Two-dimensional lithology intelligent identification method for collected rock core

Abstract

The invention provides a two-dimensional lithology intelligent identification method for a collected rock core. The method comprises the following steps: step 1, collecting and analyzing rock core and geological information of a research area; 2, performing optical scanning on the rock core to generate a rock core optical image; step 3, performing preprocessing and manual marking on the rock core optical image; 4, carrying out a machine learning-based label sample fine processing technology by utilizing the obtained optical image data and the rock core label, and generating an intelligent training label sample set; and 5, establishing a U-Net + +-based two-dimensional rock core image recognition model, and forming an artificial intelligence-based rock core scanning and digital logging technology. According to the invention, a digital target of a physical rock core library in a geological reference center and intelligent identification of unidentified rock cores can be realized.

Description

Technical Field

[0001] The present invention relates to the intelligent identification application of stored cores, and more particularly, to a method for intelligent two-dimensional lithology identification of stored cores. Background Art

[0002] Cores are the most objective results obtained in geological work. Core logging is the process of sorting, describing, classifying, numbering, packing, and warehousing the cores obtained from drilling. Its purpose is to scientifically and objectively record and arrange the original parameters inherent in the cores, describe the observation results of the cores using standardized geological terms, and register all sample numbers, sampling locations, and analysis purposes. By digitizing and scanning the cores and building a database, complete core graphic and text information can be provided for professionals, effectively improving the utilization efficiency of the cores.

[0003] With the development of modern technology, exploration and production in resource fields such as mines and oilfields have become increasingly digital. Through digitization, the efficiency of data collection and processing can be improved, and safety and reliability can be enhanced. Traditional core collection methods require the use of professional equipment and on-site operations, which are relatively complex, time-consuming, and laborious. Summary of the Invention

[0004] In view of the problem that traditional core collection methods require the use of professional equipment and on-site operations, which are relatively complex, time-consuming, and laborious, the present invention proposes a method for intelligent two-dimensional lithology identification of stored cores.

[0005] According to an intelligent two-dimensional lithology identification method for stored cores of the present invention, it includes the steps of: Step 1: Collect and analyze core and geological information in the study area; Step 2: Optically scan the cores to generate core optical images; Step 3: Perform preprocessing and manual labeling on the core optical images; Step 4: Utilize the obtained optical image data and core labels to carry out a fine processing technology for label samples based on machine learning to generate an intelligent training label sample set; Step 5: Establish a two-dimensional core image recognition model based on U-Net++, and form a core scanning and digital logging technology based on artificial intelligence.

[0006] Further, Step 1 includes: collecting geological data related to the study area, including geological profiles and lithological histograms; the data comes from geological surveys, satellite images, and borehole data; according to the geological characteristics and research task requirements of the study area, determining the classification criteria for the lithology and geological structure of the study area.

[0007] Further, Step 2 includes: conducting optical scanning of cores in the core library, and obtaining optical images of the cores through a core scanner; the scanning method is rolling scanning, and the cores are rotated 360° during the scanning process.

[0008] Further, Step 3 includes: preprocessing the obtained core images, mainly including image cropping, scaling, denoising, and unifying the image resolution; then, converting the optical images of the cores into RGB data sets. During the resampling of the resolution of the optical images, data feature engineering processing is carried out, including obtaining standard deviation, maximum entropy, and filter-based feature extraction; using an unsupervised autoencoder model to generate random multi-depth filters to scan the images, obtaining multi-depth RGB data sets, increasing the input parameters of deep learning, and improving the accuracy; performing manual labeling of the labels, judging the lithology of the obtained columnar cores, performing manual labeling, and annotating through Labelme software to obtain a label sample set for two-dimensional lithology intelligent recognition, and drawing the corresponding two-dimensional lithology conversion map.

[0009] Further, Step 4 includes: identifying, eliminating, and relabeling the mislabeled and coarsely labeled parts in the manually labeled label samples; among them, the process of reinforcement supervised learning is divided into three parts: anomaly detection, semi-supervised training, and relabeling. In the anomaly detection stage, the isolation forest algorithm is used to identify abnormal pixel points, and the labels of the abnormal points are eliminated as unlabeled data; in the semi-supervised training stage, the triple training method is used, and reliable labeled data is used for training; in the relabeling stage, the model trained by semi-supervised training is used to re-predict the samples with the labels eliminated, and labels are assigned to obtain a reliable training sample set to improve the accuracy of lithology sample training and prediction.

[0010] Further, Step 5 includes: conducting research on two-dimensional core image recognition and digital cataloging based on the deep neural network of U-Net++; constructing a network model suitable for two-dimensional lithology recognition of core images, determining network structure, input parameters, output parameters, and optimization algorithms; using the label sample set after fine processing of the label samples to conduct training of the U-Net++ network model; through repeated iteration, adjusting the model parameters until the model performance meets the accuracy requirements; using the validation data set to verify the model, evaluating the accuracy and robustness of the model; using the trained network model to conduct research on lithology prediction of well section cores, forming a two-dimensional lithology and geological structure columnar diagram of the core images; using the anisotropic diffusion method to denoise and smooth the generated images to improve the image quality; optimizing the lithology boundary and tectonic boundary to make them clearer and more accurate.

[0011] Beneficial effects: Compared with the prior art, the present invention has the following advantages: The present invention can automate the acquisition and transmission processes through digitization, and at the same time can quickly process and store data, improving the data processing efficiency. In addition, an artificial intelligence model is established between the core optical images and the core lithology labels, so as to intelligently identify a large number of unrecognized cores in the core library, thereby reducing the time-consuming of core digitization cataloging and establishing an efficient and convenient digital core database. The present invention proposes an intelligent recognition application method based on the collected cores, forming a core scanning and digitization cataloging technology based on artificial intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present invention. The illustrative embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings: Figure 1 is a specific technical route flowchart of an embodiment of the present invention.

[0013] Figure 2 is a one-dimensional lithology histogram, lithology classification, and the distribution of the number of various lithology samples in an embodiment of the present invention.

[0014] Figure 3 is an optical image of core rolling and scanning in an embodiment of the present invention.

[0015] Figure 4 is the preprocessing (denoising, cropping, etc.) of the rolling and scanning optical image in an embodiment of the present invention.

[0016] Figure 5 is a schematic diagram of an unsupervised autoencoder model and filter in an embodiment of the present invention.

[0017] Figure 6 is the optical image feature engineering processing in an embodiment of the present invention; (a) is the original optical image; (b) is the image obtained by extracting the standard deviation; (c) is the image obtained by extracting the entropy.

[0018] Figure 7 is a schematic diagram of the processing of noise label samples (a) and application effect (b) in an embodiment of the present invention.

[0019] Figure 8 is a schematic diagram of the U-Net++ network structure (a) and the result of two-dimensional lithology recognition generation (b) in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] The present invention will be described in detail below with reference to the drawings and in conjunction with the embodiments. It should be noted that, without conflict, the embodiments in the present invention and the features in the embodiments can be combined with each other.

[0021] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily describe a specific order or sequence.

[0022] Please refer to Figures 1 to 8 , the present invention provides a method for two-dimensional lithology intelligent recognition of library core samples.

[0023] Step 1: Refer to Figure 2 , collect and analyze core samples and geological information in the study area; Collect geological cross-sections, lithology columnar diagrams and other geological-related data in the study area. The data can come from geological surveys, satellite images, borehole data, etc. On this basis, according to the geological characteristics of the study area and the requirements of the research task, determine the classification criteria for the lithology and geological structure in the study area, providing a basis for the artificial intelligence training and prediction of two-dimensional lithology recognition and digital cataloging of core images.

[0024] Step 2: Refer to Figure 3 , perform optical scanning on the core samples to generate core optical images; Carry out optical scanning of core samples in the core library, and obtain the optical images of core samples through a core scanner; the scanning method is rolling scanning, and the core samples are rotated 360° during the scanning process. For complete columnar core samples, the scanning speed is slower and the obtained information is more complete.

[0025] Step 3: Refer to Figure 4 , Figure 5 and Figure 6 , perform preprocessing of the optical images of the core samples and manual labeling; Preprocess the obtained core images, mainly including image cropping, scaling, denoising and unifying the image resolution, etc. Then, convert the optical images of the core samples into RGB data sets. During the resampling process of the resolution of the optical images, carry out data feature engineering processing, such as obtaining standard deviation, maximum entropy, filter-based feature extraction, etc. In particular, use an unsupervised autoencoder model to generate random multi-depth filters to scan the images, obtain multi-depth RGB data sets, increase the input parameters of deep learning, and improve the accuracy. Among them, the random multi-depth filters generated by the autoencoder ensure the diversity and randomness of image scanning, so as to extract richer image features. Finally, perform manual labeling of the labels. This work is carried out by geological experts to judge the lithology of the obtained columnar core samples, perform manual labeling, and annotate through the Labelme software to obtain a label sample set for two-dimensional lithology intelligent recognition, and draw the corresponding two-dimensional lithology conversion diagram.

[0026] Step 4: Refer to Figure 7, the optical image data and core labels obtained in the foregoing steps are used to develop a fine processing technology for label samples based on machine learning to generate an intelligent training label sample set.

[0027] For the possible mislabeling and coarsening labeling in the manually labeled label samples, reinforcement supervised learning is used for identification, elimination, and relabeling. Among them, the process of reinforcement supervised learning can be divided into three parts: anomaly detection, semi-supervised training, and relabeling. In the anomaly detection stage, the Isolation Forest algorithm is used to identify abnormal pixel points, and the labels of the abnormal points are eliminated as unlabeled data. In the semi-supervised training stage, the triple training method is used to train with reliable labeled data. In the relabeling stage, the model trained by semi-supervised training is used to re-predict the samples with the eliminated labels and assign labels. Finally, a reliable training sample set is obtained to improve the accuracy of lithology sample training and prediction.

[0028] Step Five: Refer to Figure 8 , establish a two-dimensional core image recognition model based on U-Net++, and form a core scanning and digital cataloging technology based on artificial intelligence.

[0029] Conduct research on two-dimensional core image recognition and digital cataloging based on the deep neural network of U-Net++; construct a network model suitable for two-dimensional lithology recognition of core images, and determine the network structure, input parameters, output parameters, optimization algorithms, etc.; use the label sample set after fine processing of label samples to carry out training of the U-Net++ network model. Through repeated iteration, adjust the model parameters until the model performance meets the accuracy requirements. At the same time, use the validation data set to verify the model and evaluate the accuracy and robustness of the model; use the trained network model to carry out lithology prediction research on the core of the well section to form a two-dimensional lithology and geological structure columnar diagram of the core image; use the anisotropic diffusion method to denoise and smooth the generated image to improve the image quality; optimize the lithology boundary and tectonic boundary to make it clearer and more accurate.

[0030] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A two-dimensional lithology intelligent recognition method for a collection of core samples, characterized in that Including the steps: Step 1: Collect and analyze the core and geological information of the study area; Step 2: Conduct optical scanning on the core to generate core optical images; Step 3: Preprocess and manually label the core optical images; Step 4: Utilize the obtained optical image data and core labels to carry out a fine processing technology for label samples based on machine learning, and generate an intelligent training label sample set; Step 5: Establish a two-dimensional core image recognition model based on U-Net++, and form a core scanning and digital cataloging technology based on artificial intelligence.

2. The two-dimensional lithology intelligent identification method for the collected core according to claim 1, characterized in that Step 1 includes: Collecting geological-related data of the study area, including geological profiles and lithologic columns; the data comes from geological surveys, satellite images, and borehole data; according to the geological characteristics of the study area and the requirements of the research task, determining the classification criteria for the lithology and geological structure of the study area.

3. The two-dimensional lithology intelligent recognition method for the collection of core samples according to claim 1, characterized in that Step 2 includes: Conducting core optical scanning work on the core library, and obtaining the optical images of the core through a core scanner; the scanning method is rolling scanning, and the core is rotated 360° during the scanning process.

4. The two-dimensional lithology intelligent identification method for the collected cores according to claim 1, characterized in that Step 3 includes: Preprocessing the obtained core images, mainly including image cropping, scaling, denoising, and unifying the image resolution; then, converting the optical images of the core into an RGB data set, and during the resampling process of the optical image resolution, carrying out data feature engineering processing, including obtaining the standard deviation, maximum entropy, and filter-based feature extraction; using an unsupervised autoencoder model to generate random multi-depth filters to scan the image, obtaining a multi-depth RGB data set, increasing the input parameters of deep learning, and improving the accuracy; carrying out manual labeling of the labels, judging the lithology of the obtained columnar core, carrying out manual labeling, and annotating through the Labelme software to obtain a label sample set for two-dimensional lithology intelligent recognition, and drawing the corresponding two-dimensional lithology conversion map.

5. The two-dimensional lithology intelligent recognition method for the collected cores according to claim 1, characterized in that Step 4 includes: Identifying, eliminating, and relabeling the mislabeled and coarsely labeled parts in the manually labeled label samples; among them, the process of reinforcement supervised learning is divided into three parts: anomaly detection, semi-supervised training, and relabeling. In the anomaly detection stage, the isolation forest algorithm is used to identify abnormal pixel points, and the labels of the abnormal points are eliminated as unlabeled data; in the semi-supervised training stage, the triple training method is used, and reliable labeled data is used for training; in the relabeling stage, the model trained by semi-supervised training is used to re-predict the samples with the labels eliminated, and labels are assigned to obtain a reliable training sample set to improve the accuracy of lithology sample training and prediction.

6. The two-dimensional lithology intelligent recognition method for the collected core according to claim 1, characterized in that, Step 5 includes: conducting research on the recognition and digital cataloging of two-dimensional core images based on the deep neural network of U-Net++; constructing a network model suitable for two-dimensional lithology recognition of core images, and determining the network structure, input parameters, output parameters, and optimization algorithm; using the labeled sample set after fine processing of the labeled samples to carry out the training of the U-Net++ network model; through repeated iteration, adjusting the model parameters until the model performance meets the accuracy requirements; using the validation data set to verify the model and evaluate the accuracy and robustness of the model; using the trained network model to carry out research on lithology prediction of well section cores to form a two-dimensional lithology and geological structure columnar chart of the core images; using the anisotropic diffusion method to denoise and smooth the generated images to improve the image quality; optimizing the lithology boundary and tectonic boundary to make them clearer and more accurate.

Images