Article
Computer Science, Artificial Intelligence
Bin Cao, Yuqi Liu, Chenyu Hou, Jing Fan, Baihua Zheng, Jianwei Yin
Summary: This paper proposes ATEC, a solution that can efficiently find a preferable hyperplane by automatically tuning the error cost for between-class samples. ATEC distinguishes itself from all existing parameter tuning strategies by evaluating the effectiveness of error costs and changing them in the right direction if they are not effective.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2021)
Article
Mathematics, Interdisciplinary Applications
Shaofei Zang, Xinghai Li, Jianwei Ma, Yongyi Yan, Jinfeng Lv, Yuan Wei
Summary: This paper proposes a new variant of ELM called DELM-CDMA for unsupervised domain adaptation. It introduces CDMA into the hidden layer of ELM to reduce distribution discrepancy between domains and eliminate domain bias. It also adopts LDA to improve the discrimination of the learned model. Experimental results show that DELM-CDMA effectively extends ELM and outperforms other domain adaptation approaches.
Article
Geochemistry & Geophysics
Wei Wei, Songzheng Xu, Lei Zhang, Jinyang Zhang, Yanning Zhang
Summary: This research proposes a novel deep learning network that utilizes both labeled and unlabeled data for training, aiming to address overfitting caused by inaccurate labels in hyperspectral image classification. By exploiting the unsupervised structure knowledge in unlabeled data, the proposed method improves the accuracy of conventional supervised classification.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Giannis Lantzanakis, Zina Mitraka, Nektarios Chrysoulakis
Summary: Accurate land cover mapping of the Earth's surface using Earth observation data, especially in urban areas, is a challenging task due to the large spectral variability of man-made structures and the mixed pixel phenomenon. Support vector machines (SVMs) are commonly used for classification, but there is no rule of thumb for choosing optimal parameters in classifying satellite imagery, requiring a time-consuming trial-and-error process. Proposed advancements in the C-SVC algorithm aim to improve its performance and reduce manual parameterization.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2021)
Article
Geochemistry & Geophysics
Bing Tu, Chengle Zhou, Xiaolong Liao, Guoyun Zhang, Yishu Peng
Summary: The study introduces a novel spatial-spectral classification method for hyperspectral images based on structural-kernel collaborative representation (SKCR), which considers a weak assumption of spatial neighborhood. The method utilizes superpixel segmentation and dual kernels to achieve excellent classification performance even with relatively small training samples.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2021)
Article
Computer Science, Artificial Intelligence
Qiang Zhou, Wen'an Zhou, Shirui Wang
Summary: The paper introduces a novel domain adaptation method, Cluster adaptation Networks (CAN), to reduce domain shift, preserve the data structure in the feature space, and facilitate classification in the target domain. Experiments on various datasets validate the effectiveness of structure preservation in the proposed model.
IMAGE AND VISION COMPUTING
(2021)
Article
Automation & Control Systems
Skyler Badge, Sumit Soman, Suresh Chandra, Jayadeva
Summary: The Minimal Complexity Machine (MCM) is a kernel-based learning model that can learn very sparse models with improved generalization through data dependent optimized kernels. Results on benchmark datasets demonstrate both model sparsity and improved generalization for the MCM and a large-scale MCM variant.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2021)
Article
Mathematics, Interdisciplinary Applications
Shaofei Zang, Dongqing Li, Chao Ma, Jianwei Ma
Summary: This paper proposes a method called JTELM, which combines cross-domain mean approximation and subspace alignment to achieve transfer learning even in the absence of labeled samples. The experiments show that JTELM outperforms traditional ELM and other transfer or nontransfer learning methods.
Article
Computer Science, Artificial Intelligence
Feng Liu, Guangquan Zhang, Jie Lu
Summary: This article introduces a shared-fuzzy-equivalence-relation neural network (SFERNN) for addressing the multisource heterogeneous UDA problem, which optimizes parameters by minimizing cross-entropy loss and distributional discrepancy between source and target domains. Experimental results demonstrate that SFERNN outperforms existing single-source heterogeneous UDA methods on multiple real-world datasets.
IEEE TRANSACTIONS ON FUZZY SYSTEMS
(2021)
Article
Computer Science, Artificial Intelligence
Xiao-Lin Xu, Geng-Xin Xu, Chuan-Xian Ren, Dao-Qing Dai, Hong Yan
Summary: In this paper, a conditional independence induced unsupervised domain adaptation (CIDA) method is proposed to tackle the dataset bias problem. The method aims to find the low-dimensional and transferable feature representation by optimizing mutual information terms. Experimental results demonstrate the effectiveness of CIDA.
PATTERN RECOGNITION
(2023)
Article
Mathematics
Pengfei Ge, Yesen Sun
Summary: The discriminability and transferability of models are important for domain adaptation methods. This study proposes a new deep domain adaptation method, GPTKL, which uses Gaussian Process-based Transfer Kernel Learning to improve the discrimination ability of the model. Experimental results demonstrate the superior classification performance of GPTKL compared to state-of-the-art methods.
Article
Computer Science, Artificial Intelligence
Keqiuyin Li, Jie Lu, Hua Zuo, Guangquan Zhang
Summary: Unsupervised domain adaptation leverages knowledge from a labeled source domain to tackle a similar unlabeled target domain. Existing methods transfer knowledge from a single source domain, but this may be insufficient. This paper proposes a dynamic classifier alignment (DCA) method for multi-source domain adaptation, which aligns classifiers driven by multi-view features.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Artificial Intelligence
Xing Wei, Bin Wen, Fan Yang, Yujie Liu, Chong Zhao, Di Hu, Hui Luo
Summary: Unsupervised domain adaptation aims to transfer knowledge from labeled source domain to unlabeled target domain for improving target domain's classification performance. However, existing methods only focus on domain alignment in contrastive learning, neglecting the classification task, leading to suboptimal solutions. In this paper, we propose Task-oriented Contrastive Learning for Unsupervised Domain Adaptation (TOCL) to address the inconsistency problem. Our method performs feature weighting and extracts task-related features from source domain images after data augmentation for contrastive learning. Additionally, a delimitation discriminator is introduced to detect and minimize the output difference between two types of data-augmented samples in the target domain. Extensive experiments on public datasets demonstrate the effectiveness and adaptability of TOCL in improving classification accuracy.
EXPERT SYSTEMS WITH APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Qiang Zhou, Wen'an Zhou, Shirui Wang, Ying Xing
Summary: The study introduces a novel adversarial distribution adaptation network (ADAN) for domain adaptation, aiming to learn domain-invariant representations by reducing discrepancies in global and local distributions. It adapts the global distribution between two domains using a single-domain discriminator and aligns the local distributions between sub-domains using source decision boundaries. Extending ADAN to improved ADAN (iADAN) with a feature norm term enhances model generalization and shows superior performance on Office-Home and ImageCLEF-DA datasets compared to other state-of-the-art domain adaptation methods.
NEURAL COMPUTING & APPLICATIONS
(2021)
Article
Computer Science, Artificial Intelligence
Sean Shensheng Xu, Man-Wai Mak, Chunqi Chang
Summary: This paper proposes an unsupervised patient adaptation approach for creating patient-specific deep neural network (DNN) classifiers based on the patient-specific i-vectors from unlabeled patient-specific ECG data. Evaluation on the MIT-BIH arrhythmia dataset shows that the proposed approach outperforms existing models, making personalized ECG classification more practical.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Geochemistry & Geophysics
Abhishek Singh, Lorenzo Bruzzone
Summary: In this study, a wavelet-inspired attention-based convolutional neural network (WIANet) architecture is proposed for land cover classification. By combining wavelet convolution and attention units with the UNet-based architecture, the spectral and texture information can be better utilized to distinguish classes with high similarity in multispectral images.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Geochemistry & Geophysics
Leonardo Carrer, Davide Castelletti, Riccardo Pozzobon, Francesco Sauro, Lorenzo Bruzzone
Summary: Caves are important scientific targets due to their significant biodiversity and unique geological formations, but it is difficult to determine the accessibility of cave systems through skylights. This article proposes a methodology using high-resolution orbital synthetic aperture radar (SAR) imaging systems to estimate cave characteristics and accessibility near skylights. Experimental results demonstrate the effectiveness of this methodology in various surface conditions, and unknown cave systems were discovered near Volcan Wolf and Ecuador, Isla Isabela, Galapagos. The implications of this work extend to geological studies, ecology, and space exploration research.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Leonardo Carrer, Lorenzo Bruzzone
Summary: This article investigates the presence of lava tubes on the Moon and Mars using radar sounding data. The study determines that lava tubes exhibit self-affine fractal structures at horizontal scales relevant to radar sounding, and they have electromagnetic roughness in the VHF band. These findings have implications for current radar sounding systems' ability to detect lava tubes and for future missions devoted to lava tube detection and characterization.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Astronomy & Astrophysics
Sanchari Thakur, Elisa Sbalchiero, Lorenzo Bruzzone
Summary: The exploration of Venus is gaining importance, and EnVision has been selected as the European Space Agency's mission to study the planet's shallow crust. The development of the subsurface radar sounder requires detailed analysis based on simulations due to the lack of high-quality data. This study aims to identify hypotheses and targets for subsurface exploration of Venus.
PLANETARY AND SPACE SCIENCE
(2023)
Article
Astronomy & Astrophysics
Elisa Sbalchiero, Sanchari Thakur, Marco Cortellazzi, Lorenzo Bruzzone
Summary: The RIME radar on the JUICE mission will explore the icy moon Ganymede to search for dielectric and mechanical interfaces below its surface. The radar can potentially detect and differentiate between different geological components that support the formation hypotheses of the bright terrains on Ganymede.
Article
Sport Sciences
Chen Kunlun, Liu Xiaoqiong, Zhang Xu, Ding Lei
Summary: The Chinese Go League has developed and flourished over a period of 20 years, becoming a key part of urban sports culture. Using mathematical statistics, spatiotemporal trajectory, and geospatial analysis, this study investigates the spatiotemporal distribution pattern and diffusive evolution characteristics of Chinese Go League clubs. The results indicate that the number of Go League clubs has increased in a stepped pattern, with a spatial distribution primarily focused on municipalities and provincial capital cities. There is potential for further expansion and the future development of the Chinese Go industry can be enhanced by marketability and building relationships with host cities.
FRONTIERS IN SPORTS AND ACTIVE LIVING
(2023)
Review
Astronomy & Astrophysics
Leigh N. Fletcher, Thibault Cavalie, Davide Grassi, Ricardo Hueso, Luisa M. Lara, Yohai Kaspi, Eli Galanti, Thomas K. Greathouse, Philippa M. Molyneux, Marina Galand, Claire Vallat, Olivier Witasse, Rosario Lorente, Paul Hartogh, Francois Poulet, Yves Langevin, Pasquale Palumbo, G. Randall Gladstone, Kurt D. Retherford, Michele K. Dougherty, Jan-Erik Wahlund, Stas Barabash, Luciano Iess, Lorenzo Bruzzone, Hauke Hussmann, Leonid I. Gurvits, Ondrej Santolik, Ivana Kolmasova, Georg Fischer, Ingo Mueller-Wodarg, Giuseppe Piccioni, Thierry Fouchet, Jean-Claude Gerard, Agustin Sanchez-Lavega, Patrick G. J. Irwin, Denis Grodent, Francesca Altieri, Alessandro Mura, Pierre Drossart, Josh Kammer, Rohini Giles, Stephanie Cazaux, Geraint Jones, Maria Smirnova, Emmanuel Lellouch, Alexander S. Medvedev, Raphael Moreno, Ladislav Rezac, Athena Coustenis, Marc Costa
Summary: ESA's Jupiter Icy Moons Explorer (JUICE) will conduct a detailed investigation of the Jovian system in the 2030s, utilizing state-of-the-art instruments and a tailored orbital tour. The mission aims to gather information on the climate, meteorology, and chemistry of Jupiter's atmosphere and auroras, as well as studying phenomena on various timescales. The remote sensing payload includes spectroscopy, imaging, and sounding techniques, allowing for a comprehensive characterization of the planet's atmosphere and auroras.
SPACE SCIENCE REVIEWS
(2023)
Article
Geochemistry & Geophysics
Xuming Zhang, Yuanchao Su, Lianru Gao, Lorenzo Bruzzone, Xingfa Gu, Qingjiu Tian
Summary: This article proposes two types of lightweight self-attention modules (CLMSA and PLMSA) to reduce the memory and computation burden of the transformer model in hyperspectral image classification. A lightweight transformer (LiT) network is built with these modules, combining convolutional blocks and transformers to extract both local and long-range dependencies. Additionally, a controlled multiclass stratified (CMS) sampling strategy is used to generate appropriately sized training data and mitigate overfitting. Experimental results validate the effectiveness of the proposed design.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Engineering, Electrical & Electronic
Giulio Weikmann, Daniele Marinelli, Claudia Paris, Silke Migdall, Eva Gleisberg, Florian Appel, Heike Bach, Jim Dowling, Lorenzo Bruzzone
Summary: This article presents a novel system that produces multiyear high-resolution irrigation water demand maps for agricultural areas, enabling a new level of detail for irrigation support. The system uses a scalable distributed deep learning model trained on Sentinel-2 images and integrates satellite data and meteorological data to derive irrigation water demand. The software architecture of the system relies on the integration of the Food Security TEP and the data-intensive AI Hopsworks platform.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Zhi Sheng, Feng Zhang, Jiande Sun, Yanyan Tan, Kai Zhang, Lorenzo Bruzzone
Summary: This article proposes a unified two-stage spatial and spectral network (UTSN) for pansharpening. It constructs a branch of networks for each different satellite, in which the spatial enhancement network (SEN) improves the spatial details in the fused images, and the spectral adjustment network (SAN) captures the spectral characteristics of the specific satellite. By refining the spectral information in the intermediate image through SAN, the proposed method achieves promising pansharpening results even for a new satellite with limited training images.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Kai Zhang, Xue Zhao, Feng Zhang, Lei Ding, Jiande Sun, Lorenzo Bruzzone
Summary: In this article, a cross-temporal difference (CTD) attention mechanism is proposed to capture changes in multitemporal images. Two CTD-transformer encoders and decoders are designed to extract and improve features of changed areas. Experimental results show that the proposed method outperforms state-of-the-art methods on LEVIR-CD, WHU-CD, and CLCD datasets.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Remote Sensing
Yansheng Li, Xinwei Li, Yongjun Zhang, Daifeng Peng, Lorenzo Bruzzone
Summary: Remote sensing big data presents a challenge in efficiently mining useful information. Deep learning has been introduced, but the generation of high-quality pixel-level labels is a major obstacle. Weakly supervised deep learning is a promising solution. This review summarizes the achievements of weakly supervised deep learning-based information extraction from remote sensing big data and outlines potential research directions.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2023)
Article
Geochemistry & Geophysics
Renchu Guan, Mingming Wang, Lorenzo Bruzzone, Haishi Zhao, Chen Yang
Summary: Semantic segmentation is a challenging task in VHR remote sensing applications. DCNNs with attention mechanism have shown outstanding performance. However, existing attention-guided methods have limitations due to the estimation of a large number of parameters and limited labeled samples. This article proposes a lightweight model with multistage feature fusion to overcome these limitations and achieve promising performance.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Kai Zhang, Anfei Wang, Feng Zhang, Wenbo Wan, Jiande Sun, Lorenzo Bruzzone
Summary: In this study, we proposed a BP-driven model called S2DBPN to fuse low spatial resolution multi-spectral (LR MS) and high spatial resolution panchromatic (PAN) images. S2DBPN consists of a spatial BP network, a spectral BP network, and a reconstruction network, and integrates features from spatial and spectral BP networks to generate the desired high-resolution multi-spectral (HR MS) image. Experimental results demonstrate that SDBPN-D-2 can produce better HR MS images in terms of qualitative and quantitative evaluation metrics when compared to existing methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Engineering, Electrical & Electronic
Parth Naik, Michele Dalponte, Lorenzo Bruzzone
Summary: This study proposes an automated machine learning framework for predicting forest above-ground biomass (AGB). The framework reduces human-bias through hyper-parameter optimization and automatic feature extraction, and improves prediction accuracy through model ensembling.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
