Article
Geochemistry & Geophysics
Cao Song, Wenkai Lu, Yuqing Wang, Songbai Jin, Jinliang Tang, Lei Chen
Summary: Reservoir prediction is a significant issue in seismic interpretation. This study proposes a semisupervised deep-learning framework using a closed-loop CNN and virtual well-logging labels to improve the accuracy and spatial continuity of the predicted reservoir.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Claudia Hulbert, Romain Jolivet, Blandine Gardonio, Paul A. Johnson, Christopher X. Ren, Bertrand Rouet-Leduc
Summary: This study presents a new methodology to improve the detection and location of hidden tremors within seismic noise. By combining convolutional neural networks and neural network attribution, the researchers were able to extract core tremor signatures and accurately locate the source of tremors. This approach allows for the identification of small signals hidden within the noise and can locate more tremors than existing catalogs.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Energy & Fuels
Xuliang Liu, Wenshu Zha, Daolun Li, Xiang Li, Luhang Shen
Summary: This paper proposes an automatic well test interpretation method based on one-dimensional convolutional neural network (1D CNN) for circular reservoir with changing wellbore storage. Compared with two-dimensional convolutional neural network (2D CNN), 1D CNN significantly reduces the computational complexity and time cost. The effectiveness of this method is proved by the field data in Daqing oilfield.
JOURNAL OF ENERGY RESOURCES TECHNOLOGY-TRANSACTIONS OF THE ASME
(2023)
Article
Automation & Control Systems
Stefano Peluchetti, Stefano Favaro
Summary: Modern neural networks with a large number of layers and units per layer have shown remarkable performance. Studies reveal the undesired properties of independent and identically initialized networks as depth increases, but solutions such as shrinking distributions and establishing interplays with stochastic differential equations have been proposed. Additionally, investigations into the expressiveness of infinitely deep ResNets under different settings highlight the importance of network initialization and depth in training.
JOURNAL OF MACHINE LEARNING RESEARCH
(2021)
Review
Geosciences, Multidisciplinary
Yu An, Haiwen Du, Siteng Ma, Yingjie Niu, Dairui Liu, Jing Wang, Yuhan Du, Conrad Childs, John Walsh, Ruihai Dong
Summary: Automated seismic fault interpretation has been an active area of research, and Deep learning (DL) methods have shown promising results in this field since 2018. However, the lack of a reasonable summary of these methods has made it difficult for researchers to understand the current development process. To fill this gap, a systematic review of DL-based fault interpretation literature published between 2012 and 2022 was conducted, and 73 seismic datasets from 56 articles were summarized. The study reported the benefits of using DL for fault interpretation, but also identified challenges hindering its integration into industrial workflows, such as the lack of sufficient annotated data.
EARTH-SCIENCE REVIEWS
(2023)
Article
Computer Science, Interdisciplinary Applications
Ji Zhang, Duarte Nuno Vieira, Qi Cheng, Yongzheng Zhu, Kaifei Deng, Jianhua Zhang, Zhiqiang Qin, Qiran Sun, Tianye Zhang, Kaijun Ma, Xiaofeng Zhang, Ping Huang
Summary: Diatom testing is important in forensic medicine for drowning diagnosis, but manual identification of diatoms in samples is time-consuming and labor-intensive. This study introduces DiatomNet v1.0, a software that can automatically identify diatom frustules in whole slides. The software's performance is improved through optimization with limited new datasets.
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
(2023)
Article
Environmental Sciences
Qiyuan Yang, Xianmin Wang, Xinlong Zhang, Jianping Zheng, Yu Ke, Lizhe Wang, Haixiang Guo
Summary: Massive earthquakes often result in thousands of coseismic landslides. The automatic recognition of these landslides is crucial for post-earthquake emergency rescue, landslide risk mitigation, and city reconstruction. This study proposes a novel semantic segmentation network, EGCN, to improve the accuracy of landslide identification.
Article
Energy & Fuels
Xuliang Liu, Wenshu Zha, Zhankui Qi, Daolun Li, Yan Xing, Lei He
Summary: Well test analysis is crucial in monitoring reservoir performance. This paper proposes an intelligent reservoir model identification method using convolutional neural network (CNN), which improves classification accuracy and achieves good results.
JOURNAL OF ENERGY RESOURCES TECHNOLOGY-TRANSACTIONS OF THE ASME
(2022)
Article
Energy & Fuels
Sihan Yang, Qiguo Liu, Xiaoping Li, Youjie Xu
Summary: This study proposes an automated framework for well test model identification using syntactic pattern recognition. The framework consists of six steps and can effectively address the non-uniqueness of different reservoir models. The findings of this study are important for better understanding the process of well test experts in completing the task of model identification.
PETROLEUM SCIENCE AND TECHNOLOGY
(2022)
Article
Computer Science, Artificial Intelligence
Lei Yang, Shuai Xu, Junfeng Fan, En Li, Yanhong Liu
Summary: Defect detection is crucial for manufacturing and processing applications, but it faces challenges in weak-texture and low-contrast industrial environments. Deep learning has shown excellent performance in defect identification, but it has limitations in feature processing and temporal modeling. To address these issues, a pixel-level deep segmentation network is proposed, combining a residual attention network for effective feature representation and a bidirectional ConvLSTM block for better propagation of context features. Experimental results demonstrate that this network outperforms other state-of-the-art methods.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Dermatology
Jiachen Sun, Lin Fu, Wen Zhang, Dongjie Li, Ming Zhang, Zineng Xu, Hailong Bai, Peng Ding
Summary: The study developed a CNN-based deep learning method for automatic diagnosis and graduation of skin frostbite. The approach demonstrated higher accuracy and efficiency compared to two residents from the burns department.
INTERNATIONAL WOUND JOURNAL
(2023)
Article
Health Care Sciences & Services
Albert T. Young, Kristen Fernandez, Jacob Pfau, Rasika Reddy, Nhat Anh Cao, Max Y. von Franque, Arjun Johal, Benjamin V. Wu, Rachel R. Wu, Jennifer Y. Chen, Raj P. Fadadu, Juan A. Vasquez, Andrew Tam, Michael J. Keiser, Maria L. Wei
Summary: This study systematically assessed the performance of artificial intelligence models, specifically convolutional neural networks, on real-world non-curated images using computational stress tests. Results showed inconsistent predictions on images subjected to repeated capture or simple transformations, indicating the need for further validation of models meeting conventionally reported metrics.
NPJ DIGITAL MEDICINE
(2021)
Article
Computer Science, Artificial Intelligence
Hareem Ayesha, Sajid Iqbal, Mehreen Tariq, Muhammad Abrar, Muhammad Sanaullah, Ishaq Abbas, Amjad Rehman, Muhammad Farooq Khan Niazi, Shafiq Hussain
Summary: Automatic natural language interpretation of medical images is a challenging field that combines computer vision and natural language processing. This comprehensive review discusses the methods, performance, strengths, limitations of recent research in medical image captioning, providing recommendations for the future.
PATTERN RECOGNITION
(2021)
Article
Construction & Building Technology
Sheng Zhang, Xinling Deng, Yumin Lu, Shaozheng Hong, Zhengyi Kong, Yongli Peng, Ye Luo
Summary: This paper proposes a Channel Attention based Metallic Corrosion Detection method (CAMCD) that can automatically detect corroded regions with multiple distinct levels. By embedding SE blocks into the deep residual network and highlighting important features of various corroded regions using learned weights of channel attentions, the CAMCD network shows superior performance in discriminating patch-wise features among different corrosion levels, as validated by experimental results on a collected metallic corrosion dataset and visualizations of feature maps.
JOURNAL OF BUILDING ENGINEERING
(2021)
Article
Energy & Fuels
Peng Dong, Zhiming Chen, Xinwei Liao, Wei Yu
Summary: The study utilizes a one-dimensional convolutional neural network to build an automatic interpretation model for well test data, showing high accuracy and reliability in curve classification and parameter inversion compared to traditional artificial neural networks and two-dimensional convolutional neural networks, with faster training speed.
JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING
(2022)