judgments is the shadow, ignoring the parallel bedding features composed of light-colored clay minerals.
[0106] In this embodiment, in order to further improve the recognition accuracy of the model, the initial approach is to remove all 135 images in the original data set that have been misidentified in the 10 versions of the original model to form an optimized data set, and randomly divide the training set and the validation set to obtain 10 versions. The same model algorithm is used for training again, and the recognition results of the optimized model show that the top1 accuracy is 89.36% to 92.34%, with an average of 91.11% (see Table 3). The arithmetic and the blind test results of the 10 versions of the validation set form the confusion matrix of the optimized model validation set (see Table 3). Figure 8 (As shown in Figure B). Compared with the original model, the recognition accuracy (91.11%) and F1-score (88.97%) increased slightly by more than 1%. Since the parallel bedding optimization eliminated the most interference images, the F1-score of the parallel bedding increased by 4 percentage points.
[0107] Reasonable training set optimization strategy. Through the performance comparison between the original model and the optimized model (the comparison table is shown in Table 4), it can be seen that the optimized model is better in accuracy, recall, precision, and F1-score. However, the improvement in the accuracy of the optimized model is much lower than the proportion of easily misidentified images eliminated in the training set, that is, eliminating 13% of easily misidentified training images only brings an average accuracy improvement of 1.1%. In other words, the generalization ability of the optimized model has decreased, and some core images correctly identified in the original model have been misidentified in the optimized model. For example, in a parallel bedding image annotated in a verification set, the parallel bedding confidence given by the original model is 87.23%, while that of the optimized model is only 0.14%. As a result, the optimized model misidentifies the parallel bedding image as wavy bedding. Although the easily misidentified images of the original model have some interference information, such as non-orthogonal shooting angles, marked core surfaces, uneven rock surfaces, etc., they can improve the anti-interference ability of the model. It is the elimination of easily misidentified images that leads to a decrease in the generalization ability of the model. In order to make the model better applicable to production practice, it is not necessary to deliberately avoid the common core image interference information in actual work when constructing the training set. Only in this way can the future sedimentary structure intelligent recognition model better serve the production work of the oil field.
[0108] Table 4: Performance comparison of the original model and the optimized model
[0109]
[0110] Step S8: After the model passes the model test, the model is published and applied online to intelligently identify the core sedimentary structure.
[0111] After the model passes the test, it can be applied for release. It can be deployed on the public cloud or on a local server as needed. After being released to the public cloud, the intelligent recognition of sedimentary structure images can be performed online on a computer or mobile device. Single image recognition can be performed, or batch intelligent recognition can be performed.
[0112] In this embodiment, after the model passes the test, it is applied to be published on the public cloud. After the public cloud deployment is completed, the link provided by the cloud server can be called to perform single and batch image recognition. The cloud platform also provides a simple mobile experience app (attached Fig.10 ), and then perform intelligent recognition of sedimentary structure images online on the mobile phone. Fig.11 The mobile experience app interface and sedimentary structure picture intelligent recognition results provided in the embodiment of this application are as follows: Fig.11 The recognition results of three images are shown in the figure. The more characteristic lens bedding and parallel bedding are accurately identified with a probability close to 100%. The sedimentary structure of the third image contains both wavy bedding and parallel bedding. The recognition result is wavy bedding with a probability of 75%, which is also accurate.
[0113] By testing the model on outcrop photos in the field, relatively obvious sedimentary structures can also be accurately identified. Therefore, the application scope of the model is not limited to core-scale images, but also has a certain help for relevant petroleum geology experts and college students in field geological surveys, and young students in geological travel and learning. However, due to the lack of training images and the extremely small number of tests, the current accuracy cannot be used as usable evidence.
[0114] Finally, it should be noted that the above-described embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solution of the present invention, rather than to limit it, and the protection scope of the present invention is not limited to this. The Baidu Open Intelligent Cloud Platform used in the embodiment is only to show the technical process of the present invention. Researchers in this field can use other cloud platforms to follow the steps of this embodiment to obtain a new sedimentary structure intelligent recognition model. The current data set of the embodiment provides 5 types of samples, and for this reason, the model also provides a default option where the recognition result is outside the five types of sedimentary structures. Researchers can also annotate more types of sedimentary structure images under the guidance of this patent method, make a larger training set, and improve the practicality of the model, but they should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.
[0115] Corresponding to the above-mentioned embodiments, the present application also provides a method and device for intelligently identifying core sedimentary structures.
[0116] See also Fig.12 , is a structural block diagram of a core sedimentary structure intelligent identification method device provided in an embodiment of the present application. Fig.12 As shown, it mainly includes the following modules.
[0117] The image acquisition module 1201 is used to select the sedimentary structure type and obtain the original image of the core;
[0118] An image processing module 1202 is used to screen and process the original core image to obtain a core image;
[0119] An image annotation module 1203 is used to classify and annotate the core image according to the sedimentary structure type;
[0120] An image expansion module 1204 is used to perform image enhancement and expansion on the core image to obtain a data set;
[0121] A model building module 1205 is used to build a model based on a deep learning image classification algorithm;
[0122] Model testing module 1206, used for model training result analysis and model testing;
[0123] A model updating module 1207 is used to optimize the data set and iteratively update the model when the model fails the model test;
[0124] The model publishing module 1208 is used for model publishing and online application to intelligently identify the core sedimentary structure.
[0125] It should be pointed out that the specific contents involved in the embodiments of the present application can be found in the description of the above method embodiments, and for the sake of brevity, they will not be repeated here.
[0126] Corresponding to the above embodiment, an embodiment of the present application further provides an electronic device.
[0127] See also Fig.13 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Fig.13 As shown, the electronic device 1300 may include: a processor 1301, a memory 1302 and a communication unit 1303. These components communicate through one or more buses. Those skilled in the art can understand that the electronic device structure shown in the figure does not constitute a limitation on the embodiments of the present application. It can be a bus structure or a star structure, and can also include more or fewer components than shown in the figure, or combine certain components, or arrange components differently.
[0128] The communication unit 1303 is used to establish a communication channel so that the electronic device can communicate with other devices.
[0129] The processor 1301 is the control center of the electronic device. It uses various interfaces and lines to connect various parts of the entire electronic device. It runs or executes software programs and / or modules stored in the memory 1302, and calls data stored in the memory to perform various functions of the electronic device and / or process data. The processor can be composed of an integrated circuit (IC), for example, it can be composed of a single packaged IC, or it can be composed of multiple packaged ICs with the same or different functions. For example, the processor 1301 can include only a central processing unit (CPU). In the embodiment of the present application, the CPU can be a single computing core or multiple computing cores.
[0130] Memory 1302 is used to store execution instructions of processor 1301. Memory 1302 can be implemented by any type of volatile or non-volatile storage device or a combination of them, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.
[0131] When the execution instructions in the memory 1302 are executed by the processor 1301 , the electronic device 1300 is enabled to execute part or all of the steps in the above method embodiments.
[0132] Corresponding to the above embodiment, the embodiment of the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium may store a program, wherein when the program is running, the device where the computer-readable storage medium is located may be controlled to perform some or all of the steps in the above method embodiment. In a specific implementation, the computer-readable storage medium may be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).
[0133] Corresponding to the above-mentioned embodiment, the embodiment of the present application further provides a computer program product, which includes executable instructions. When the executable instructions are executed on a computer, the computer executes some or all of the steps in the above-mentioned method embodiment.
[0134] In the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.
[0135] Those of ordinary skill in the art will appreciate that the various units and algorithm steps described in the embodiments disclosed herein can be implemented in a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.
[0136] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0137] In several embodiments provided in the present application, any function can be stored in a computer-readable storage medium if it is implemented in the form of a software functional unit and sold or used as an independent product. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., and other media that can store program codes.
[0138] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.
Claims
1. A method for intelligent identification of core sedimentary structures, characterized in that: include: Select the sedimentary structure type and obtain the original image of the core; Screening and processing the original core image to obtain a core image; Classifying and labeling the core image according to sedimentary structure type; Performing image enhancement and expansion on the core image to obtain a data set; Build models based on deep learning image classification algorithms; Conduct model training result analysis and model testing; If the model fails the model test, the data set is optimized and the model is iteratively updated; After the model passes the model test, it is published and applied online to intelligently identify the core sedimentary structure.
2. The method according to claim 1, characterized in that Sedimentary structural types include: grain-sequence bedding, parallel bedding, wavy bedding, lens bedding and charcoal foliation.
3. The method according to claim 1, characterized in that The original core image is screened and processed, specifically including: Remove images that are too blurred, have unclear sedimentary structures, or have too much interference information, and select the original core images with obvious sedimentary structures; Identify the sedimentary structure types of the screened core raw images; Delete the interference information of the filtered core original image; The length, width and size of the image are processed to obtain a single sedimentary structure image.
4. The method according to claim 1, characterized in that The image enhancement and expansion includes: adjusting image brightness, adjusting image color balance, adjusting image clarity, randomly blocking a portion of the image, flipping the image left and right, rotating the image, and adjusting the color of the image.
5. The method according to claim 1, characterized in that Perform model training result analysis and model testing, specifically: Importing the data set into the model, randomly selecting a portion of images in the data set to form a training set, and the remaining images to form a test set, to perform the first model training; When the sample quality is high, the classification feature discrimination is reasonable, and the model accuracy reaches the target range, the data set is randomly divided into a training set and a test set, and dozens of trainings are performed to obtain the model accuracy; For any type of sedimentary structure, the F1-score index is further used to evaluate the model classification results to obtain the F1-score index of the model; When both the model accuracy and F1-score indicators are qualified, the model is trained using a data set that has never participated in the model training set and test set. If the model accuracy reaches more than 85%, the model passes the test.
6. The method according to claim 5, characterized in that The data set is optimized and the model is iteratively updated, specifically: If the model accuracy is not up to standard, check the F1-score index of each sedimentary structure type and determine the sedimentary structure type with the lowest score; The datasets of the sedimentary structural types with the lowest scores were optimized by removing interfering images and supplementing them with more images with obvious features; Use the optimized dataset to iteratively update the model.
7. The method according to claim 1, characterized in that The model can intelligently identify core sedimentary structures through a mobile phone.
8. Intelligent identification device for core sedimentary structure, characterized in that: include: An image acquisition module is used to select the sedimentary structure type and obtain the original image of the core; An image processing module is used to screen and process the original core image to obtain a core image; An image annotation module, used to classify and annotate the core image according to sedimentary structure type; An image expansion module, used to perform image enhancement and expansion on the core image to obtain a data set; Model building module, used to build models based on deep learning image classification algorithms; Model testing module, used for model training result analysis and model testing; A model updating module, used for optimizing the data set and iteratively updating the model when the model fails the model test; The model publishing module is used for model publishing and online application, and for intelligent identification of core sedimentary structures.
9. An electronic device, characterized in that: include: processor; Memory; And a computer program, wherein the computer program is stored in the memory, and the computer program includes instructions, and when the instructions are executed by the processor, the electronic device executes the method according to any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored program, wherein when the program is executed, the device where the computer-readable storage medium is located is controlled to execute the method according to any one of claims 1 to 7.
Citation Information
Patent Citations
Clastic rock lithology identification method and system based on deep learning method
CN113128477A
Sedimentary rock category identification method and device, electronic equipment and storage medium
CN113807449A
Artificial intelligence-based microscopic rock and ore image recognition method
CN114638992A
Cited By
Rock sample bedding identification method and device based on machine learning
CN121437907A
A rock sample bedding identification method and device based on machine learning
CN121437907B
