Rock Images Research Articles

Multi-scaling Properties of 2D Reservoir Micro-pore Heterogeneity Based on Digital Casting Thin-Section Images

In recent years, in order to study the micro-pore structures of reservoirs, to define the types of reservoir space and to classify and evaluate reservoir capabilities, the technology of numerical image analysis has been concerned widely with the study of pore heterogeneity with fractals and multifractals. This is of great challenge due to developments in image resolutions and quantitative characterization methods. In this paper, the pore structure of digital casting thin-section images of six samples of volcanic and volcano-sedimentary rocks from the Dagang area (Huanghua depression, China) was studied, and then the pore spaces of the reservoir samples were characterized automatically using multifractal parameters advantageous to describe the irregularity and disorder of geometric objects. The method of moments was used to derive the parameters of multifractality and heterogeneity of pore structures. The results show well that, together with cluster analysis, 12 fractal and multifractal parameters can be applied to describe effectively the irregular degrees of pore spaces. The six rock samples have been classified into three groups with various reservoir capabilities. The work presented here can provide new avenues for analysis of the spatial arrangement of mineral particles, aggregates and pores in spaces, and that the multi-scaling nature of casting thin-section images of rocks suggests using these algorithms as a basis for reservoir capability assessment.

CNN-PFVS: Integrating Neural Network and Finite Volume Models to Accelerate Flow Simulation on Pore Space Images

Direct numerical simulations of flow on micro-computed tomography (micro-CT) images are extensively used in many disciplines of science and engineering. Recently, we have developed a pore-scale finite volume solver (PFVS) to directly solve for flow on micro-CT images and predict permeability of digital cores. The solver assigns a local conductivity to each voxel based on geometrical and topological constraints. The local conductivity term in PFVS is conventionally calculated by an iterative local scanning algorithm, where the number of iterations depends on the size of the largest flow channel. This can increase the computation time of PFVS significantly if the largest flow channel is reasonably large. In this paper, we apply convolutional neural networks (CNN) to predict local conductivity for each voxel, thus bypassing the iterative algorithm while also preserving the mass conservation in the system by still solving for flow using conventional methods. The network is trained to convert segmented binary images of rocks into a numerical map required for flow simulation by the use of paired image-to-image translation using a ResNet-Style architecture. Comparison of the generated and original coefficient maps shows that the average error is within 1% over the 3D pore geometries used in this study. Then, we compare the absolute permeability results obtained from the original PFVS and the CNN-PFVS and the errors are within 20% with the average of 13.8%. Machine learning improves the computation time significantly especially on the images with large domain size and flow channels. On the samples tested, the speedup factor is 10 times using CNN compared to iterative calculations.

Efficient extraction of pore networks from massive tomograms via geometric domain decomposition

Image-based modelling of porous media to study the transport and reaction processes has become an essential tool. The availability of increasingly large image datasets at high resolution creates a need to develop algorithms that can process massive size images at a low computational cost. This study presents an efficient workflow to extract pore networks from large size porous domains using a watershed segmentation with geometrical domain decomposition. The method subdivides a porous image into smaller overlapping subdomains and performs a watershed segmentation on each subdomain in parallel or serial modes of operation to save CPU time or memory RAM, respectively. The computational performance of the algorithm was analyzed on a large size image and found to consume 50 percent less memory and upto 7 times less CPU time than the standard watershed implementation. Pore networks of four massive digital rock images were extracted and the the effective permeability predicted by the networks agreed well with previously investigated values illustrating the accuracy of the method. An additional application of this method, taking advantage of the reduced computational cost, is the upgrading of low-resolution image. It was found that increasing the resolution of a coarse image leads to more accurate predictions by helping the watershed segmenation prouduce a more faithful pore network model. The developed algorithm is implemented in Python, and included in the open source project PoreSpy. It uses highly optimized and efficient modules such as Dask and Numba to obtain the maximum performance. The domain decomposition approach used here will also lend itself well to processing on distributed memory clusters, enabling the processing of even larger porous domains.

Permeability-porosity relations from single image of natural rock: Subsampling approach

Natural rock is heterogeneous at all scales. This challenges the digital rock physics concept since the images of natural rock are mm-sized, while the results should be applicable at much larger spatial scales. To address this challenge we subsample a digital volume and compute the permeability of its subvolumes, as well as that of the entire volume, using the lattice-Boltzmann method (LBM). We reassemble the resulting values in 3D and use Darcy's flow simulator to re-compute the permeability of the entire host volume. In spite of significant heterogeneity of the subvolume properties, the Darcy-upscaled entire-volume permeability falls very close to that LBM-computed. Next, by using random reconstructions of the subvolume permeability matrix, we obtain physically meaningful permeability versus porosity transforms that can arguably be used at a scale much greater than that of a digital CT-scan image. This simple principle “to see a trend in a grain of sand” is the novel contribution of our work. Arguably, it can be implemented using images of drill cuttings collected within an interval in a well. We discuss examples of this principle for three basic rock types: high-porosity unconsolidated sand, medium-porosity sandstone, and medium-porosity carbonate.

Estimating dry rock frame moduli of high-resolution 3D digital rock images using the contact-mechanics-based effective medium approach

We have developed a method to estimate the dry frame elastic moduli of high-resolution 3D digital rock images using the contact-mechanics-based effective medium theory (EMT) model. The existing EMT models often are used to predict the effective dry frame elastic moduli of granular aggregates as a function of porosity, average number of contacts per grain, grain radius, contact radius, and contact stiffnesses of an elastic two-grain combination. But, it is almost impossible to measure the number of contacts per grain, contact radius distribution, or contact stiffness distribution in complex rocks. Therefore, explicit assumptions based on simplified microstructural geometries often are made to predict these contact properties in granular aggregates. As a result, the predictions of dry frame elastic moduli using EMT models may fail to match the observed properties because of numerous simplified assumptions, which can be violated in complex rocks. Our method uses the morphological contact properties (i.e., the grain-to-grain contact radius distribution, grain radius distribution, and coordination number distribution) directly extracted from 3D digital rock images to improve the prediction accuracy of dry frame elastic moduli using the EMT models. With integration of digital rocks technology, there is no longer a need to assume the size and shape of the grains, contact size, and number of contacts. The prediction accuracy of our method is validated on high-resolution 3D micro-CT digital rock images of miniplugs extracted from plugs with ultrasonic velocity measurements under dry conditions at different confining pressures. Image-computed dry frame elastic moduli using the EMT model are consistent with laboratory-measured moduli extrapolated to ambient conditions.

Прояв тенденцій музичного і хореографічного мистецтв ХХ - початку ХХІ ст. у творах за мотивами поем Ованеса Туманяна

The purpose of the study is to analyze the ballets and the dance fragment of the opera created on the basis of Hovhannes Tumanyan’s poems in the context of the trends in musical and choreographic art of the 20th and beginning of the 21st centuries; to identify the artistic means by which it is possible to embody the images and situations of a literary work in opera and ballet art. Methodology. The operas “Almast” by A. Spendiarov and “Salome” by R. Strauss were analyzed using the method of comparative musicology. An art history analysis of the ballets’ videos was carried out, their comparison with the poems of O. Tumanyan in order to identify the motives that formed the basis of their choreographic embodiment. The scientific novelty. For the first time, an art criticism analysis of ballet performances created on the basis of O. Tumanyan’s poems was carried out through the prism of the 20th and early 21st centuries main trends. (image of rock, visualization of an internal psychological split, interweaving of reality and science fiction, symbolism). Conclusions. The operas of A. Spendiarov “Almast” and R. Strauss “Salome”, in particular the fate of the main characters, have many parallels, which is due to the artistic trends of the late 19th and early 20th centuries. The image of rock (merciless fate), implied or embodied as a specific stage character, was most clearly manifested in A. Spendiarov’s opera Almast, in the dance and plastic theatrics “Anush Olmak” (Anush. To be) L. Taberyan and the ballet “Maro” directed by N. Megrabyan. In the ballet “Anush” directed by R. Kharatyan, the predetermined fate of Anush from above is less expressed. It is replaced by the inexorability of Adat (coming from the ancestors of the indestructible law). The visualization of the internal psychological split of the heroes – their black and white, evil and kind, angelic and diabolical forms, and the clash of conflicting emotions of their inner world are most clearly manifested in the image of Almast in the musical development of her image by A. Spendiarov. In the dance and plastic theatrics “Anush. Olmak” by L. Taberyan, the internal psychological split-bifurcation of the heroine is visualized, since the heroine has two scenic roles. The rock, the evil fate that gravitates over the heroine, is also represented in a concrete stage manner. Beliefs, the interweaving of reality and science fiction, are presented in the choreographic performances of N. Megrabyan “Sako Lori”, “Maro”. The folk rites are reproduced in “Anush. Olmak” (Anush. To be) by L. Taberyan. Symbolism is characteristic of the production of R. Kharatyan’s ballet “Anush” both at the musical and choreographic-production levels. The study proved a significant interpretative potential of the Hovhannes Tumanyan’s works for the opera and ballet art.

Classification of mineral components of granitoid rocks by using methods of digital petrography and machine learning

Machine learning methods have begun to be used in petrography relatively recently. However, thanks to the rapid programming development, more powerful algorithms and tools appear, the use of which to solve petrographic tasks hasn’t yet been considered. That’s why the purpose of this work was to use modern machine learning methods to identify mineral components from macro images of rock samples, as well as to use digital image processing methods. This article presents the method of determination of quantitative characteristics and the method of classification of minerals on macro images of rocks. An open source program for analyzing and processing images ImageJ, and its plugin Trainable Weka Segmentation were used as a toolkit. Macro images are obtained by scanning polished granite samples. Seven macro images of various representatives of the granites were selected for the experiment. Training with a teacher was conducted, where the decision tree method was used for classification. Based on this data set, classes were created for each of the rock-forming minerals: quartz (Q), potassium feldspar (Fps), plagioclase (Pl) and biotite (Bi). Regions of interest were prepared and stored in one database on the basis of which the classifier was trained. Based on the obtained classification data, masks of each mineral were created. A quantitative analysis was performed based on these masks: the percentage content and amount of grains of each mineral were determined. Results are presented in tabular and graphical forms.

Representation and meaning in rock art: the case of Algonquian rock images

ABSTRACT Over the past decade, discussions on ontology, animism and semiotics have generated numerous avenues of research on the meanings of rock images. In a context marked by the increasing recognition of Indigenous knowledges, a number of archaeologists have questioned representationalism and the idea that rock art has a fixed meaning in the past. In this paper, we discuss these new approaches with reference to the rock art of the Algonquian-speaking peoples of Central and Eastern Canada. We show how, for many Algonquian groups, rock art signified and continues to signify their experiences and practices in a continually emergent world. As such, we recognize that meaning within Algonquian relational ontologies is fluid without compromising the material-semiotic representations and intentions of the original creators. To illustrate this point, we examine the layers of significance associated with the rock art of the Canadian Shield and discuss a petroglyph site from the Canadian Maritimes.

Electromagnetic rock properties’ characterization and modeling using 3D micro-CT rock images

ABSTRACT This paper mainly focuses on the electromagnetic modeling aspect of the digital rock physics to understand the relationship between the micro-structures and contents within the structure to their electrical properties. This is achieved by using the 3D rock micro-CT images and performing electromagnetic modeling using the 3D numerical mixing law (finite difference method based), which is a robust numerical method with high efficiency. Based on the results of this investigation, one can establish the relationship between the rock structure/porous space-filling materials and the buck electrical properties, such as the permittivity and the conductivity. Two different 3D rock (sandstone and dacite) images with different image reconstruction processes were used to demonstrate the accuracy and efficiency of this method. Additionally, single-pole Debye model was also used to be compared with our results to further validate our proposed numerical method and also provide a potential approach to predict the oil saturation for future research study.

A Deep Learning Model for Quick and Accurate Rock Recognition with Smartphones

In the geological survey, the recognition and classification of rock lithology are an important content. The recognition method based on rock thin section leads to long recognition period and high recognition cost, and the recognition accuracy cannot be guaranteed. Moreover, the above method cannot provide an effective solution in the field. As a communication device with multiple sensors, smartphones are carried by most geological survey workers. In this paper, a smartphone application based on the convolutional neural network is developed. In this application, the phone’s camera can be used to take photos of rocks. And the types and lithology of rocks can be quickly and accurately identified in a very short time. This paper proposed a method for quickly and accurately recognizing rock lithology in the field. Based on ShuffleNet, a lightweight convolutional neural network used in deep learning, combined with the transfer learning method, the recognition model of the rock image was established. The trained model was then deployed to the smartphone. A smartphone application for identifying rock lithology was designed and developed to verify its usability and accuracy. The research results showed that the accuracy of the recognition model in this paper was 97.65% on the verification data set of the PC. The accuracy of recognition on the test data set of the smartphone was 95.30%, among which the average recognition time of the single sheet was 786 milliseconds, the maximum value was 1,045 milliseconds, and the minimum value was 452 milliseconds. And the single-image accuracy above 96% accounted for 95% of the test data set. This paper presented a new solution for the rapid and accurate recognition of rock lithology in field geological surveys, which met the needs of geological survey personnel to quickly and accurately identify rock lithology in field operations.

Research on Coal and Rock Image Recognition Based on 2DPCA Algorithm

The PCA algorithm in coal and rock recognition had some fundamental flaws such as large computation and low efficiency. In this paper, a coal and rock image recognition method based on 2DPCA algorithm was proposed. The experimental results showed that this method could not only reduce the dimension of the original feature, but also reduce the complicated calculation and improve the recognition accuracy effectively.

Wettability Quantification in Mixed-Wet Rocks Using a New NMR-Based Method

Summary Nuclear magnetic resonance (NMR) measurements have been attractive options for wettability characterization of reservoir rocks as they are sensitive to the type of fluid in contact with the grain surface. Several NMR-based wettability indices are documented in previous publications. Most of these methods require extensive calibration or involve complex inversion algorithms, which makes them computationally expensive and complicates their applicability in mixed-wet multimodal rocks. In this paper, we introduce a new NMR-based wettability index and verify its reliability in pore-scale and core-scale domains using numerical simulations and experimental measurements, respectively. This new index requires calibration at fully water-saturated water-wet and fully hydrocarbon-saturated hydrocarbon-wet states and can be applied to mixed-wet rocks at any fluid saturation level. This new NMR-based wettability index is a function of the transverse magnetization (T2) of mixed-wet rocks, the bulk relaxivity and saturation of each fluid, and the T2 distributions for fully water-wet and hydrocarbon-wet samples of the same rock type. The reliability of the new index was first tested in the pore-scale domain. For this part, we selected several pore-scale microcomputed tomography (CT) images of carbonate and sandstone rocks. We used a previously developed finite volume simulator to model the T2 responses in these images at fully water-wet and fully hydrocarbon-wet wettability states. Then we generated synthetic partially saturated mixed-wet samples and simulated T2 responses in these synthetic images. We used the simulated T2 results for determining their NMR-based wettability index and verified its applicability in the pore-scale domain. Next, we tested the reliability of the new NMR-based wettability index in the core-scale domain using NMR measurements in four Texas Cream (TC) rock samples, obtained from the Edwards formation. We altered the wettability of the cores to be water-, hydrocarbon-, and mixed-wet by injecting brine, a naphthenic acid in decane solution, or an anionic surfactant solution. We quantified the wettability of these samples using the Amott-Harvey (AH) index and contact angle measurements. Next, we measured the T2 distribution of these samples at different fluid saturation levels. Finally, we quantified the wettability values of these core samples using the new NMR-based index and compared them to those obtained from the AH index and contact angle measurements. We documented successful verification of the proposed method on samples with wettability ranging from −0.90 to 0.98 and from −0.6 to 0.5 (independently quantified using the AH method) in the pore- and core-scale domains, respectively. Results demonstrated that the new NMR-based wettability index reliably estimates the wettability of mixed-wet rocks in a wide range of wettability states. The new wettability index can potentially improve the speed and reliability of NMR-based wettability characterization and is promising for log-scale wettability assessment in mixed-wet rocks.

PoreFlow-Net: A 3D convolutional neural network to predict fluid flow through porous media

We present the PoreFlow-Net, a 3D convolutional neural network architecture that provides fast and accurate fluid flow predictions for 3D digital rock images. We trained our network to extract spatial relationships between the porous medium morphology and the fluid velocity field. Our workflow computes simple geometrical information from 3D binary images to train a deep neural network (the PoreFlow-Net) optimized to generalize the problem of flow through porous materials. Our results show that the extracted information is sufficient to obtain accurate flow field predictions in less than a second, without performing expensive numerical simulations providing a speed-up of several orders of magnitude. We also demonstrate that our model, trained with simple synthetic geometries, is able to provide accurate results in real samples spanning granular rocks, carbonates, and slightly consolidated media from a variety of subsurface formations, which highlights the ability of the model to generalize the porous media flow problem. The workflow presented here shows the successful application of a disruptive technology (physics-based training of machine learning models) to the digital rock physics community.

Super-resolution of real-world rock microcomputed tomography images using cycle-consistent generative adversarial networks.

Digital rock imaging plays an important role in studying the microstructure and macroscopic properties of rocks, where microcomputed tomography (MCT) is widely used. Due to the inherent limitations of MCT, a balance should be made between the field of view (FOV) and resolution of rock MCT images-a large FOV at low resolution (LR) or a small FOV at high resolution (HR). However, large FOV and HR are both expected for reliable analysis results in practice. Super-resolution (SR) is an effective solution to break through the mutual restriction between the FOV and resolution of rock MCT images, for it can reconstruct an HR image from a LR observation. Most of the existing SR methods cannot produce satisfactory HR results on real-world rock MCT images. One of the main reasons for this is that paired images are usually needed to learn the relationship between LR and HR rock images. However, it is challenging to collect such a dataset in a real scenario. Meanwhile, the simulated datasets may be unable to accurately reflect the model in actual applications. To address these problems, we propose a cycle-consistent generative adversarial network (CycleGAN)-based SR approach for real-world rock MCT images, namely, SRCycleGAN. In the off-line training phase, a set of unpaired rock MCT images is used to train the proposed SRCycleGAN, which can model the mapping between rock MCT images at different resolutions. In the on-line testing phase, the resolution of the LR input is enhanced via the learned mapping by SRCycleGAN. Experimental results show that the proposed SRCycleGAN can greatly improve the quality of simulated and real-world rock MCT images. The HR images reconstructed by SRCycleGAN show good agreement with the targets in terms of both the visual quality and the statistical parameters, including the porosity, the local porosity distribution, the two-point correlation function, the lineal-path function, the two-point cluster function, the chord-length distribution function, and the pore size distribution. Large FOV and HR rock MCT images can be obtained with the help of SRCycleGAN. Hence, this work makes it possible to generate HR rock MCT images that exceed the limitations of imaging systems on FOV and resolution.

Altered mineral segmentation in thin sections using an incremental-dynamic clustering algorithm

Intelligent mineral segmentation in thin section images of rocks still remains a challenging task in modern computational mineralogy. The objective of the paper is segmenting minerals in geological thin section’s images with special attention on altered mineral segmentation. In this paper, an efficient incremental-dynamic clustering algorithm is developed for segmentation of minerals in thin sections containing altered and non-altered minerals. In the clustering algorithm, there is no need for determining the number of clusters (minerals) existed in thin section images, and also it is able to deal with color changing and non-evident boundaries in altered minerals. We have solved two main existing limitations: segmentation of mineral pixels that are frequently labeled as background pixels, and segmentation of thin sections containing altered minerals. Moreover, we created an open database (Alborz Mineralogical Database), as a benchmark database in computational geosciences regarding image studies of mineral. The proposed method is validated based on the results provided by the segmentation maps, and experimental results indicate that the proposed method is very efficient and outperforms previous segmentation methods for altered minerals in thin section images. The proposed method can be applied in mining engineering, rock mechanics engineering, geotechnique engineering, mineralogy, petrography, and applications such as NASA’s Mars Rover Explorations (MRE).

Stochastic reconstruction of 3D porous media from 2D images using generative adversarial networks

Micro computed tomography (CT) provides petrophysics laboratories with the ability to image three dimensional porous media at pore scale. However, evaluating flow properties requires the acquisition of a large number of representative images, which is often unfeasible. Stochastic reconstruction methods are algorithms able to generate novel, realistic rock images from a small sample, thus avoiding a large acquisition process. A more convenient approach would use only two dimensional images, replacing 3D scans with images of the rock cuttings made during the drilling. This would extend the technique to media having pores smaller than the resolution of the micro-CT, but that can be imaged by microscopy.We introduce a novel method for 2D-to-3D reconstruction of the structure of porous media by applying generative adversarial neural networks. We compare several measures of pore morphology between simulated and acquired images. Experiments include bead pack, Berea sandstone, and Ketton limestone images. Results show that our GANs-based method can reconstruct three-dimensional images of porous media at different scales that are representative of the morphology of the original images. Also, compared to classical stochastic methods of image reconstruction, the generation of multiple images is much faster.

Training Deep Networks to Construct a Psychological Feature Space for a Natural-Object Category Domain

Many successful formal models of human categorization have been developed, but these models have been tested almost exclusively using artificial categories, because deriving psychological representations of large sets of natural stimuli using traditional methods such as multidimensional scaling (MDS) has been an intractable task. Here, we propose a novel integration in which MDS representations are used to train deep convolutional neural networks (CNNs) to automatically derive psychological representations for unlimited numbers of natural stimuli. In an example application, we train an ensemble of CNNs to produce the MDS coordinates of images of rocks, and we show that the ensemble can predict the MDS coordinates of new sets of rocks, even those not part of the original MDS space. We then show that the CNN-predicted MDS representations, unlike off-the-shelf CNN representations, can be used in conjunction with a formal psychological model to predict human categorization behavior. We further show that the CNNs can be trained to produce additional dimensions that extend the original MDS space and provide even better model fits to human category-learning data. Our integrated method provides a promising approach that can be instrumental in allowing researchers to extend traditional psychological-scaling and category-learning models to the complex, high-dimensional domains that exist in the natural world.

Autonomous Martian rock image classification based on transfer deep learning methods

In Mars exploration, rocks are good targets for compositional analysis with spectrometers. Their shape, size, and texture could provide a wealth of information for study of planetary geology. However, imitations on communications between Mars and earth lead to operations latencies and slow progress in planetary surface missions. Increasing the autonomy of rovers has become an important research direction. Autonomy is the ability to choose which scientific data to collect and which ones to send back to Earth. One of the aims is to recognize the rocks independently. The AEGIS system adopts the method of edge detection to select potential rock targets for following observation, but the type of rocks cannot be distinguished. Convolutional neural network (CNN) is getting more attention due to its performance in computer vision. However, a common issue of CNN is that it requires large amount of rock images for training, which are difficult to get. Transfer learning provides a good way to overcome the problem of lack of dataset. In this work, CNN based on vgg-16 architecture with deep transfer learning is used to automatically classify 4 groups of Martian rocks. The proposed model achieves accuracy of 100% on Martian rock images we collected from MSL Analyst ‘s Notebook. Moreover, a comparison between the VGG-16 transfer model and other models is made, and it can be found that the proposed model has the best performance in Martian rock classification.

Research on Intelligent Identification of Rock Types Based on Faster R-CNN Method

In the mining process of underground metal mines, the misjudgment of rock types by on-site technicians will have a serious negative impact on the stability evaluation of rock mass and the formulation of support schemes, which will result in the loss of economic benefits and potential safety hazards of mining enterprises. In order to realize the precise and intelligent identification of rock types, the image data of peridotite, basalt, marble, gneiss, conglomerate, limestone, granite, magnetite quartzite are amplified. Under the target detection framework of Faster R-CNN deep learning, the extraction network based on simplified VGG16 is used to extract and learn features of rock images, and finally the rock type identification system is successfully trained. The experimental verification shows that the system is correct for single-type rock image recognition and the accuracy is more than 96%. In order to realize accurate and intelligent identification of the surrounding rock surface under complex lithological conditions, the multi-type rocks hybrid images are also identified. The results show that the recognition effect is great and the accuracy rate is over 80%. Therefore, this system can accurately identify rock types with similar image features, which proves that the model has strong robustness and generalization ability. It has broad application prospects in rock mass stability evaluation and rock classification in underground mining.

Recognizing Multiple Types of Rocks Quickly and Accurately Based on Lightweight CNNs Model

The recognition and classification of rock lithology is an extremely important task of geological surveys. This paper proposes a new method for quickly identifying multiple types of rocks suitable for geological survey work field. Based on the two lightweight convolutional neural networks (CNNs), SqueezeNet and MobileNets, and combined with the transfer learning method, a rock lithology recognition model was established. The model was embedded into a smart phone for testing. This method was used to identify and classify the images of 28 kinds of rocks. Through a comprehensive comparison of the two models, the accuracy of SqueezeNet and MobileNets in the test dataset is 94.55% and 93.27%, respectively. Via the two models, the average recognition time of a single rock image is 557 and 836 milliseconds, and rock images with a recognition accuracy of over 96% accounted for 95% and 93% of the entire test dataset. Compared with the classification method based on rock thin section images, this method does not need to make rock thin sections. This paper meets the requirements of workers to quickly and accurately identify rocks in the work field, which improves the work efficiency and limits identification costs.