Pub Date : 2025-07-04DOI: 10.1016/j.aiig.2025.100140
Vinh V. Le , HongGiang Nguyen , Nguyen Huu Ngu
This paper presents a deep learning architecture combined with exploratory data analysis to estimate maximum wall deflection in deep excavations. Six major geotechnical parameters were studied. Statistical methods, such as pair plots and Pearson correlation, highlighted excavation depth (correlation coefficient = 0.82) as the most significant factor. For method prediction, five deep learning models (CNN, LSTM, BiLSTM, CNN-LSTM, and CNN-BiLSTM) were built. The CNN-BiLSTM model excelled in training performance (R2 = 0.98, RMSE = 0.02), while BiLSTM reached superior testing results (R2 = 0.85, RMSE = 0.06), suggesting greater generalization ability. Based on the feature importance analysis from model weights, excavation depth, stiffness ratio, and bracing spacing were ranked as the highest contributors. This point verified a lack of prediction bias on residual plots and high model agreement with measured values on Taylor diagrams (correlation coefficient 0.92). The effectiveness of integrated techniques was reliably assured for predicting wall deformation. This approach facilitates more accurate and efficient geotechnical design and provides engineers with improved tools for risk evaluation and decision-making in deep excavation projects.
{"title":"Deep learning approaches for estimating maximum wall deflection in excavations with inconsistent clay stratigraphy","authors":"Vinh V. Le , HongGiang Nguyen , Nguyen Huu Ngu","doi":"10.1016/j.aiig.2025.100140","DOIUrl":"10.1016/j.aiig.2025.100140","url":null,"abstract":"<div><div>This paper presents a deep learning architecture combined with exploratory data analysis to estimate maximum wall deflection in deep excavations. Six major geotechnical parameters were studied. Statistical methods, such as pair plots and Pearson correlation, highlighted excavation depth (correlation coefficient = 0.82) as the most significant factor. For method prediction, five deep learning models (CNN, LSTM, BiLSTM, CNN-LSTM, and CNN-BiLSTM) were built. The CNN-BiLSTM model excelled in training performance (R<sup>2</sup> = 0.98, RMSE = 0.02), while BiLSTM reached superior testing results (R<sup>2</sup> = 0.85, RMSE = 0.06), suggesting greater generalization ability. Based on the feature importance analysis from model weights, excavation depth, stiffness ratio, and bracing spacing were ranked as the highest contributors. This point verified a lack of prediction bias on residual plots and high model agreement with measured values on Taylor diagrams (correlation coefficient 0.92). The effectiveness of integrated techniques was reliably assured for predicting wall deformation. This approach facilitates more accurate and efficient geotechnical design and provides engineers with improved tools for risk evaluation and decision-making in deep excavation projects.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 2","pages":"Article 100140"},"PeriodicalIF":0.0,"publicationDate":"2025-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144581267","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-30DOI: 10.1016/j.aiig.2025.100142
Baoling Gui, Anshuman Bhardwaj, Lydia Sam
Rapid urbanization and land-use changes are placing immense pressure on resources, infrastructure, and environmental sustainability. To address these, accurate urban simulation models are essential for sustainable development and governance. Among them, Cellular Automata (CA) models have become key tools for predicting urban expansion, optimizing land-use planning, and supporting data-driven decision-making. This review provides a comprehensive examination of the development of urban cellular automata (UCA) models, presenting a new framework to enhance individual UCA sub-modules within the context of emerging technologies, sustainable environments, and public governance. By addressing gaps in prior UCA modelling reviews—particularly in the integration and optimization of UCA sub-module technologies—this framework is designed to simplify UCA model understanding and development. We systematically review pioneering case studies, deconstruct current UCA operational processes, and explore modern technologies, such as big data and artificial intelligence, to optimize these sub-modules further. We discuss current limitations within UCA models and propose future pathways, emphasizing the necessity of comprehensive analyses for effective UCA simulations. Proposed solutions include strengthening our understanding of urban growth mechanisms, examining spatial positioning and temporal evolution dynamics, and enhancing urban geographic simulations with deep learning techniques to support sustainable transitions in public governance. These improvements offer data-driven decision support for environmental management, advancing policies that foster sustainable urban development.
{"title":"Cellular automata models for simulation and prediction of urban land use change: Development and prospects","authors":"Baoling Gui, Anshuman Bhardwaj, Lydia Sam","doi":"10.1016/j.aiig.2025.100142","DOIUrl":"10.1016/j.aiig.2025.100142","url":null,"abstract":"<div><div>Rapid urbanization and land-use changes are placing immense pressure on resources, infrastructure, and environmental sustainability. To address these, accurate urban simulation models are essential for sustainable development and governance. Among them, Cellular Automata (CA) models have become key tools for predicting urban expansion, optimizing land-use planning, and supporting data-driven decision-making. This review provides a comprehensive examination of the development of urban cellular automata (UCA) models, presenting a new framework to enhance individual UCA sub-modules within the context of emerging technologies, sustainable environments, and public governance. By addressing gaps in prior UCA modelling reviews—particularly in the integration and optimization of UCA sub-module technologies—this framework is designed to simplify UCA model understanding and development. We systematically review pioneering case studies, deconstruct current UCA operational processes, and explore modern technologies, such as big data and artificial intelligence, to optimize these sub-modules further. We discuss current limitations within UCA models and propose future pathways, emphasizing the necessity of comprehensive analyses for effective UCA simulations. Proposed solutions include strengthening our understanding of urban growth mechanisms, examining spatial positioning and temporal evolution dynamics, and enhancing urban geographic simulations with deep learning techniques to support sustainable transitions in public governance. These improvements offer data-driven decision support for environmental management, advancing policies that foster sustainable urban development.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 2","pages":"Article 100142"},"PeriodicalIF":0.0,"publicationDate":"2025-06-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144518220","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-24DOI: 10.1016/j.aiig.2025.100139
Paul Theophily Nsulangi , Werneld Egno Ngongi , John Mbogo Kafuku , Guan Zhen Liang
This study compared the predictive performance and processing speed of an artificial neural network (ANN) and a hybrid of a numerical reservoir simulation (NRS) and artificial neural network (NRS-ANN) models in estimating the oil production rate of the ZH86 reservoir block under waterflood recovery. The historical input variables: reservoir pressure, reservoir pore volume containing hydrocarbons, reservoir pore volume containing water and reservoir water injection rate used as inputs for ANN models. To create the NRS-ANN hybrid models, 314 data sets extracted from the NRS model, which included reservoir pressure, reservoir pore volume containing hydrocarbons, reservoir pore volume containing water and reservoir water injection rate were used. The output of the models was the historical oil production rate (HOPR in m3 per day) recorded from the ZH86 reservoir block. Models were developed using MATLAB R2021a and trained with 25 models in three replicate conditions (2, 4 and 6), each at 1000 epochs. A comparative analysis indicated that, for all 25 models, the ANN outperformed the NRS-ANN in terms of processing speed and prediction performance. ANN models achieved an average of R2 and MAE of 0.8433 and 8.0964 m3/day values, respectively, while NRS-ANN hybrid models achieved an average of R2 and MAE of 0.7828 and 8.2484 m3/day values, respectively. In addition, ANN models achieved a processing speed of 49 epochs/sec, 32 epochs/sec, and 24 epochs/sec after 2, 4, and 6 replicates, respectively. Whereas the NRS-ANN hybrid models achieved lower average processing speeds of 45 epochs/sec, 23 epochs/sec and 20 epochs/sec. In addition, the ANN optimal model outperforms the NRS-ANN model in terms of both processing speed and accuracy. The ANN optimal model achieved a speed of 336.44 epochs/sec, compared to the NRS-ANN hybrid optimal model, which achieved a speed of 52.16 epochs/sec. The ANN optimal model achieved lower RMSE and MAE values of 7.9291 m3/day and 5.3855 m3/day in the validation dataset compared with the hybrid ANS optimal model, which achieved 13.6821 m3/day and 9.2047 m3/day, respectively. The study also showed that the ANN optimal model consistently achieved higher R2 values: 0.9472, 0.9284 and 0.9316 in the training, test and validation data sets. Whereas the NRS-ANN hybrid optimal yielded lower R2 values of 0.8030, 0.8622 and 0.7776 for the training, testing and validation datasets. The study showed that ANN models are a more effective and reliable tool, as they balance both processing speed and accuracy in estimating the oil production rate of the ZH86 reservoir block under the waterflooding recovery method.
{"title":"Comparison of processing speed of NRS-ANN hybrid and ANN models for oil production rate estimation of reservoir under waterflooding","authors":"Paul Theophily Nsulangi , Werneld Egno Ngongi , John Mbogo Kafuku , Guan Zhen Liang","doi":"10.1016/j.aiig.2025.100139","DOIUrl":"10.1016/j.aiig.2025.100139","url":null,"abstract":"<div><div>This study compared the predictive performance and processing speed of an artificial neural network (ANN) and a hybrid of a numerical reservoir simulation (NRS) and artificial neural network (NRS-ANN) models in estimating the oil production rate of the ZH86 reservoir block under waterflood recovery. The historical input variables: reservoir pressure, reservoir pore volume containing hydrocarbons, reservoir pore volume containing water and reservoir water injection rate used as inputs for ANN models. To create the NRS-ANN hybrid models, 314 data sets extracted from the NRS model, which included reservoir pressure, reservoir pore volume containing hydrocarbons, reservoir pore volume containing water and reservoir water injection rate were used. The output of the models was the historical oil production rate (HOPR in m<sup>3</sup> per day) recorded from the ZH86 reservoir block. Models were developed using MATLAB R2021a and trained with 25 models in three replicate conditions (2, 4 and 6), each at 1000 epochs. A comparative analysis indicated that, for all 25 models, the ANN outperformed the NRS-ANN in terms of processing speed and prediction performance. ANN models achieved an average of R<sup>2</sup> and MAE of 0.8433 and 8.0964 m<sup>3</sup>/day values, respectively, while NRS-ANN hybrid models achieved an average of R<sup>2</sup> and MAE of 0.7828 and 8.2484 m<sup>3</sup>/day values, respectively. In addition, ANN models achieved a processing speed of 49 epochs/sec, 32 epochs/sec, and 24 epochs/sec after 2, 4, and 6 replicates, respectively. Whereas the NRS-ANN hybrid models achieved lower average processing speeds of 45 epochs/sec, 23 epochs/sec and 20 epochs/sec. In addition, the ANN optimal model outperforms the NRS-ANN model in terms of both processing speed and accuracy. The ANN optimal model achieved a speed of 336.44 epochs/sec, compared to the NRS-ANN hybrid optimal model, which achieved a speed of 52.16 epochs/sec. The ANN optimal model achieved lower RMSE and MAE values of 7.9291 m<sup>3</sup>/day and 5.3855 m<sup>3</sup>/day in the validation dataset compared with the hybrid ANS optimal model, which achieved 13.6821 m<sup>3</sup>/day and 9.2047 m<sup>3</sup>/day, respectively. The study also showed that the ANN optimal model consistently achieved higher R<sup>2</sup> values: 0.9472, 0.9284 and 0.9316 in the training, test and validation data sets. Whereas the NRS-ANN hybrid optimal yielded lower R<sup>2</sup> values of 0.8030, 0.8622 and 0.7776 for the training, testing and validation datasets. The study showed that ANN models are a more effective and reliable tool, as they balance both processing speed and accuracy in estimating the oil production rate of the ZH86 reservoir block under the waterflooding recovery method.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 2","pages":"Article 100139"},"PeriodicalIF":0.0,"publicationDate":"2025-06-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144653505","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-23DOI: 10.1016/j.aiig.2025.100128
Junfei Zhang , Huisheng Cheng , Ninghui Sun , Zehui Huo , Junlin Chen
Ternary geopolymers incorporating multiple solid wastes such as steel slag (SS), fly ash (FA), and granulated blast furnace slag (GBFS) are considered environmentally friendly and exhibit enhanced performance. However, the mechanisms governing strength development and the design of optimal mixtures are not fully understood due to the complexity of their components. This study presents the development of four machine learning models—Artificial Neural Network (ANN), Support Vector Regression (SVR), Extremely Randomized Tree (ERT), and Gradient Boosting Regression (GBR)—for predicting the unconfined compressive strength (UCS) of ternary geopolymers. The models were trained using a dataset comprising 120 mixtures derived from laboratory tests. Shapley Additive Explanations analysis was employed to interpret the machine learning models and elucidate the influence of different components on the properties of ternary geopolymers. The results indicate that ANN exhibits the highest predictive accuracy for UCS (R = 0.949). Furthermore, the UCS of ternary geopolymers is most sensitive to the content of GBFS. This study provides valuable insights for optimizing the mix proportions in ternary blended geopolymer mixtures.
{"title":"Interpretable machine learning models for evaluating strength of ternary geopolymers","authors":"Junfei Zhang , Huisheng Cheng , Ninghui Sun , Zehui Huo , Junlin Chen","doi":"10.1016/j.aiig.2025.100128","DOIUrl":"10.1016/j.aiig.2025.100128","url":null,"abstract":"<div><div>Ternary geopolymers incorporating multiple solid wastes such as steel slag (SS), fly ash (FA), and granulated blast furnace slag (GBFS) are considered environmentally friendly and exhibit enhanced performance. However, the mechanisms governing strength development and the design of optimal mixtures are not fully understood due to the complexity of their components. This study presents the development of four machine learning models—Artificial Neural Network (ANN), Support Vector Regression (SVR), Extremely Randomized Tree (ERT), and Gradient Boosting Regression (GBR)—for predicting the unconfined compressive strength (UCS) of ternary geopolymers. The models were trained using a dataset comprising 120 mixtures derived from laboratory tests. Shapley Additive Explanations analysis was employed to interpret the machine learning models and elucidate the influence of different components on the properties of ternary geopolymers. The results indicate that ANN exhibits the highest predictive accuracy for UCS (R = 0.949). Furthermore, the UCS of ternary geopolymers is most sensitive to the content of GBFS. This study provides valuable insights for optimizing the mix proportions in ternary blended geopolymer mixtures.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 2","pages":"Article 100128"},"PeriodicalIF":0.0,"publicationDate":"2025-06-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144523230","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-03DOI: 10.1016/j.aiig.2025.100126
Andrey V. Soromotin , Dmitriy A. Martyushev , João Luiz Junho Pereira
Permeability is one of the main oil reservoir characteristics. It affects potential oil production, well-completion technologies, the choice of enhanced oil recovery methods, and more. The methods used to determine and predict reservoir permeability have serious shortcomings. This article aims to refine and adapt machine learning techniques using historical data from hydrocarbon field development to evaluate and predict parameters such as the skin factor and permeability of the remote reservoir zone. The article analyzes data from 4045 wells tests in oil fields in the Perm Krai (Russia). An evaluation of the performance of different Machine Learning (ML) algorithms in the prediction of the well permeability is performed. Three different real datasets are used to train more than 20 machine learning regressors, whose hyperparameters are optimized using Bayesian Optimization (BO). The resulting models demonstrate significantly better predictive performance compared to traditional methods and the best ML model found is one that never was applied before to this problem. The permeability prediction model is characterized by a high R2 adjusted value of 0.799. A promising approach is the integration of machine learning methods and the use of pressure recovery curves to estimate permeability in real-time. The work is unique for its approach to predicting pressure recovery curves during well operation without stopping wells, providing primary data for interpretation. These innovations are exclusive and can improve the accuracy of permeability forecasts. It also reduces well downtime associated with traditional well-testing procedures. The proposed methods pave the way for more efficient and cost-effective reservoir development, ultimately supporting better decision-making and resource optimization in oil production.
{"title":"On the application of machine learning algorithms in predicting the permeability of oil reservoirs","authors":"Andrey V. Soromotin , Dmitriy A. Martyushev , João Luiz Junho Pereira","doi":"10.1016/j.aiig.2025.100126","DOIUrl":"10.1016/j.aiig.2025.100126","url":null,"abstract":"<div><div>Permeability is one of the main oil reservoir characteristics. It affects potential oil production, well-completion technologies, the choice of enhanced oil recovery methods, and more. The methods used to determine and predict reservoir permeability have serious shortcomings. This article aims to refine and adapt machine learning techniques using historical data from hydrocarbon field development to evaluate and predict parameters such as the skin factor and permeability of the remote reservoir zone. The article analyzes data from 4045 wells tests in oil fields in the Perm Krai (Russia). An evaluation of the performance of different Machine Learning (ML) algorithms in the prediction of the well permeability is performed. Three different real datasets are used to train more than 20 machine learning regressors, whose hyperparameters are optimized using Bayesian Optimization (BO). The resulting models demonstrate significantly better predictive performance compared to traditional methods and the best ML model found is one that never was applied before to this problem. The permeability prediction model is characterized by a high R<sup>2</sup> adjusted value of 0.799. A promising approach is the integration of machine learning methods and the use of pressure recovery curves to estimate permeability in real-time. The work is unique for its approach to predicting pressure recovery curves during well operation without stopping wells, providing primary data for interpretation. These innovations are exclusive and can improve the accuracy of permeability forecasts. It also reduces well downtime associated with traditional well-testing procedures. The proposed methods pave the way for more efficient and cost-effective reservoir development, ultimately supporting better decision-making and resource optimization in oil production.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 2","pages":"Article 100126"},"PeriodicalIF":0.0,"publicationDate":"2025-06-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144330611","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-01DOI: 10.1016/j.aiig.2025.100120
M. Giselle Fernández-Godino , Wai Tong Chung , Akshay A. Gowardhan , Matthias Ihme , Qingkai Kong , Donald D. Lucas , Stephen C. Myers
High-resolution spatiotemporal simulations effectively capture the complexities of atmospheric plume dispersion in complex terrain. However, their high computational cost makes them impractical for applications requiring rapid responses or iterative processes, such as optimization, uncertainty quantification, or inverse modeling. To address this challenge, this work introduces the Dual-Stage Temporal Three-dimensional UNet Super-resolution (DST3D-UNet-SR) model, a highly efficient deep learning model for plume dispersion predictions. DST3D-UNet-SR is composed of two sequential modules: the temporal module (TM), which predicts the transient evolution of a plume in complex terrain from low-resolution temporal data, and the spatial refinement module (SRM), which subsequently enhances the spatial resolution of the TM predictions. We train DST3D-UNet-SR using a comprehensive dataset derived from high-resolution large eddy simulations (LES) of plume transport. We propose the DST3D-UNet-SR model to significantly accelerate LES of three-dimensional (3D) plume dispersion by three orders of magnitude. Additionally, the model demonstrates the ability to dynamically adapt to evolving conditions through the incorporation of new observational data, substantially improving prediction accuracy in high-concentration regions near the source.
{"title":"A staged deep learning approach to spatial refinement in 3D temporal atmospheric transport","authors":"M. Giselle Fernández-Godino , Wai Tong Chung , Akshay A. Gowardhan , Matthias Ihme , Qingkai Kong , Donald D. Lucas , Stephen C. Myers","doi":"10.1016/j.aiig.2025.100120","DOIUrl":"10.1016/j.aiig.2025.100120","url":null,"abstract":"<div><div>High-resolution spatiotemporal simulations effectively capture the complexities of atmospheric plume dispersion in complex terrain. However, their high computational cost makes them impractical for applications requiring rapid responses or iterative processes, such as optimization, uncertainty quantification, or inverse modeling. To address this challenge, this work introduces the Dual-Stage Temporal Three-dimensional UNet Super-resolution (DST3D-UNet-SR) model, a highly efficient deep learning model for plume dispersion predictions. DST3D-UNet-SR is composed of two sequential modules: the temporal module (TM), which predicts the transient evolution of a plume in complex terrain from low-resolution temporal data, and the spatial refinement module (SRM), which subsequently enhances the spatial resolution of the TM predictions. We train DST3D-UNet-SR using a comprehensive dataset derived from high-resolution large eddy simulations (LES) of plume transport. We propose the DST3D-UNet-SR model to significantly accelerate LES of three-dimensional (3D) plume dispersion by three orders of magnitude. Additionally, the model demonstrates the ability to dynamically adapt to evolving conditions through the incorporation of new observational data, substantially improving prediction accuracy in high-concentration regions near the source.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 1","pages":"Article 100120"},"PeriodicalIF":0.0,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144185115","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-01DOI: 10.1016/j.aiig.2025.100118
Stalyn Paucar, Christian Mejia-Escobar, Victor Collaguazo
The identification and characterization of rock types is a core activity in geology and related fields, including mining, petroleum, environmental science, industry, and construction. Traditionally, this task is performed by human specialists who analyze and describe the type, composition, texture, shape, and other properties of rock samples, whether collected in-situ or prepared in a laboratory. However, the process is subjective, dependent on the specialist’s experience, and time-consuming. This study proposes an artificial intelligence-based approach that combines computer vision and natural language processing to generate both textual and verbal descriptions from images of rock thin sections. A dataset of images and corresponding textual descriptions is used to train a hybrid deep learning model. Features extracted from the images using EfficientNetB7 are processed by a Transformer network to generate textual descriptions, which are then converted into verbal responses using a speech synthesis service. The experimental results show an accuracy of 0.892 and a BLEU score of 0.71. This model offers potential utility for research, professional, and academic applications and has been deployed as a web application for public use.
{"title":"Automatic description of rock thin sections: A web application","authors":"Stalyn Paucar, Christian Mejia-Escobar, Victor Collaguazo","doi":"10.1016/j.aiig.2025.100118","DOIUrl":"10.1016/j.aiig.2025.100118","url":null,"abstract":"<div><div>The identification and characterization of rock types is a core activity in geology and related fields, including mining, petroleum, environmental science, industry, and construction. Traditionally, this task is performed by human specialists who analyze and describe the type, composition, texture, shape, and other properties of rock samples, whether collected in-situ or prepared in a laboratory. However, the process is subjective, dependent on the specialist’s experience, and time-consuming. This study proposes an artificial intelligence-based approach that combines computer vision and natural language processing to generate both textual and verbal descriptions from images of rock thin sections. A dataset of images and corresponding textual descriptions is used to train a hybrid deep learning model. Features extracted from the images using EfficientNetB7 are processed by a Transformer network to generate textual descriptions, which are then converted into verbal responses using a speech synthesis service. The experimental results show an accuracy of 0.892 and a BLEU score of 0.71. This model offers potential utility for research, professional, and academic applications and has been deployed as a web application for public use.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 1","pages":"Article 100118"},"PeriodicalIF":0.0,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144212094","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-01DOI: 10.1016/j.aiig.2025.100123
Omar M. Saad , Matteo Ravasi , Tariq Alkhalifah
Seismic data denoising is a critical process usually applied at various stages of the seismic processing workflow, as our ability to mitigate noise in seismic data affects the quality of our subsequent analyses. However, finding an optimal balance between preserving seismic signals and effectively reducing seismic noise presents a substantial challenge. In this study, we introduce a multi-stage deep learning model, trained in a self-supervised manner, designed specifically to suppress seismic noise while minimizing signal leakage. This model operates as a patch-based approach, extracting overlapping patches from the noisy data and converting them into 1D vectors for input. It consists of two identical sub-networks, each configured differently. Inspired by the transformer architecture, each sub-network features an embedded block that comprises two fully connected layers, which are utilized for feature extraction from the input patches. After reshaping, a multi-head attention module enhances the model’s focus on significant features by assigning higher attention weights to them. The key difference between the two sub-networks lies in the number of neurons within their fully connected layers. The first sub-network serves as a strong denoiser with a small number of neurons, effectively attenuating seismic noise; in contrast, the second sub-network functions as a signal-add-back model, using a larger number of neurons to retrieve some of the signal that was not preserved in the output of the first sub-network. The proposed model produces two outputs, each corresponding to one of the sub-networks, and both sub-networks are optimized simultaneously using the noisy data as the label for both outputs. Evaluations conducted on both synthetic and field data demonstrate the model’s effectiveness in suppressing seismic noise with minimal signal leakage, outperforming some benchmark methods.
{"title":"Self-supervised multi-stage deep learning network for seismic data denoising","authors":"Omar M. Saad , Matteo Ravasi , Tariq Alkhalifah","doi":"10.1016/j.aiig.2025.100123","DOIUrl":"10.1016/j.aiig.2025.100123","url":null,"abstract":"<div><div>Seismic data denoising is a critical process usually applied at various stages of the seismic processing workflow, as our ability to mitigate noise in seismic data affects the quality of our subsequent analyses. However, finding an optimal balance between preserving seismic signals and effectively reducing seismic noise presents a substantial challenge. In this study, we introduce a multi-stage deep learning model, trained in a self-supervised manner, designed specifically to suppress seismic noise while minimizing signal leakage. This model operates as a patch-based approach, extracting overlapping patches from the noisy data and converting them into 1D vectors for input. It consists of two identical sub-networks, each configured differently. Inspired by the transformer architecture, each sub-network features an embedded block that comprises two fully connected layers, which are utilized for feature extraction from the input patches. After reshaping, a multi-head attention module enhances the model’s focus on significant features by assigning higher attention weights to them. The key difference between the two sub-networks lies in the number of neurons within their fully connected layers. The first sub-network serves as a strong denoiser with a small number of neurons, effectively attenuating seismic noise; in contrast, the second sub-network functions as a signal-add-back model, using a larger number of neurons to retrieve some of the signal that was not preserved in the output of the first sub-network. The proposed model produces two outputs, each corresponding to one of the sub-networks, and both sub-networks are optimized simultaneously using the noisy data as the label for both outputs. Evaluations conducted on both synthetic and field data demonstrate the model’s effectiveness in suppressing seismic noise with minimal signal leakage, outperforming some benchmark methods.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 1","pages":"Article 100123"},"PeriodicalIF":0.0,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144243298","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-01DOI: 10.1016/j.aiig.2025.100114
{"title":"Thank you reviewers!","authors":"","doi":"10.1016/j.aiig.2025.100114","DOIUrl":"10.1016/j.aiig.2025.100114","url":null,"abstract":"","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 1","pages":"Article 100114"},"PeriodicalIF":0.0,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144366260","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-06-01DOI: 10.1016/j.aiig.2025.100127
M.A. Dalhat, Sami A. Osman
This study employed convolutional neural networks (CNNs) for the classification of rock minerals based on 3179 RGB-scale original microstructural images of undisturbed broken surfaces. The image dataset covers 40 distinct rock mineral-types. Three CNN architectures (Simple model, SqueezeNet, and Xception) were evaluated to compare their performance and feature extraction capabilities. Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to visualize the features influencing model predictions, providing insights into how each model distinguishes between mineral classes. Key discriminative attributes included texture, grain size, pattern, and color variations. Texture and grain boundaries were identified as the most critical features, as they were strongly activated regions by the best model. Patterns such as banding and chromatic contrasts further enhanced classification accuracy. Performance analysis revealed that the Simple model had limited ability to isolate fine-grained details, producing broad and less specific activations (0.84 test accuracy). SqueezeNet demonstrated improved localization of discriminative features but occasionally missed finer textural details (0.95 test accuracy). The Xception model outperformed the others, achieving the highest classification accuracy (0.98 test accuracy) by exhibiting precise and tightly focused activations, capturing intricate textures and subtle chromatic variations. Its superior performance can be attributed to its deep architecture and efficient depth-wise separable convolutions, which enabled hierarchical and detailed feature extraction. Results underscores the importance of texture, pattern, and chromatic features in accurate mineral classification and highlights the suitability of deep, efficient architectures like Xception for such tasks. These findings demonstrate the potential of CNNs in geoscience research, offering a framework for automated mineral identification in industrial and scientific applications.
{"title":"Deep learning based identification of rock minerals from un-processed digital microscopic images of undisturbed broken-surfaces","authors":"M.A. Dalhat, Sami A. Osman","doi":"10.1016/j.aiig.2025.100127","DOIUrl":"10.1016/j.aiig.2025.100127","url":null,"abstract":"<div><div>This study employed convolutional neural networks (CNNs) for the classification of rock minerals based on 3179 RGB-scale original microstructural images of undisturbed broken surfaces. The image dataset covers 40 distinct rock mineral-types. Three CNN architectures (Simple model, SqueezeNet, and Xception) were evaluated to compare their performance and feature extraction capabilities. Gradient-weighted Class Activation Mapping (Grad-CAM) was employed to visualize the features influencing model predictions, providing insights into how each model distinguishes between mineral classes. Key discriminative attributes included texture, grain size, pattern, and color variations. Texture and grain boundaries were identified as the most critical features, as they were strongly activated regions by the best model. Patterns such as banding and chromatic contrasts further enhanced classification accuracy. Performance analysis revealed that the Simple model had limited ability to isolate fine-grained details, producing broad and less specific activations (0.84 test accuracy). SqueezeNet demonstrated improved localization of discriminative features but occasionally missed finer textural details (0.95 test accuracy). The Xception model outperformed the others, achieving the highest classification accuracy (0.98 test accuracy) by exhibiting precise and tightly focused activations, capturing intricate textures and subtle chromatic variations. Its superior performance can be attributed to its deep architecture and efficient depth-wise separable convolutions, which enabled hierarchical and detailed feature extraction. Results underscores the importance of texture, pattern, and chromatic features in accurate mineral classification and highlights the suitability of deep, efficient architectures like Xception for such tasks. These findings demonstrate the potential of CNNs in geoscience research, offering a framework for automated mineral identification in industrial and scientific applications.</div></div>","PeriodicalId":100124,"journal":{"name":"Artificial Intelligence in Geosciences","volume":"6 1","pages":"Article 100127"},"PeriodicalIF":0.0,"publicationDate":"2025-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144255175","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号