Article
Engineering, Industrial
Chayan Maiti, Sreekumar Muthuswamy
Summary: The purpose of this research is to automate material identification and classification task using image processing and machine learning techniques. A dataset of four material surfaces is generated and RGB data are extracted as input features. Convolutional Neural Network (CNN) and other classification methods are utilized to classify material images based on the dataset, achieving 100% accuracy in training and testing. The proposed methodology is demonstrated to accurately identify materials under various illumination environments and camera positions.
INTERNATIONAL JOURNAL OF PRODUCTION RESEARCH
(2023)
Article
Mathematics
Saurabh Agarwal, Hyenki Kim, Ki-Hyun Jung
Summary: In this paper, a new steganalysis scheme is proposed to detect stego-images. This scheme utilizes predefined kernels and a Leaky ReLU layer to improve detection performance, and it is verified to be more effective than existing steganalysis schemes on multiple steganography schemes.
Article
Computer Science, Information Systems
Pengfei Xian, Lai-Man Po, Yuzhi Zhao, Wing-Yin Yu, Kwok-Wai Cheung
Summary: This paper presents CLIP Driven Few-shot Panoptic Segmentation (CLIP-FPS), which leverages the knowledge of Contrastive Language-Image Pre-training (CLIP) model. The proposed method uses a center indexing attention mechanism to facilitate knowledge transfer and represents objects in an image as centers with pixel offsets. The model consists of a decoder for generating object center-offset groups and a self-attention module for producing a feature attention map. The results show that our method outperforms existing panoptic segmentation techniques in terms of Panoptic Quality (PQ) metrics.
Article
Computer Science, Artificial Intelligence
Dawid Polap, Marcin Wozniak
Summary: Modifying existing classifier operation models aims to enhance efficiency and reduce training time. The proposed method can implement various classifiers using federated learning and parallelism. Analysis and selection of the best classifier, as well as augmentation techniques, are important elements to improve operation in federated learning.
Article
Mathematics
Stefan Rohrmanstorfer, Mikhail Komarov, Felix Moedritscher
Summary: With the increasing amount of image data, automatic search and analysis of image information has become necessary. By studying and implementing different methods, features are successfully extracted from fashion data using convolutional neural networks and TensorFlow to build image classification models.
Article
Environmental Sciences
Jiaqing Zhang, Jie Lei, Weiying Xie, Daixun Li
Summary: The fusion of hyperspectral and LiDAR images is crucial for accurate classification and recognition in remote sensing. This study proposes a method that combines a binary convolutional neural network and a graph convolutional network with invariant attributes to overcome the challenges of constructing effective graph structures. The method utilizes a joint detection framework to simultaneously learn features from regular and irregular regions, resulting in an enhanced structural representation of the images. Experimental results demonstrate the superior performance of the proposed method in hyperspectral image analysis tasks.
Article
Materials Science, Multidisciplinary
Xiaolong Pei, Yu hong Zhao, Liwen Chen, Qingwei Guo, Zhiqiang Duan, Yue Pan, Hua Hou
Summary: Appropriate image preprocessing can enhance machine learning performance. However, the robustness of machine learning to different preprocessing methods in micrograph datasets with notable sample differences remains unexplored. In this study, we collected hundreds of optical micrographs with variations in color, contrast, size, brightness, and tensile strength. Through various preprocessing techniques, including color transformation, size adjustment, normalization, and image enhancement, we established classification and prediction models using transfer learning and the VGG16 model. The results demonstrated comparable accuracy between grayscale and color micrographs in classification models, with different preprocessing methods yielding varying degrees of improvement in accuracy and coefficient of determination (R2) in the regression models.
MATERIALS & DESIGN
(2023)
Article
Automation & Control Systems
Dawid Polap, Marta Wlodarczyk-Sielicka, Natalia Wawrzyniak
Summary: This paper presents an alternative approach to automatic image analysis using various artificial intelligence techniques. The system utilizes collaborative learning to continually increase its knowledge during operation, and it benefits from quick implementation and a small database.
Article
Geochemistry & Geophysics
Guoqing Zhou, Weiguang Liu, Qiang Zhu, Yanling Lu, Yu Liu
Summary: In recent years, models based on fully convolutional neural networks have been proposed to improve accuracy but ignored computational efficiency. This research presents an innovative deep learning model, ECA-MobileNetV3(large)+SegNet, which simultaneously considers both aspects. By modifying the encoder and decoder structures, the proposed model achieves significant improvement in performance and reduces the number of parameters.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Computer Science, Information Systems
Ben Chen, Feiwei Qin, Yanli Shao, Jin Cao, Yong Peng, Ruiquan Ge
Summary: The study proposes a novel method for diagnosing leukemia by classifying white blood cells in bone marrow using the WBC-GLAformer model. The model combines the features of convolutional neural networks and transformers to enrich the features and improve classification accuracy by selecting discriminative regions.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2023)
Article
Ecology
Diego Andre Sant'Ana, Marcio Carneiro Brito Pache, Jose Martins, Gilberto Astolfi, Wellington Pereira Soares, Sebastiao Lucas Neves de Melo, Natalia da Silva Heimbach, Vanessa Aparecida de Moraes Weber, Rodrigo Goncalves Mateus, Hemerson Pistori
Summary: This paper presents an experiment comparing four different convolutional neural network architectures for classifying sheep superpixels. The results show that DenseNet201 is the best architecture for this task, providing a method for segmenting mixed-breed sheep images and supporting non-invasive methods for animal tracking and weight prediction.
ECOLOGICAL INFORMATICS
(2022)
Article
Biotechnology & Applied Microbiology
Hui Cai, Shihan Shan, Xiaoping Wang
Summary: This article introduces a method for rapidly detecting plankton using a convolutional neural network, and the results show that this method has higher speed and accuracy compared to traditional methods.
ALGAL RESEARCH-BIOMASS BIOFUELS AND BIOPRODUCTS
(2022)
Article
Computer Science, Artificial Intelligence
Gao Huang, Zhuang Liu, Geoff Pleiss, Laurens van der Maaten, Kilian Q. Weinberger
Summary: Recent work has shown that adding shorter connections in convolutional networks can make the network deeper, more accurate, and more efficient in training. This paper introduces Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward manner. DenseNets alleviate the vanishing-gradient problem, encourage feature reuse, and improve parameter efficiency, leading to significant improvements in object recognition tasks.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2022)
Article
Automation & Control Systems
Yasir Ali Farrukh, Syed Wali, Irfan Khan, Nathaniel D. Bastian
Summary: The exponential growth of internet and inter-connectivity has led to an increase in network traffic and new attacks that pose challenges to network security. To address this, a new approach called SeNet-I is proposed, which leverages computer vision capabilities to develop high-level representations of network traffic using a deep concatenated convolutional neural network model. The experimental results show that the proposed method outperforms existing methods in network intrusion detection.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2023)
Article
Chemistry, Multidisciplinary
Francesco de Gioia, Luca Fanucci
Summary: This study introduces a novel data-driven model for demosaicing in digital imaging, taking into account the different requirements for reconstruction of image Luma and Chrominance channels. The model is a parallel composition of two reconstruction networks with individual architecture and trained with distinct loss functions. To address overfitting, a dataset containing patches with common chromatic and spectral characteristics was prepared.
APPLIED SCIENCES-BASEL
(2021)
Article
Materials Science, Multidisciplinary
Arun Baskaran, Elizabeth J. Kautz, Aritra Chowdhary, Wufei Ma, Bulent Yener, Daniel J. Lewis
Summary: This work reviews the application of IDML to materials characterization, defines a hierarchy of six action steps to compartmentalize problems, and evaluates studies through the decisions made in these steps for a granular assessment of the field. The importance of interpretability and explainability is discussed, along with an overview of two emerging techniques in the field.
Correction
Materials Science, Multidisciplinary
A. D. Boccardo, M. Tong, S. B. Leen, D. Tourret, J. Segurado
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Tao Li, Qing Hou, Jie-chao Cui, Jia-hui Yang, Ben Xu, Min Li, Jun Wang, Bao-qin Fu
Summary: This study investigates the thermal and defect properties of AlN using molecular dynamics simulation, and proposes a new method for selecting interatomic potentials, developing a new model. The developed model demonstrates high computational accuracy, providing an important tool for modeling thermal transport and defect evolution in AlN-based devices.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Shin-Pon Ju, Chao-Chuan Huang, Hsing-Yin Chen
Summary: Amorphous boron nitride (a-BN) is a promising ultralow-dielectric-constant material for interconnect isolation in integrated circuits. This study establishes a deep learning potential (DLP) for different forms of boron nitride and uses molecular dynamics simulations to investigate the mechanical behaviors of a-BN. The results reveal the structure-property relationships of a-BN, providing useful insights for integrating it in device applications.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
M. Salman, S. Schmauder
Summary: Shape memory polymer foams (SMPFs) are lightweight cellular materials that can recover their undeformed shape through external stimulation. Reinforcing the material with nano-clay filler improves its physical properties. Multiscale modeling techniques can be used to study the thermomechanical response of SMPFs and show good agreement with experimental results.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Laura Gueci, Francesco Ferrante, Marco Bertini, Chiara Nania, Dario Duca
Summary: This study investigates the acidity of 30 Bronsted sites in the beta-zeolite framework and compares three computational methods. The results show a wide range of deprotonation energy values, and the proposed best method provides accurate calculations.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
K. A. Lopes Lima, L. A. Ribeiro Junior
Summary: Advancements in nanomaterial synthesis and characterization have led to the discovery of new carbon allotropes, including biphenylene network (BPN). The study finds that BPN lattices with a single-atom vacancy exhibit higher CO2 adsorption energies than pristine BPN. Unlike other 2D carbon allotropes, BPN does not exhibit precise CO2 sensing and selectivity by altering its band structure configuration.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Jay Kumar Sharma, Arpita Dhamija, Anand Pal, Jagdish Kumar
Summary: In this study, the quaternary Heusler alloys LiAEFeSb were investigated for their crystal structure, electronic properties, and magnetic behavior. Density functional theory calculations revealed that LiSrFeSb and LiBaFeSb exhibit half-metallic band structure and 100% spin polarization, making them excellent choices for spintronic applications.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Roman A. Eremin, Innokentiy S. Humonen, Alexey A. Kazakov, Vladimir D. Lazarev, Anatoly P. Pushkarev, Semen A. Budennyy
Summary: Computational modeling of disordered crystal structures is essential for studying composition-structure-property relations. In this work, the effects of Cd and Zn substitutions on the structural stability of CsPbI3 were investigated using DFT calculations and GNN models. The study achieved accurate energy predictions for structures with high substitution contents, and the impact of data subsampling on prediction quality was comprehensively studied. Transfer learning routines were also tested, providing new perspectives for data-driven research of disordered materials.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Zhixin Sun, Hang Dong, Yaohui Yin, Ai Wang, Zhen Fan, Guangyong Jin, Chao Xin
Summary: In this study, the crystal structure, electronic structure, and optical properties of KH2PO4: KDP crystals under different pressures were investigated using the generalized gradient approximate. It was found that high pressure caused a phase transition in KDP and greatly increased the band gap. The results suggest that high pressure enhances the compactness of KDP and improves the laser damage threshold.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Tingting Yu
Summary: This study presents atomistic simulations revealing that an increase in driving force may result in slower grain boundary movement and switches in the mode of grain boundary shear coupling migration. Shear coupling behavior is found to effectively alleviate stress and holds potential for stress relaxation and microstructure manipulation in materials.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Y. Zhang, X. Q. Deng, Q. Jing, Z. S. Zhang
Summary: The electronic properties of C2N/antimonene van der Waals heterostructure are investigated using density functional theory. The results show that by applying horizontal strain, vertical strain, electric field, and interlayer twist, the electronic structure can be adjusted. Additionally, the band alignment and energy states of the heterostructure can be significantly changed by applying vertical strain on the twisted structure. These findings are important for controlling the electronic properties of heterostructures.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Chad E. Junkermeier, Evan Larmand, Jean-Charles Morais, Jedediah Kobebel, Kat Lavarez, R. Martin Adra, Jirui Yang, Valeria Aparicio Diaz, Ricardo Paupitz, George Psofogiannakis
Summary: This study investigates the adsorption properties of carbon dioxide (CO2), methane (CH4), and dihydrogen (H2) in carbophenes functionalized with different groups. The results show that carbophenes can be promising adsorbents for these gases, with high adsorption energies and low desorption temperatures. The design and combination of functional groups can further enhance their adsorption performance.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Y. Borges, L. Huber, H. Zapolsky, R. Patte, G. Demange
Summary: Grain boundary structure is closely related to solute atom segregation, and machine learning can predict the segregation energy density. The study provides a fresh perspective on the relationship between grain boundary structure and segregation properties.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
M. R. Jones, L. T. W. Fey, I. J. Beyerlein
Summary: In this work, a three-dimensional ab-initio informed phase-field-dislocation dynamics model combined with Langevin dynamics is used to investigate glide mechanisms of edge and screw dislocations in Nb at finite temperatures. It is found that the screw dislocation changes its mode of glide at two distinct temperatures, which coincides with the thermal insensitivity and athermal behavior of Nb yield strengths.
COMPUTATIONAL MATERIALS SCIENCE
(2024)
Article
Materials Science, Multidisciplinary
Joshua A. Vita, Dallas R. Trinkle
Summary: This study introduces a new machine learning model framework that combines the simplicity of spline-based potentials with the flexibility of neural network architectures. The simplified version of the neural network potential can efficiently describe complex datasets and explore the boundary between classical and machine learning models. Using spline filters for encoding atomic environments results in interpretable embedding layers that can incorporate expected physical behaviors and improve interpretability through neural network modifications.
COMPUTATIONAL MATERIALS SCIENCE
(2024)