Rock Images Research Articles

Rock Image Classification Based on EfficientNet and Triplet Attention Mechanism

Rock image classification is a fundamental and crucial task in the creation of geological surveys. Traditional rock image classification methods mainly rely on manual operation, resulting in high costs and unstable accuracy. While existing methods based on deep learning models have overcome the limitations of traditional methods and achieved intelligent image classification, they still suffer from low accuracy due to suboptimal network structures. In this study, a rock image classification model based on EfficientNet and a triplet attention mechanism is proposed to achieve accurate end-to-end classification. The model was built on EfficientNet, which boasts an efficient network structure thanks to NAS technology and a compound model scaling method, thus achieving high accuracy for rock image classification. Additionally, the triplet attention mechanism was introduced to address the shortcoming of EfficientNet in feature expression and enable the model to fully capture the channel and spatial attention information of rock images, further improving accuracy. During network training, transfer learning was employed by loading pre-trained model parameters into the classification model, which accelerated convergence and reduced training time. The results show that the classification model with transfer learning achieved 92.6% accuracy in the training set and 93.2% Top-1 accuracy in the test set, outperforming other mainstream models and demonstrating strong robustness and generalization ability.

Upscaling Permeability Using Multiscale X‐Ray‐CT Images With Digital Rock Modeling and Deep Learning Techniques

AbstractThis study presents a workflow to predict the upscaled absolute permeability of the rock core direct from CT images whose resolution is not sufficient to allow direct pore‐scale permeability computation. This workflow exploits the deep learning technique with the data of raw CT images of rocks and their corresponding permeability value obtained by performing flow simulation on high‐resolution CT images. The permeability map of a much larger region in the rock core is predicted by the trained neural network. Finally, the upscaled permeability of the entire rock core is calculated by the Darcy flow solver, and the results showed a good agreement with the experiment data. This proposed deep learning based upscaling method allows estimating the permeability of large‐scale core samples while preserving the effects of fine‐scale pore structure variations due to the local heterogeneity.

Processing of micro-CT images of granodiorite rock samples using convolutional neural networks (CNN), Part II: Semantic segmentation using a 2.5D CNN

X-ray computed tomography (XCT) is routinely used in geosciences for the purpose of rock characterisation. High-quality micro-CT images are successfully used for fracture characterisation, as well as analysis of grains and pores. In contrast, the use of XCT for mineral identification is uncommon and often ineffective. Implementation of micro-CT imaging techniques for mineral identification is affected by the accuracy and precision of the image segmentation results. Conventional segmentation methods such as thresholding, watershed, and active contouring are user-biased and do not provide the robust distinction between various heavy accessory minerals in granite rocks. Heavy ore minerals such as pyrite, chalcopyrite, molybdenite, and ilmenite are readily recognised in grey-scale micro-CT images because of their high attenuation coefficient, but further differentiation between these minerals using only traditional segmentation methods is challenging. Conversely, deep convolutional neural networks (CNNs) are fully self-trained, and they have demonstrated accurate semantic segmentation results for rock images. However, the application of CNN semantic segmentation for igneous rocks is not well documented. In this research, the U-Net 2.5D CNN was deployed to train the neural network on a combination of high-resolution micro-CT and mineral liberation analysis (MLA) images to identify different accessory mineral regions of interest (ROIs). The image segmentation results were assessed using MLA and SEM data, and the accuracy of segmentation was found to be greater than 97%. The methodology developed in this study can be extended to map the mineralogy of granite samples unseen by the CNN to further validate the robustness of the approach.

Novel Rock Image Classification: The Proposal and Implementation of HKUDES_Net

Novel Rock Image Classification: The Proposal and Implementation of HKUDES_Net

Retraction Note: Novel Rock Image Classification: The Proposal and Implementation of RockNet

Retraction Note: Novel Rock Image Classification: The Proposal and Implementation of RockNet

Micro-Scale Lattice Boltzmann Simulation of Two-Phase CO2–Brine Flow in a Tighter REV Extracted from a Permeable Sandstone Core: Implications for CO2 Storage Efficiency

Deep saline permeable sandstones have the potential to serve as sites for CO2 storage. However, unstable CO2 storage in pores can be costly and harmful to the environment. In this study, we used lattice Boltzmann (LB) simulations to investigate the factors that affect steady-state CO2–brine imbibition flow in sandstone pores, with a focus on improving CO2 storage efficiency in deep saline permeable sandstone aquifers. We extracted three representative element volumes (REVs) from a digital rock image of a sandstone core and selected a tighter REV in the upper subdomain so that its permeability would apparently be lower than that of the other two based on single-phase LB simulation for further analysis. The results of our steady-state LB simulations of CO2–brine imbibition processes in the tighter REV under four differential pressures showed that a threshold pressure gradient of around 0.5 MPa/m exists at a differential pressure of 200 Pa, and that higher differential pressures result in a greater and more linear pressure drop and stronger channelization after the flow are initiated. Furthermore, we conducted simulations over a range of target brine saturations in the tighter REV at the optimal differential pressure of 400 Pa. Our findings showed that the relative permeability of CO2 is greatly reduced as the capillary number falls below a certain threshold, while the viscosity ratio has a smaller but still significant effect on relative permeability and storage efficiency through the lubrication effect. Wettability has a limited effect on the storage efficiency, but it does impact the relative permeability within the initial saturation range when the capillary number is low and the curves have not yet converged. Overall, these results provide micro-scale insights into the factors that affect CO2 storage efficiency in sandstones.

Signatures of Life Detected in Images of Rocks Using Neural Network Analysis Demonstrate New Potential for Searching for Biosignatures on the Surface of Mars.

Microorganisms play a role in the construction or modulation of various types of landforms. They are especially notable for forming microbially induced sedimentary structures (MISS). Such microbial structures have been considered to be among the most likely biosignatures that might be encountered on the martian surface. Twenty-nine algorithms have been tested with images taken during a laboratory experiment for testing their performance in discriminating mat cracks (MISS) from abiotic mud cracks. Among the algorithms, neural network types produced excellent predictions with similar precision of 0.99. Following that step, a convolutional neural network (CNN) approach has been tested to see whether it can conclusively detect MISS in images of rocks and sediment surfaces taken at different natural sites where present and ancient (fossil) microbial mat cracks and abiotic desiccation cracks were observed. The CNN approach showed excellent prediction of biotic and abiotic structures from the images (global precision, sensitivity, and specificity, respectively, 0.99, 0.99, and 0.97). The key areas of interest of the machine matched well with human expertise for distinguishing biotic and abiotic forms (in their geomorphological meaning). The images indicated clear differences between the abiotic and biotic situations expressed at three embedded scales: texture (size, shape, and arrangement of the grains constituting the surface of one form), form (outer shape of one form), and pattern of form arrangement (arrangement of the forms over a few square meters). The most discriminative components for biogenicity were the border of the mat cracks with their tortuous enlarged and blistered morphology more or less curved upward, sometimes with thin laminations. To apply this innovative biogeomorphological approach to the images obtained by rovers on Mars, the main physical and biological sources of variation in abiotic and biotic outcomes must now be further considered.

Image segmentation and flow prediction of digital rock with U-net network

Computed tomography (CT) images of sandstone contain rich reservoir information. Analyzing digital rock images is important for geological research and the flow in the subsurface. This paper presents a workflow for assessing digital rock petrophysical properties based on machine learning techniques, including 1) automatic segmentation of sandstone rock images using U-net networks, 2) permeability prediction using machine learning, and 3) flow simulation by deep learning. First, using the U-net network, the rock images are binary-segmented into matrix and pore, and multisegmented into the matrix, pore, and mineral. The accuracy and intersection over union (IOU) are used to evaluate the performance of image segmentation. The accuracy and IOU of binary segmentation results are 99.87% and 0.9986, and the results for multi-segmentation are 96.77% and 0.7281, respectively. Then, the key features of CT images influencing sandstone permeability are extracted, and the analysis of image features reveals that the hydraulic radius is the most important parameter for permeability prediction. After that, the sandstone permeability is predicted by long short-term memory (LSTM) and random forest (RF) and then compared with the permeability calculated by the lattice Boltzmann (LBM) method. The mean square error (MSE), mean absolute error (MAE), and root mean square error (RMSE) are used to quantitatively evaluate the error of permeability prediction. The studies show that the precision of RF in permeability prediction is higher than that of LSTM, and when all the feature parameters are used as input, the accuracy of permeability prediction is a little higher than that when only the hydraulic radius is used as input. Finally, this paper refines a new U-net model to predict the flow velocity field from CT images, and this new U-net model can reduce the computation time by 98.59% compared with the LBM method. This study will be significant for applying deep learning in simulate the flow in digital rock.

Rock image classification using deep residual neural network with transfer learning

Rock image classification is a significant part of geological research. Compared with traditional image classification methods, rock image classification methods based on deep learning models have the great advantage in terms of automatic image features extraction. However, the rock classification accuracies of existing deep learning models are unsatisfied due to the weak feature extraction ability of the network model. In this study, a deep residual neural network (ResNet) model with the transfer learning method is proposed to establish the corresponding rock automatic classification model for seven kinds of rock images. ResNet34 introduces the residual structure to make it have an excellent effect in the field of image classification, which extracts high-quality rock image features and avoids information loss. The transfer learning method abstracts the deep features from the shallow features, and better express the rock texture features for classification in the case of fewer rock images. To improve the generalization of the model, a total of 3,82,536 rock images were generated for training via image slicing and data augmentation. The network parameters trained on the Texture Library dataset which contains 47 types of texture images and reflect the characteristics of rocks are used for transfer learning. This pre-trained weight is loaded when training the ResNet34 model with the rock dataset. Then the model parameters are fine-tuned to transfer the model to the rock classification problem. The experimental results show that the accuracy of the model without transfer learning reached 88.1%, while the model using transfer learning achieved an accuracy of 99.1%. Aiming at geological engineering field investigation, this paper studies the embedded deployment application of the rock classification network. The proposed rock classification network model is transplanted to an embedded platform. By designing a rock classification system, the off-line rock classification is realized, which provides a new solution for the rock classification problem in the geological survey. The deep residual neural network and transfer learning method used in this paper can automatically classify rock features without manually extracting. These methods reduce the influence of subjective factors and make the rock classification process more automatic and intelligent.

How petrophysical properties influence the seismic signature of carbonate fault damage zone: Insights from forward-seismic modelling

Seismic imaging is the most common tool to detect subsurface structures in continental and marine settings. Despite technological advancement, seismic analysis of carbonates is still challenging because they are strongly influenced by petrophysical heterogeneities, being this even more difficult when further heterogeneities are added by the presence of faults. In this work, unmigrated forward-seismic models are developed to understand the seismic response changes related to carbonate-bearing fault systems and their deformation behaviour. We focus on the carbonate ramp of the Majella Massif (central Italy), that is characterised by the presence of porous and faulted carbonate lithologies. This carbonate system represents an analogue of carbonate reservoirs worldwide. Field and laboratory data of fault rocks sampled at increasing distances from the slip planes highlight a damage zone/fault core architecture with a decreasing porosity approaching fault planes. This observation is also confirmed by the thin sections analysis and the calculation of the shear modulus. Seismic images of fault rocks presenting lower porosity than the host rock show weak diffraction hyperbolas while diffraction hyperbolas are more evident in the forward-seismic models of fault rocks with greater porosity than the host rock. Moreover, factors such as the variation in the dip angle of the fault or the thickness of the damage zone can enhance or reduce the hyperbolas. Performing the migration process to stacked sections would not provide evidence of increased porosity in the damage zone due to the suppression of the diffractive component. The presence of weak hyperbolas in unmigrated seismic images is thus interpreted as evidence of a lower porosity damage zone with respect to the host rock. This can be related to deformation mechanisms leading to porosity loss in the damage zone for porous rocks as observed in the study area. However, this could also be related to the confining stress or fracture filling that counteract the fracture-related porosity increase when faults are hosted by tight rocks.

A Pixel-Wise k-Immediate Neighbour-Based Image Analysis Approach for Identifying Rock Pores and Fractures from Grayscale Image Samples

The purpose of the current study is to propose a novel meta-heuristic image analysis approach using multi-objective optimization, named ‘Pixel-wise k-Immediate Neighbors’ to identify pores and fractures (both natural and induced, even in the micro-level) in the wells of a hydrocarbon reservoir, which presents better identification accuracy in the presence of the grayscale sample rock images. Pores and fractures imaging is currently being used extensively to predict the amount of petroleum under adequate trap conditions in the oil and gas industry. These properties have tremendous applications in contaminant transport, radioactive waste storage in the bedrock, and CO2 storage. A few strategies to automatically identify the pores and fractures from the images can be found in the contemporary literature. Several researchers employed classification technique using support vector machines (SVMs), whereas a few of them adopted deep learning systems. However, in these cases, the reported accuracy was not satisfactory in the presence of grayscale, low quality (poor resolution and chrominance), and irregular geometric-shaped images. The classification accuracy of the proposed multi-objective method outperformed the most influential contemporary approaches using deep learning systems, although with a few restrictions, which have been articulated later in the current work.

НАСКАЛЬНЫЕ ИЗОБРАЖЕНИЯ ХОДЖИКЕНТА

The limestone rock with petroglyphs in the village of Khodzikent (Uzbek: Xo'jakent) is one of the most well-known and accessible rock art sites of Uzbekistan, with a rich history of discovery and study. The first mentions of signs on the rock have been known in print since 1879, but the systematic study of rock images has been organized by many researchers since the middle of the 20th century, such as H.A. Alpysbaev, G.V. Parfenov, G.V. Shatsky, D. Kabirov, M.M. Khuzhanazarov, etc. In addition to the published data, this work includes unique archival materials that shed new light on the research of the Khodzikent petroglyphs by the Bostandyk archaeological team led by G.V. Parfenov. The analysis of historiographical materials reveals a number of controversial issues related primarily to the quality of fixation of pictorial sources and discrepancies in their chronological attribution. A continuous survey of promising sites and a documentation technique, developed taking into account the specifics of the Khodzikent petroglyphs (the smoothness of their contours), allowed the authors to obtain qualitatively new results: to identify a previously unknown surface with the image of a “rosette”, fix 119 rock images on the “central” rock and determine the sequence of their creation using palimpsests and planigraphy analysis, reflect in digital drawings an integral image of the site of rock art. The relative chronology and analysis of iconography and the pictorial series indicate a possible broad chronological framework for the creation of Khodzikent rock images: from the Eneolithic to the early Middle Ages.

Content-based image retrieval for industrial material images with deep learning and encoded physical properties

Abstract Industrial materials images are an important application domain for content-based image retrieval. Users need to quickly search databases for images that exhibit similar appearance, properties, and/or features to reduce analysis turnaround time and cost. The images in this study are 2D images of millimeter-scale rock samples acquired at micrometer resolution with light microscopy or extracted from 3D micro-CT scans. Labeled rock images are expensive and time-consuming to acquire and thus are typically only available in the tens of thousands. Training a high-capacity deep learning (DL) model from scratch is therefore not practicable due to data paucity. To overcome this “few-shot learning” challenge, we propose leveraging pretrained common DL models in conjunction with transfer learning. The “similarity” of industrial materials images is subjective and assessed by human experts based on both visual appearance and physical qualities. We have emulated this human-driven assessment process via a physics-informed neural network including metadata and physical measurements in the loss function. We present a novel DL architecture that combines Siamese neural networks with a loss function that integrates classification and regression terms. The networks are trained with both image and metadata similarity (classification), and with metadata prediction (regression). For efficient inference, we use a highly compressed image feature representation, computed offline once, to search the database for images similar to a query image. Numerical experiments demonstrate superior retrieval performance of our new architecture compared with other DL and custom-feature-based approaches.

Using Digital Rock Physics to Evaluate Novel Percussion Core Quality

This paper details a benchmarking and validation workflow using digital rock physics (DRP) to evaluate the effectiveness of various percussion sidewall core (PSWC) acquisition methods. The workflow consists of obtaining digital rock properties and images of reference material to compare with material obtained from laboratory percussion sidewall acquisition, including novel designs. This workflow allows insight into the acquisition processes and potential sources of damage in the PSWC technique, and potentially other rock sampling techniques, and offers an opportunity to evaluate its appropriateness as a subsurface rock sample acquisition method. Sample cubes from six outcrop sandstone formations of known properties were used in the testing program to cover a wide range of particularly low and medium unconfined compressive strengths (UCS). Multiple control rotary plug samples were cut from each sandstone formation. The sample cubes were then used as the parent material in laboratory testing of various designs of PSWC bullets. The PSWC bullets, including novel designs, were shot in simulated downhole environments. Both control rotary plug samples and PSWC test core samples were imaged with high-resolution X-ray micro-computed tomography (micro-CT) at resolutions between 2 and 11 μm, and digital rock analysis was conducted on all samples. Using pre- and post-test images, the damage could be identified, and petrophysical properties, including porosity and permeability, could be determined and directly compared with DRP results from control samples and available routine core analysis (RCA) results. DRP provides unique insight to evaluate and quantify changes (or lack of changes) to the sample material subjected to the PSWC acquisition. Damage encountered in the test samples includes grain crushing and compaction that degrades storage and transport properties and dilatant zones that locally enhance transport properties. The presence, frequency, and distribution of these zones are dependent on experimental parameters. In all cases, undisturbed rock fabric could be identified in each sample, and intact texture was verified by comparison with reference material. A novel and efficient method for acquiring and evaluating subsurface samples was developed and benchmarked. Lab results indicate this method may be equally applicable to low- and mid-range UCS rocks. This approach enables a cost-effective reservoir characterization strategy. By optimizing PSWC bullet design and coupling this with mature, image-based digital rock technology, this work demonstrated that the samples and results obtained by this method are representative and that the controls on storage and transport properties are well understood.

Digital Core Analysis of Capillary Pressure in Sandstone

This research aimed to obtain the petrophysical parameters of the capillary pressure of a sandstonefrom 3D modelling of high-resolution rock images. By integrating a number of 2D slice results, a plot of capillarypressure vs water saturation was built. When performing digital simulations using 512 x 512 x 512 pixels, theeffect on the resulting image was clearly seen using interactive thresholding. The obtained porosity was 29.5%while the permeability was obtained through iteration as 3942 mD, a quite large value since it is a synthetic core.Meanwhile, at the depth of free water level of 984 ft, the capillary pressure is found to be 167.36 psi with watersaturation at 4%.

Microscale THMC Modeling of Pressure Solution in Salt Rock: Impacts of Geometry and Temperature

Pressure solution, a mechanism that involves tight coupling between the geometry and thermal-hydro-mechanical-chemical (THMC) processes, plays an important role in diagenesis. In this study, we make the first attempt to conduct microscale THMC modeling to understand and quantify the impacts of geometry and temperature on pressure solution, taking natural salt rock as an example. This modeling capability is achieved by expanding a novel MC code that we developed previously (Hu et al. J Geophys Res: Solid Earth 126:e2021JB023112, 2021) to include temperature effects. We first conduct a simulation of an example that involves a single brine inclusion within a single halite grain and find that the temperature impact is limited for that case. We then extract geometry from an image of a natural salt rock and conduct simulations with different cases: (A) only temperature and no stress, (B) only stress and no temperature, and (C) with both stress and temperature. These different cases result in quite different phenomena. In case A, dissolution and precipitation occur across the entire system due to isolated pore space reaching a localized mass balance between dissolution, precipitation, and diffusion. In case B, intense geometric features (e.g., major asperities, inclusions) in one area undergo stress concentration, thus dominating pressure solution in that area. In case C, pressure solution is spread out at contacting highly stressed geometric features close to the hotter side. We conclude that geometric features dominate stress distribution, thus dominating pressure solution in a natural salt rock that may be affected by the temperature if a sufficient temperature gradient is applied.

Coal rock image recognition method based on improved CLBP and receptive field theory

AbstractRapid coal‐rock identification is one of the key technologies for intelligent and unmanned coal mining. Currently, the existing image recognition algorithms cannot satisfy practical needs in terms of recognition speed and accuracy. In view of the evident differences between coal and rock in visual attributes such as color, gloss and texture, the complete local binary pattern (CLBP) image feature descriptor is introduced for coal and rock image recognition. Given that the original algorithm oversimplifies local texture features by ignoring imaging information from higher‐order pixels and the concave and convex areas between adjacent sampling points, this paper proposes a higher‐order differential median CLBP image feature descriptor to replace the original CLBP center pixel gray with a local gray median, and replace the binary differential with a second‐order differential. Meanwhile, for the high dimensionality of CLBP descriptor histogram and feature redundancy, deep learning perceptual field theory is introduced to realize data nonlinear dimensionality reduction and deep feature extraction. With relevant experiments conducted, the following conclusion can be drawn: (1) Compared with that of the original CLBP, the recognition accuracy of the improved CLBP algorithm is greatly improved and finally stabilized above 94.3% under strong noise interference; (2) Compared with that of the original CLBP model, the single image recognition time of the coal rock image recognition model fusing the improved CLBP and the receptive field theory is 0.0035 s, a reduction of 71.0%; compared with the improved CLBP model (without the fusion of receptive field theory), it can shorten the recognition time by 97.0%, but the accuracy rate still maintains more than 98.5%. The method offers a valuable technical reference for the fields of mineral development and deep mining.

Enhancing resolution of micro-CT images of reservoir rocks using super resolution

Current hardware configuration of micro-CT detectors puts a lower limit on voxel size that can be acquired while maintaining a sufficiently large field of view. This limits the degree to which rock pores can be resolved in a micro-CT image and thus restricting the application envelope of Digital Rock technology. Super resolution techniques can refine voxel size while maintaining a sufficiently large field of view using pairs of low- and high-resolution images for training. However, for interpretation of quality of Digital Rock results, image quality is not determined by voxel size alone but by the degree to which a feature such as pore throat is resolved, which depends on both the physical size of the feature and voxel size. Furthermore, artificially down-sampling finer voxel size images to obtain images of coarser voxel size for training deep learning networks is not sufficient to capture the mapping between images acquired at different resolutions. This is especially true for reservoir rocks because the noise and artifacts introduced during imaging and reconstruction are more complex than that captured by simple down-sampling operation. We overcome these two limitations, by (1) using the ratio of pore throat size and voxel size (N) to group training dataset instead of voxel size and (2) using pairs of registered micro-CT images acquired using a state-of-the-art detector instead of synthetically down-sampled images. We show that combination of these two techniques produced images with better sharpness and contrast and enabled us to refine voxel size significantly beyond what is possible using the current imaging technology while maintaining the field of view.

Quotation of Stage Images of Hard Rock and Heavy Metal Musicians of the 1970-90s in Clothing Design of the 21st Century

Research aim. The article is devoted to the review of modern clothing design projects in the context of quoting the stage images of hard rock and heavy metal musicians of the 1970s-1990s. Research methods. The source analysis method was used to find out the level of scientific development of the problem. Collecting factual material method was used to search for visual information. To describe and process the visual information of stage images of musicians and shows of leading designers’ collections, the Art history methods, aesthetic and structural-compositional analysis, as well as the comparative method, were used. The theoretical generalization method was used to formulate the conclusions. Visual material is taken from printed publications and the Internet. Scientific novelty is in identifying the main stylistic features of the stage costume of performers of the hard rock and heavy metal musical genres in the 70s and 90s of the 20th century, as well as in comparing and finding quotes of stage images of musicians in modern catwalk clothing collections. Conclusions. The development of the stage costume design of hard rock and heavy metal genres with its wide influence on street fashion, as well as “official fashion”, was due to a significant spread of media technologies popularization at the end of the 20th century. Stage images of artists were imitated and introduced into an everyday wardrobe, firstly, by their fans; secondly, its distribution was facilitated by high-profile advertising companies promoting bands and performers; thirdly, it spread through the fashion industry and the work of talented designers who sensed the mood of society and were able to predict the commercial success of the new aesthetic. In the 21st century well-known brands and designers actively quote performers’ stage costumes. Ways of quoting are very diverse: from completely copying images or borrowing a separate assortment, copying forms and details of clothes to the imitation of cut, decoration, and trimming. The most popular quoting is the use of band mascots or logos to print and design materials, various patches, and stickers. Borrowing a characteristic combination of the assortment in creating a general image has recently become the most popular. From the general trends of today, we come to the conclusion that experiments with the form itself and its material components are significantly inferior to experiments with the emotional filling of images, which are successfully provided by quoting the stage costume with its hyperbolized spectacularity and expressiveness.

Hybrid of Deep Learning and Word Embedding in Generating Captions: Image-Captioning Solution for Geological Rock Images.

Captioning is the process of assembling a description for an image. Previous research on captioning has usually focused on foreground objects. In captioning concepts, there are two main objects for discussion: background object and foreground object. In contrast to the previous image-captioning research, generating captions from the geological images of rocks is more focused on the background of the images. This study proposed image captioning using a convolutional neural network, long short-term memory, and word2vec to generate words from the image. The proposed model was constructed by a convolutional neural network (CNN), long short-term memory (LSTM), and word2vec and gave a dense output of 256 units. To make it properly grammatical, a sequence of predicted words was reconstructed into a sentence by the beam search algorithm with K = 3. An evaluation of the pre-trained baseline model VGG16 and our proposed CNN-A, CNN-B, CNN-C, and CNN-D models used BLEU score methods for the N-gram. The BLEU scores achieved for BLEU-1 using these models were 0.5515, 0.6463, 0.7012, 0.7620, and 0.5620, respectively. BLEU-2 showed scores of 0.6048, 0.6507, 0.7083, 0.8756, and 0.6578, respectively. BLEU-3 performed with scores of 0.6414, 0.6892, 0.7312, 0.8861, and 0.7307, respectively. Finally, BLEU-4 had scores of 0.6526, 0.6504, 0.7345, 0.8250, and 0.7537, respectively. Our CNN-C model outperformed the other models, especially the baseline model. Furthermore, there are several future challenges in studying captions, such as geological sentence structure, geological sentence phrase, and constructing words by a geological tagger.