Article
Geochemistry & Geophysics
Xiaoshu Chen, Shaoming Pan, Yanwen Chong
Summary: The paper proposes a region and category adaptive domain discriminator (RCA-DD) to emphasize the differences in regions and categories during the alignment process by introducing an entropy-based regional attention module and class-clear module, aiming to improve the effectiveness of unsupervised domain adaptation methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Yuchi Ma, Zhengwei Yang, Zhou Zhang
Summary: Recently, supervised machine learning methods have been widely used for crop yield prediction using remote sensing data. However, due to domain shift, these models often lack spatial transferability. To address this issue, a multisource maximum predictor discrepancy (MMPD) neural network is proposed as an unsupervised domain adaptation approach for corn yield prediction. The MMPD model effectively reduces domain shifts and outperforms other state-of-the-art deep learning and domain adaptation methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Kuiliang Gao, Anzhu Yu, Xiong You, Chunping Qiu, Bing Liu
Summary: This article proposes a novel cross-domain multiprototypes learning method to address the problem of class-level confusion in unsupervised domain adaptation of remote sensing images (RSIs). Multiple prototypes are learned for each class to achieve better domain alignment. Additionally, two masked consistency learning strategies are designed to explore the contextual structure of target RSIs and improve the quality of pseudo labels for prototype updating.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Shunping Zhou, Yuting Feng, Shengwen Li, Daoyuan Zheng, Fang Fang, Yuanyuan Liu, Bo Wan
Summary: This article proposes a framework to introduce digital surface model (DSM) data for the unsupervised semantic segmentation of high-resolution remote sensing imagery (RSI). The proposed method combines RSI with DSM through multipath encoder (MPE) and multitask decoder (MTD), and aligns global data distribution in the source and target domains with a unsupervised domain adaptation (UDA) module. Experimental results show that the proposed method substantially improves the semantic segmentation performance on high-resolution RSI and outperforms state-of-the-art methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Xianping Ma, Xiaokang Zhang, Zhiguo Wang, Man-On Pun
Summary: This work investigates unsupervised domain adaptation (UDA)-based semantic segmentation of very high-resolution (VHR) remote sensing (RS) images from different domains. To address the limitations of existing UDA methods, this paper proposes a mutually boosted attention transformer (MBATrans) to capture cross-domain dependencies and reduce domain discrepancies. The proposed MBATrans-augmented GAN architecture achieves superior performance on two large-scale VHR RS datasets.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Computer Science, Artificial Intelligence
Shuang Wang, Dong Zhao, Chi Zhang, Yuwei Guo, Qi Zang, Yu Gu, Yi Li, Licheng Jiao
Summary: This paper proposes a novel cluster alignment framework for unsupervised domain adaptation (UDA), which aims to improve the adaptability of models in the target domain by mining domain-specific knowledge. The proposed method utilizes a multi-prototype clustering strategy to tightly distribute pixel features within the same class, aligns the domain distributions using a contrastive strategy, and learns task-specific decision boundaries with an affinity-based normalized cut loss.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2022)
Article
Geochemistry & Geophysics
Jianhua Guo, Jingyu Yang, Huanjing Yue, Kun Li
Summary: In order to reduce the labeling cost of cloud detection methods, this study proposes an unsupervised domain adaptation approach that can generalize the model trained on labeled images of the source satellite to unlabeled images of the target satellite. By using grouped features alignment and entropy minimization methods, domain-invariant representations are extracted to improve the cloud detection accuracy of different satellite images.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Pourya Shamsolmoali, Masoumeh Zareapoor, Huiyu Zhou, Ruili Wang, Jie Yang
Summary: A new model is introduced to apply structured domain adaption for synthetic image generation and road segmentation, incorporating a feature pyramid network into generative adversarial networks to minimize the difference between the source and target domains and improve road extraction accuracy and completeness.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2021)
Article
Geochemistry & Geophysics
Liang Yan, Bin Fan, Shiming Xiang, Chunhong Pan
Summary: Semantic segmentation of remote sensing images has made significant progress with supervised deep learning, but struggles with domain shift between different data domains. To address this challenge, a cross mean teacher (CMT) UDA method is proposed, which utilizes global alignment and a cross teacher-student network for effective pixel utilization.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Radiology, Nuclear Medicine & Medical Imaging
Bo Li, Xinge You, Jing Wang, Qinmu Peng, Shi Yin, Ruinan Qi, Qianqian Ren, Ziming Hong
Summary: The study addresses the issue of poor performance in neonatal brain MRI segmentation when trained on a source domain and applied in a target domain. A novel framework called IAS-NET is proposed for intraclass feature alignment, showing improved results compared to current UDA methods in segmenting neonatal brain MR images.
Article
Environmental Sciences
Junbo Zhang, Shifeng Xu, Jun Sun, Dinghua Ou, Xiaobo Wu, Mantao Wang
Summary: This work utilizes an unsupervised adversarial domain adaptation method to train a neural network for unsupervised agricultural land extraction by reducing the gap between the source and target domains. The approach consists of two phases: inter-domain adaptation and intra-domain adaptation. Experimental results demonstrate the effectiveness of the method in comparison to other unsupervised domain adaptation techniques.
Article
Engineering, Electrical & Electronic
Luhan Wang, Pengfeng Xiao, Xueliang Zhang, Xinyang Chen
Summary: This study proposes a novel fine-grained adaptation framework that combines global-local alignment and category-level alignment to address the problems in cross-domain segmentation of remote sensing images (RSIs). Experiments demonstrate that local adaptation and category-level adaptation of RSIs are complementary in cross-domain segmentation, and the integrated framework helps achieve outstanding performance for unsupervised domain adaptation (UDA) semantic segmentation of RSIs.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Jia-Xin Wang, Si-Bao Chen, Chris H. Q. Ding, Jin Tang, Bin Luo
Summary: The article introduces a semi-supervised remote sensing image semantic segmentation method, RanPaste, which combines labeled and unlabeled images to enhance segmentation performance. By combining consistency regularization and pseudo label, and utilizing thresholds to gradually improve model performance, the method enables the model to learn more underlying information from unlabeled data.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Chunhui Zhao, Boao Qin, Shou Feng, Wenxiang Zhu, Lifu Zhang, Jinchang Ren
Summary: In this paper, a novel unsupervised domain adaptation framework is proposed for cross-scene hyperspectral image classification. The framework aligns task-related features and learns task-specific decision boundaries, improving the classification performance.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Sarmad F. Ismael, Koray Kayabol, Erchan Aptoula
Summary: This letter proposes a new lightweight unsupervised domain adaptation method for semantic segmentation of high-resolution remote sensing images, using an image-to-image translation approach where latent content representations are mixed across domains, and a perceptual network module and loss function enforce visual semantic consistency. Cross-domain comparative experiments demonstrate that this method can generate semantically consistent images, operate with a single target domain sample (one-shot), and achieve better performance than state-of-the-art methods with a fraction of the required parameters.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Review
Computer Science, Artificial Intelligence
Jie Gui, Zhenan Sun, Yonggang Wen, Dacheng Tao, Jieping Ye
Summary: In this paper, a comprehensive review of various GAN methods is provided from the perspectives of algorithms, theory, and applications. The motivations, mathematical representations, and structures of most GAN algorithms are detailed and compared. Theoretical issues related to GANs are also investigated, and the typical applications of GANs in various fields are discussed.
IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING
(2023)
Article
Computer Science, Information Systems
Pengfei Wang, Changxing Ding, Wentao Tan, Mingming Gong, Kui Jia, Dacheng Tao
Summary: Unsupervised Domain Adaptive (UDA) object re-identification (Re-ID) aims to adapt a model trained on a labeled source domain to an unlabeled target domain. To address the issue of label noise caused by clustering algorithms, we propose an uncertainty-aware clustering framework (UCF) for UDA tasks. Our UCF method consistently achieves state-of-the-art performance in multiple UDA tasks for object Re-ID and significantly reduces the performance gap between unsupervised and supervised Re-ID.
IEEE TRANSACTIONS ON MULTIMEDIA
(2023)
Article
Computer Science, Artificial Intelligence
Kai Han, Yunhe Wang, Hanting Chen, Xinghao Chen, Jianyuan Guo, Zhenhua Liu, Yehui Tang, An Xiao, Chunjing Xu, Yixing Xu, Zhaohui Yang, Yiman Zhang, Dacheng Tao
Summary: Transformer, a deep neural network with a self-attention mechanism, has been initially used in natural language processing and is now gaining attention in computer vision tasks. Transformer-based models perform as well as or even better than convolutional and recurrent neural networks in various visual benchmarks. This paper reviews vision transformer models, categorizes them based on different tasks, and analyzes their advantages and disadvantages. The discussed categories include backbone network, high/mid-level vision, low-level vision, and video processing. Efficient methods for applying transformer in real device-based applications are also explored. The challenges and further research directions for vision transformers are discussed as well.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Book Review
Education & Educational Research
Jing Zhang
EDUCATIONAL PHILOSOPHY AND THEORY
(2023)
Book Review
Education & Educational Research
Jing Zhang
JOURNAL OF EDUCATION FOR TEACHING
(2023)
Article
Computer Science, Artificial Intelligence
Jinlong Fan, Jing Zhang, Dacheng Tao
Summary: This paper proposes a novel self-supervised image rectification method based on the idea that the rectified results of distorted images from different lenses should be the same. A new network architecture is designed with a shared encoder and multiple prediction heads, and a differentiable warping module is used to generate rectified and re-distorted images. The self-supervised learning scheme achieves comparable or better performance than the supervised baseline method and state-of-the-art methods.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Qiming Zhang, Yufei Xu, Jing Zhang, Dacheng Tao
Summary: Vision transformers have shown promise in computer vision tasks due to their ability to model long-range dependency. However, they lack an intrinsic bias in modeling local visual structures and dealing with scale variance. This paper introduces the ViTAE transformer, which utilizes two biases and achieves superior performance on various datasets.
INTERNATIONAL JOURNAL OF COMPUTER VISION
(2023)
Article
Computer Science, Artificial Intelligence
Jingwei Zhang, Tongliang Liu, Dacheng Tao
Summary: In this article, a new analysis of generalization in deep neural networks (DNNs) is proposed from an optimal transport perspective. Upper bounds on the generalization error of learning algorithms are derived based on the algorithmic transport cost, and various conditions for loss functions are studied. The main result shows that the generalization error in DNNs decreases exponentially to zero as the number of layers increases.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Benteng Ma, Jing Zhang, Yong Xia, Dacheng Tao
Summary: Neural architecture search (NAS) is a popular research topic for identifying better architectures. Recently, differential neural architecture search methods have gained attention for their effectiveness. This paper proposes a novel inter-layer transition NAS method to investigate the dependency between edges in a network.
PATTERN RECOGNITION
(2023)
Article
Automation & Control Systems
Yuqing Ma, Xianglong Liu, Shihao Bai, Lei Wang, Aishan Liu, Dacheng Tao, Edwin R. Hancock
Summary: In this study, a generic inpainting framework is proposed to handle incomplete images with both contiguous and discontiguous large missing areas. By employing an adversarial modeling and regionwise operations, the framework is able to generate semantically reasonable and visually realistic images, outperforming existing methods on large contiguous and discontiguous missing areas, as demonstrated by qualitative and quantitative experiments.
IEEE TRANSACTIONS ON CYBERNETICS
(2023)
Article
Automation & Control Systems
Xinyu Wu, Jinke Li, Liu Liu, Dacheng Tao
Summary: The lower limb power-assist exoskeletons are expected to assist paraplegic individuals in walking again. However, most exoskeletons only work in known environments, making it challenging to understand the user's intention and plan footstep sequences in unknown scenes. This study proposes a visual footstep planning system based on the Bezier curve, integrating Hololens and Realsense for environment understanding and user behavior intention recognition. Experimental results demonstrate that the planning time is reduced by 67.46% compared to traditional search algorithms, validating the effectiveness of the proposed system on a visual interaction platform.
IEEE TRANSACTIONS ON SYSTEMS MAN CYBERNETICS-SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Sheng Wan, Yibing Zhan, Shuo Chen, Shirui Pan, Jian Yang, Dacheng Tao, Chen Gong
Summary: Contrastive learning is a key technique for self-supervised representation learning, but the uniform negative sampling strategy limits the expressive power of contrastive models. To address this, the article proposes an adaptive sampling strategy called AdaS and introduces an auxiliary polarization regularizer to improve the performance of graph contrastive learning.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Computer Science, Artificial Intelligence
Tinglei Feng, Yingjie Zhai, Jufeng Yang, Jie Liang, Deng-Ping Fan, Jing Zhang, Ling Shao, Dacheng Tao
Summary: Image complexity is an important visual perception for humans to understand an image. Evaluating image complexity is challenging due to its subjective nature and the diversity of real-world images. To address this, we have created a large-scale dataset with 9,600 well-annotated images and developed a base model to predict complexity scores and density maps. The model is effective and correlates well with human perception. Additionally, exploring image complexity can enhance the performance of computer vision tasks. The dataset and source code can be found at https://github.com/tinglyfeng/IC9600.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Long Lan, Xiao Teng, Jing Zhang, Xiang Zhang, Dacheng Tao
Summary: In this study, an unsupervised person re-identification method is proposed, which has achieved great progress by training with pseudo labels. To purify the feature and label noise, multi-view features and the knowledge of a teacher model are utilized. Experimental results demonstrate the effectiveness of this approach for unsupervised person re-identification.
IEEE TRANSACTIONS ON IMAGE PROCESSING
(2023)
Article
Computer Science, Artificial Intelligence
Mingjin Zhang, Jingwei Xin, Jing Zhang, Dacheng Tao, Xinbo Gao
Summary: This article addresses the problem of detecting hardware Trojan from microscope chip images. It proposes a novel MCI super-resolution method using a curvature consistent network, which can recover more delicate circuit lines and improve HT detection performance. Experiments on a new benchmark dataset called MCI demonstrate the superiority of the proposed method over representative SR methods. The MCI dataset is available on GitHub.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)