Article
Computer Science, Information Systems
Ran Shao, Xiao-Jun Bi
Summary: This paper proposes a hybrid model that combines Transformer and CNN to improve the classification ability of Transformers on small datasets. By introducing more convolution operations, the model achieves state-of-the-art results on 4 small datasets, opening up new paths for the application of Transformers on small datasets.
Article
Computer Science, Artificial Intelligence
Xu Wang, Xiaoming Chen, Yanping Wang
Summary: With the popularization of intelligent transportation systems, the demand for vision-based algorithms and performance increases. Vehicle detection techniques have advanced, but face challenges, leading to the proposal of a method using GAN to generate high-resolution images.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Computer Science, Information Systems
Weihao Weng, Xin Zhu, Lei Jing, Mianxiong Dong
Summary: This paper introduces a novel architecture, smooth attention branch (SAB), that simplifies the understanding of long-range pixel-pixel dependencies in small-scale biomedical image segmentation. SAB is a modified attention operation that implements a subnetwork using reshaped feature maps rather than directly calculating a softmax value for attention scores. SAB fuses multilayer attentive feature maps to learn visual attention in multilevel features.
Article
Mathematics
Jose E. Valdez-Rodriguez, Edgardo M. Felipe-Riveron, Hiram Calvo
Summary: Early detection and treatment of glaucoma are crucial to prevent vision loss. A CNN-based methodology focusing on the extraction of optic disc information from retinal images achieved high accuracy in glaucoma diagnosis. Further exploration with different RGB channels and depth information did not significantly improve classification performance.
Article
Computer Science, Information Systems
Qingping Sun, Yi Xiao, Jie Zhang, Shizhe Zhou, Chi-Sing Leung, Xin Su
Summary: This study proposes a hybrid model to estimate a 3D human body mesh by combining model-based and model-free approaches. It first utilizes a convolutional neural network (CNN) to generate a coarse human mesh, and then refines the mesh's vertex coordinates using a graph convolutional neural network (GCN). Experimental results demonstrate that this hybrid model outperforms existing methods in human reconstruction.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Information Systems
Lorenzo Brigato, Bjoern Barz, Luca Iocchi, Joachim Denzler
Summary: This article addresses the research status and issues in image classification with small datasets, and proposes a systematic overview and a shared benchmark to facilitate objective comparisons and further research in the field.
Article
Engineering, Civil
Tianjie Zhang, Donglei Wang, Amanda Mullins, Yang Lu
Summary: In this study, an integrated APC-GAN and AttuNet framework is proposed for the automated pixel-level segmentation of pavement surface cracks. The proposed APC-GAN is designed as an image augmentation method and the AttuNet structure incorporates an attention module into the convolutional network. The performance of the framework is evaluated using the DeepCrack dataset, which contains only 300 training images. The results show that APC-GAN outperforms DCGAN and traditional image augmentation methods in generating clear and diverse pavement images, and the proposed framework achieves the highest performance metrics compared to other models.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2023)
Article
Engineering, Multidisciplinary
Mia C. Morrell, Kyle Hickmann, Brandon M. Wilson
Summary: Particle image velocimetry (PIV) is an effective tool for extracting flow fields, and convolutional neural networks (CNNs) have been successfully applied to PIV analysis with uncertainty quantification. By comparing different Bayesian CNN models, it is found that utilizing interrogation region cross-correlation maps as inputs improves performance. Additionally, the best performing BCNN shows promising results on both synthetic and real image pairs from the 1st International PIV Challenge, demonstrating potential for future applications in multi-pass PIV algorithms.
MEASUREMENT SCIENCE AND TECHNOLOGY
(2021)
Article
Computer Science, Artificial Intelligence
Andre Luiz C. Ottoni, Raphael M. de Amorim, Marcela S. Novo, Dayana B. Costa
Summary: This paper discusses the importance and challenges of using deep learning methods in building construction image classification, and proposes a method for tuning data augmentation hyperparameters to improve classification accuracy. Experimental results show that the recommended hyperparameter configuration achieved high accuracy in both case studies.
INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS
(2023)
Article
Biology
Guy M. Hagen, Justin Bendesky, Rosa Machado, Tram-Anh Nguyen, Tanmay Kumar, Jonathan Ventura
Summary: Fluorescence microscopy is important in biological research, but photobleaching and phototoxicity are limiting factors. Machine learning methods can improve signal-to-noise ratio and reduce phototoxicity. High-quality data is essential for training deep learning methods.
Article
Chemistry, Analytical
Emmanuel Pintelas, Ioannis E. Livieris, Panagiotis E. Pintelas
Summary: The study introduces a novel approach using a convolutional autoencoder topological model to address the issue of noise and redundant information affecting deep learning models, leading to a significant performance improvement by compressing and filtering initial high-dimensional input images.
Article
Biochemical Research Methods
Yuansong Zeng, Zhuoyi Wei, Weijiang Yu, Rui Yin, Yuchen Yuan, Bingling Li, Zhonghui Tang, Yutong Lu, Yuedong Yang
Summary: The rapid development of spatial transcriptomics enables the prediction of gene expression from histology images. Hist2ST, a deep learning-based model, is developed to predict RNA-seq expression by capturing the relationships between 2D vision features and the spatial dependency between spots. Comprehensive tests show that Hist2ST outperforms existing methods in gene expression prediction and spatial region identification.
BRIEFINGS IN BIOINFORMATICS
(2022)
Article
Computer Science, Information Systems
Dongyang Zhang, Jie Shao, Zhenwen Liang, Lianli Gao, Heng Tao Shen
Summary: This study introduces a cascaded super-resolution convolutional neural network (CSRCNN) to address the aliasing artifacts and high computational costs caused by existing methods that use interpolation during the beginning stage. Experimental results show that the proposed network achieves superior performance, especially with an 8x upsampling factor.
IEEE TRANSACTIONS ON MULTIMEDIA
(2021)
Article
Plant Sciences
Giovanni Galli, Felipe Sabadin, Rafael Massahiro Yassue, Cassia Galves, Humberto Fanelli Carvalho, Jose Crossa, Osval Antonio Montesinos-Lopez, Roberto Fritsche-Neto
Summary: This study evaluates the performance of MLP and CNN in genomic prediction and determines the best models through a case study. MLP and CNN show competitive results for both regression and classification tasks, and provide new insights into automated machine learning for genomic prediction.
FRONTIERS IN PLANT SCIENCE
(2022)
Article
Genetics & Heredity
Zheyu Niu, Xin Gao, Zhaozhi Xia, Shuchao Zhao, Hongrui Sun, Heng Wang, Meng Liu, Xiaohan Kong, Chaoqun Ma, Huaqiang Zhu, Hengjun Gao, Qinggong Liu, Faji Yang, Xie Song, Jun Lu, Xu Zhou
Summary: miRNAs play a crucial role in various biological processes and human diseases, and are considered as therapeutic targets for small molecules. In order to predict novel SM-miRNA associations, we propose a miRNA and small molecule association prediction model (GCNNMMA) based on ensemble learning, graph neural networks (GNNs), and convolutional neural networks (CNNs). Experimental results show that GCNNMMA outperforms other comparison models in cross-validation tests on two different datasets.
FRONTIERS IN GENETICS
(2023)
Article
Chemistry, Multidisciplinary
Shichao Peng, Yiwei Xie, Linying Wang, Wenjuan Liu, Hua Li, Zhaochao Xu, Mao Ye, Zhongmin Liu
Summary: This study investigates the molecular transport in single zeolite crystals using super-resolution structured illumination microscopy. The diffusion behaviors on the center and surface planes of the crystals were used to monitor the diversity of surface barriers. This provides a new perspective for studying the molecular transport mechanisms and origins of surface barriers in nanoporous materials.
ANGEWANDTE CHEMIE-INTERNATIONAL EDITION
(2022)
Article
Engineering, Environmental
Anqi Li, Shuanghe Meng, Kai Huang, Wuqiang Yang, Mao Ye
Summary: In this study, the effects of concentration models on the electrical capacitance tomography (ECT) measurements of gas-fluidized beds were experimentally investigated. The results showed that the concentration models can significantly influence the measured solid concentrations and the threshold selection for identifying bubbles and emulsion phase in gas-fluidized beds. A new concentration model based on a power function was proposed, and the optimal exponent alpha was determined through experimental fitting.
CHEMICAL ENGINEERING JOURNAL
(2022)
Article
Engineering, Chemical
Jibin Zhou, Mao Ye, Zhongmin Liu
Summary: The minimum fluidization velocity (Umf), as a significant parameter in fluidized beds design and operation, has received great attention. This study evaluates empirical correlations with four different formulae for predicting Umf based on a newly established database. Modifications to these correlations have been made to achieve satisfactory performance in predicting Umf for different particle types.
CHEMICAL ENGINEERING SCIENCE
(2022)
Article
Engineering, Chemical
Deyang Gao, Xue Li, Baolin Hou, Fang Lu, Mao Ye, Aiqin Wang, Xiaodong Wang
Summary: This paper investigates the behavior of bubbles in high-viscosity systems using high-speed imaging. The results show that the liquid viscosity plays a crucial role in determining the shape and dynamics of bubbles. The correlation between bubble diameter and shape for the relevant viscous system is obtained by fitting the experimental results. This study is of great importance for optimizing the design of relevant industrial processes.
CHEMICAL ENGINEERING SCIENCE
(2022)
Article
Engineering, Chemical
Mingbin Gao, Hua Li, Junyi Yu, Mao Ye, Zhongmin Liu
Summary: This study proposes a quantitative principle for the shape-selectivity of zeolite catalysis and validates it using methanol-to-hydrocarbons (MTH) as a model. By combining molecular simulations and infrared imaging, the essence of zeolite shape-selective catalysis is revealed. This study provides a method for searching for zeolite catalysts with higher selectivity.
Article
Engineering, Environmental
Chengxiu Wang, Zhihui Li, Zeneng Sun, Xingying Lan, Jinsen Gao, Mao Ye, Jesse Zhu
Summary: A two-step Otsu's method was used to analyze the gas-solids flow field in a TFB at different axial levels. The results showed that the solids holdup increased with increasing H0/D ratio, while the mass and volume fraction of cluster clouds decreased. The study also extracted parameters related to cluster cores and found that clusters had equal chances to move towards the center and the wall of the TFB at a low H0/D ratio with high Ug.
CHEMICAL ENGINEERING JOURNAL
(2023)
Article
Engineering, Environmental
Chengxiu Wang, Mengjie Luo, Xin Su, Xingying Lan, Zeneng Sun, Jinsen Gao, Mao Ye, Jesse Zhu
Summary: This study proposes a new signal processing method to identify particle clusters in CFB risers using a sliding window and a non-linear threshold curve. By considering the fluctuation of the gas-solids flow, a more realistic characterization of the clusters in HDCFB and LDCFB is obtained.
CHEMICAL ENGINEERING JOURNAL
(2023)
Article
Engineering, Chemical
Jinqiang Liang, Danzhu Liu, Shuliang Xu, Mao Ye
Summary: Light olefins, such as ethylene and propylene, are crucial chemical materials. This study introduces the Olefins Production Security Index (OPSI) as a metric for evaluating the techno-economic and security performance of different production routes. The results reveal that coal-based routes (CMTO and CMTP) exhibit higher security ranks and favorable economic performance compared to natural gas-based routes (NFTO) and deep catalytic cracking (DCC). Carbon dioxide-based routes (CDMTO and CDMTP) have lower emissions but higher costs. CMTO, CMTP, and natural gas-based routes (NMTO) achieve the top OPSI grades, indicating their competitiveness in terms of security. Sensitivity analysis identifies the security shortcomings of each production route. This framework aids in selecting the most suitable production route for light olefins, prioritizing security considerations.
CHEMICAL ENGINEERING RESEARCH & DESIGN
(2023)
Article
Engineering, Chemical
Likun Ma, Sina Kashanj, Xue Li, Shuliang Xu, David S. Nobes, Mao Ye
Summary: Porous particles, commonly used in chemical reactors, have not been extensively studied in terms of their flow characteristics and drag coefficients at the single particle level. This study investigates the flow around a porous cube using particle image velocimetry, revealing an enlarged recirculation region and higher drag coefficients compared to a solid cube. These findings are attributed to the strong push effect of penetration flow and the transition of external flow to internal flow through the porous cube.
CHEMICAL ENGINEERING SCIENCE
(2023)
Article
Engineering, Chemical
Kai Huang, Shuanghe Meng, Tao Zhang, Mao Ye, Wuqiang Yang, Zhongmin Liu
Summary: This study investigated the transition of fluidization behavior from Geldart B particles to A particles induced by temperature change using a developed high-temperature electrical capacitance tomography (ECT) sensor. Silica particles with a Sauter mean diameter of 237 μm and density of 2650 kg/m^3, typically Geldart B particles under ambient conditions, were fluidized in a 5.5 cm column from 20 to 600 degrees C. The ECT measurements showed that with increasing temperature, there was a decrease in minimum bubbling velocity (U(mb)), no bed expansion characteristic in the homogeneous fluidization regime, an absence of multiple-bubbles regime, and a larger bubble size. The pressure drop against superficial gas velocity curves at elevated temperatures confirmed homogeneous fluidization between the minimum fluidization velocity (U(mf)) and U(mb). Our analysis demonstrated that cohesive interparticle forces, which increase linearly with temperature, are responsible for the fluidization behavior transition of the silica particles.
INDUSTRIAL & ENGINEERING CHEMISTRY RESEARCH
(2023)
Article
Multidisciplinary Sciences
Jiamin Yuan, Mingbin Gao, Zhiqiang Liu, Xiaomin Tang, Yu Tian, Gang Ma, Mao Ye, Anmin Zheng
Summary: The ultrafast transport of adsorbates in confined spaces is a goal pursued by scientists. However, diffusion will be generally slower in nano-channels, as confined spaces inhibit motion. Here, researchers demonstrate that the movement of long-chain molecules increases with a decrease in pore size, indicating that confined spaces promote transport. Inspired by a hyperloop running on a railway, they establish a superfast pathway for molecules in zeolites with nano-channels. This unique hyperloop-like diffusion offers insights into molecule diffusion under confinement and has potential applications in selecting efficient catalysts for rapid transport in the industrial field.
NATURE COMMUNICATIONS
(2023)
Article
Chemistry, Physical
Dan Zhao, Mingbin Gao, Xinxin Tian, Dmitry E. Doronkin, Shanlei Han, Jan-Dierk Grunwaldt, Uwe Rodemerck, David Linke, Mao Ye, Guiyuan Jiang, Haijun Jiao, Evgenii V. Kondratenko
Summary: This study investigated the effects of diffusion on the performance of ZnOx catalysts in propane and isobutane dehydrogenation reactions. Molecular dynamics simulations showed that mass transport limitations do not play a significant role in the PDH reaction, but do affect the iBDH reaction. X-ray absorption spectroscopy revealed that the nature of active ZnOx sites depends on the support material.
Article
Green & Sustainable Science & Technology
Danzhu Liu, Jinqiang Liang, Shuliang Xu, Mao Ye
Summary: According to the traditional principle of producer responsibility, national or regional carbon emissions do not take into account the embodied carbon emissions caused by inter-regional trade in product consumption. This study uses a multiregional input-output table and carbon emission data to analyze China's domestic regional carbon flow network and identifies three clusters of carbon emission characteristic regions using the k-means clustering algorithm of machine learning. It is found that certain provinces, such as Beijing, Zhejiang, and Guangdong, are net input areas for embodied carbon emissions, transferring responsibility for carbon emissions through trade. Carbon emissions accounting on a worldwide/countrywide scale should consider both production responsibility and trade income. This research provides a novel approach to national or regional classification based on embodied carbon emissions and calls for an equitable regional distribution system of carbon emission rights. Inter-regional cooperation is crucial for achieving carbon neutrality, with economically developed regions offering assistance to improve energy efficiency or optimize energy structures in less developed regions through capital investment and technology transformation.
Article
Chemistry, Applied
Shanfan Lin, Yuchun Zhi, Wenna Zhang, Xiaoshuai Yuan, Chengwei Zhang, Mao Ye, Shutao Xu, Yingxu Wei, Zhongmin Liu
Summary: Using a colorimetric analysis method, the production of formaldehyde (HCHO) in methanol-to-olefins (MTO) and dimethyl ether (DME)-to-olefins (DTO) reactions over SAPO-34 was quantitatively monitored. It was found that HCHO was present in both slight and conspicuous amounts at the initial and deactivation stages of the reactions, as well as when co-fed with water or high-pressure H2. The study also revealed the weaker hydrogen transfer ability of DME compared to methanol, leading to a more prominent olefins-based cycle and suppression of reactant-induced hydrogen transfer and deactivation in DTO.
CHINESE JOURNAL OF CATALYSIS
(2023)
Article
Biotechnology & Applied Microbiology
Mingbin Gao, Mao Ye, Zhongmin Liu
Summary: This study examines the importance of energy management in catalyst design and catalysis process development, and explores the impact of temperature on solid-based catalysis. By analyzing innovative heating techniques and temperature-monitoring technologies, the study demonstrates the potential for transformative industrial catalytic processes.
CURRENT OPINION IN CHEMICAL ENGINEERING
(2023)