Article
Geochemistry & Geophysics
Wei-Ye Wang, Heng-Chao Li, Yang-Jun Deng, Li-Yang Shao, Xiao-Qiang Lu, Qian Du
Summary: This study introduces a novel generative adversarial network (GAN) for hyperspectral image (HSI) classification, which uses artificial sample generation for data augmentation to improve classification performance. Experimental results show that the proposed method outperforms several state-of-the-art deep classification methods on two real HSI datasets.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2021)
Article
Engineering, Electrical & Electronic
Ziping He, Kewen Xia, Pedram Ghamisi, Yuhen Hu, Shurui Fan, Baokai Zu
Summary: Generative adversarial networks (GANs) have shown great potential in hyperspectral image (HSI) classification. However, due to the class imbalance problem in HSI data, GANs tend to misclassify minority-class samples. In this paper, we propose a semisupervised generative adversarial network called HyperViTGAN, which incorporates a transformer and an external semisupervised classifier to address this issue. The HyperViTGAN utilizes adversarial learning to generate HSI patches and captures semantic context and low-level textures to preserve critical information. Experimental results demonstrate that the proposed model achieves competitive performance in comparison with state-of-the-art classification models on three well-known HSI datasets.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2022)
Article
Computer Science, Artificial Intelligence
Anyang Song, Huixian Duan, Haodong Pei, Lei Ding
Summary: This paper proposes a novel fusion model for infrared and visible image fusion using a triple-discriminator generative adversarial network, which achieves a balance between clear boundaries and rich details. The difference image obtained from image subtraction highlights the difference information, extracts image details, and retains the contrast of infrared targets and texture details in visible images during fusion.
Article
Environmental Sciences
Zhitao Chen, Lei Tong, Bin Qian, Jing Yu, Chuangbai Xiao
Summary: The article introduces a Self-Attention-Based Conditional Variational Autoencoder Generative Adversarial Network (SACVAEGAN) that aims to improve hyperspectral image classification performance by generating enhanced virtual samples, focusing on global information, and incorporating the WGAN-GP loss function to enhance model stability. Experimental results and comparisons with state-of-the-art methods demonstrate the significant advantages of SACVAEGAN in accuracy.
Article
Geochemistry & Geophysics
Fan Zhang, Jing Bai, Jingsen Zhang, Zhu Xiao, Changxing Pei
Summary: The translation describes how GAN is applied to HSI classification, proposing a new method combining PG-GAN and WGAN-GP ideas, which achieves good training results on public datasets.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2021)
Article
Computer Science, Artificial Intelligence
Hamed Jabbari, Nooshin Bigdeli
Summary: Male infertility negatively impacts infertile couples, and sperm head morphology plays a crucial role in evaluating male infertility. However, classifying sperm head images is challenging due to an imbalance between abnormal and normal samples. Capsule neural networks (CapsNets) offer a promising solution for imbalanced classification by considering spatial relationships of features, and generative adversarial networks (GANs) help improve this classification by generating synthetic samples. This paper proposes and evaluates a new architecture based on CapsNets and GANs for imbalanced classification of human sperm head images, achieving superior performance compared to other deep learning networks.
NEURAL COMPUTING & APPLICATIONS
(2023)
Article
Environmental Sciences
Andrew Hennessy, Kenneth Clarke, Megan Lewis
Summary: The study explores the use of Generative Adversarial Networks to generate realistic synthetic hyperspectral vegetation data while maintaining class relationships, leading to improved classification accuracy and efficiency in handling vegetation spectral data.
Article
Environmental Sciences
Zhongwei Li, Xue Zhu, Ziqi Xin, Fangming Guo, Xingshuai Cui, Leiquan Wang
Summary: The proposed CSSVGAN model achieves excellent hyperspectral image classification performance by utilizing crossed spatial and spectral interactions.
Article
Engineering, Electrical & Electronic
Wei Liu, Jie You, Joonwhoan Lee
Summary: This article introduces a novel generative adversarial network for conditional HSI generation and classification, utilizing a multi-stage progressive training method and a multipole GAN framework to simulate the spatial-spectral distribution features of HSIs. The inclusion of a spectral classifier helps stabilize and optimize the model, resulting in high accuracy and diversity in experimental results.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2021)
Article
Geochemistry & Geophysics
Jing Bai, Yang Zhang, Zhu Xiao, Fawang Ye, You Li, Mamoun Alazab, Licheng Jiao
Summary: Hyperspectral image classification (HIC) algorithm based on deep learning has been widely studied and achieved better results than traditional algorithms. However, using small samples for HIC leads to poor and volatile results with GAN methods. To address this issue, researchers propose a novel immune evolutionary generative adversarial network (HIEGAN) that leverages the evolutionary and immune strategies. HIEGAN overcomes the drawbacks of GAN, improves stability, and enhances classification efficiency.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Engineering, Electrical & Electronic
Caihao Sun, Xiaohua Zhang, Hongyun Meng, Xianghai Cao, Jinhua Zhang, Licheng Jiao
Summary: In this article, a semisupervised dual-branch spectral-spatial adversarial representation learning (SSARL) method is proposed for HSI classification. SSARL adaptively assigns attention weights to different pixels and adds a spectral constraint to spatial features. Experimental results demonstrate that SSARL outperforms competitive methods on small-sized labeled samples and exhibits superior performance for boundary test pixels.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Renlong Hang, Feng Zhou, Qingshan Liu, Pedram Ghamisi
Summary: In this article, a multitask generative adversarial network (MTGAN) is proposed to address the issue of deep learning models heavily depending on the quantity of available training samples in hyperspectral image classification. By utilizing rich information from unlabeled samples and employing an adversarial learning method, the MTGAN model is able to indirectly improve the discrimination and generalization ability of the classification task. Additionally, skip-layer connections are used to fully explore useful information from shallow layers, resulting in higher performance compared to other state-of-the-art deep learning models.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2021)
Article
Remote Sensing
Cuiping Shi, Tianyu Zhang, Diling Liao, Zhan Jin, Liguo Wang
Summary: This paper proposes a dual hybrid convolutional generative adversarial network (DHCGAN) for hyperspectral image classification, which effectively addresses the issues of chequerboard artifacts and mode collapse. The experimental results demonstrate significant performance improvement compared to competing methods.
INTERNATIONAL JOURNAL OF REMOTE SENSING
(2022)
Article
Geochemistry & Geophysics
Siyuan Hao, Yufeng Xia, Yuanxin Ye
Summary: In recent years, generative adversarial networks (GANs) have made significant progress in hyperspectral image classification. However, existing methods struggle to extract spectral sequence information effectively. In this paper, we propose a new framework called Transformer with residual upscale GAN (TRUG) to address the problem of insufficient training samples in hyperspectral images. Experimental results demonstrate that TRUG outperforms other models on three datasets.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Engineering, Electrical & Electronic
Baorui Wang, Shun Zhang, Yan Feng, Shaohui Mei, Sen Jia, Qian Du
Summary: This paper proposes a Hyperspectral Imagery Spatial Super-Resolution algorithm based on Generative Adversarial Network (HSSRGAN) to enhance spatial features and refine spectral information for increased spatial resolution and reduced spectral distortion.
IEEE TRANSACTIONS ON COMPUTATIONAL IMAGING
(2021)
Article
Computer Science, Artificial Intelligence
Haonan Qin, Weiying Xie, Yunsong Li, Kai Jiang, Jie Lei, Qian Du
Summary: This study proposes a two-stage detection framework based on adversarial learning for hyperspectral target detection. It extracts spectral features in latent space through background reconstruction and utilizes a generative adversarial network to estimate the background in a weakly supervised manner. Experimental results demonstrate the effectiveness of the proposed framework.
PATTERN RECOGNITION
(2023)
Article
Engineering, Electrical & Electronic
Wenhui Hou, Na Chen, Jiangtao Peng, Weiwei Sun, Qian Du
Summary: This article proposes a semisupervised deep learning method for hyperspectral image classification, which integrates active learning, self-paced learning, and deep learning. It achieves better classification performance and reduces labeling cost.
IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Swalpa Kumar Roy, Ankur Deria, Chiranjibi Shah, Juan M. Haut, Qian Du, Antonio Plaza
Summary: In recent years, convolutional neural networks (CNNs) have received significant attention in the classification of hyperspectral images (HSIs). The vision transformer (ViT), which incorporates a self-attention mechanism, has shown promising classification performance compared to CNNs. However, there is still room for improvement as the current version does not utilize spatial-spectral features.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Yang-Jun Deng, Heng-Chao Li, Si-Qiao Tan, Junhui Hou, Qian Du, Antonio Plaza
Summary: This article proposes a t-linear tensor subspace learning (tLTSL) model for robust feature extraction of hyperspectral images (HSIs) based on t-product projection. The t-product projection, a newly defined tensor transformation method, maximally captures the intrinsic structure of tensor data. Through the integrated tensor low-rank and sparse decomposition, the model effectively removes noise corruption and maps the high-order hyperspectral data into a subspace with comprehensive information. Additionally, a proposition related to tensor rank is proved to interpret the meaning of the tLTSL model. Extensive experiments on HSI data corrupted by simulated and real noise validate the effectiveness of tLTSL.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Hong Fang, Shanchuan Guo, Xin Wang, Sicong Liu, Cong Lin, Peijun Du
Summary: This study proposed a novel automatic binary scene-level change detection approach based on deep learning, which utilizes pretrained VGG-16 network and change vector analysis for scene-level direct predetection, implements pixel-level classification using decision tree, and designs a pixel-level to scene-level conversion strategy. A novel scene change detection triplet network (SCDTN) is then trained to produce the binary scene-level change map by integrating a late-fusion subnetwork and an early fusion subnetwork. Experimental results demonstrate the effectiveness and superiority of the proposed approach.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Environmental Sciences
Linya Zhao, Kun Tan, Xue Wang, Jianwei Ding, Zhaoxian Liu, Huilin Ma, Bo Han
Summary: This paper proposed a feature selection framework called RLFSR to improve the efficiency and stability of hyperspectral feature selection through the introduction of reinforcement learning. The Markov Decision Process was used to simulate the band selection process, and reinforcement learning agents were introduced to improve model performance. Two spectral feature evaluation methods were used to comprehensively evaluate all hyperspectral bands related to soil. The proposed RLFSR framework showed better performance in capturing the spectral characteristics of soil organic matter compared to existing methods.
Article
Geography, Physical
Peng Zhang, Cong Lin, Shanchuan Guo, Wei Zhang, Hong Fang, Peijun Du
Summary: This paper proposes an index-guided semantic segmentation framework for accurate urban vegetation mapping. By calculating a novel cross-scale vegetation index and training a semantic segmentation model, the proposed method outperforms existing RGB vegetation indices in highlighting vegetation and suppressing complex backgrounds. Experimental results show satisfactory performance comparable to fully-supervised methods.
INTERNATIONAL JOURNAL OF DIGITAL EARTH
(2023)
Review
Computer Science, Information Systems
Na Liu, Wei Li, Yinjian Wang, Ran Tao, Qian Du, Jocelyn Chanussot
Summary: The ability of hyperspectral images (HSIs) to capture fine spectral discriminative information allows them to observe, detect, and identify objects with subtle spectral discrepancies. However, HSIs may not accurately represent the true distribution of ground objects due to environmental disturbances, atmospheric effects, and hardware limitations. These degradations significantly reduce the quality and usefulness of HSIs. Low-rank tensor approximation (LRTA) has gained attention in the HSI restoration community and is effective in addressing convex and non-convex inverse optimization problems. This survey provides a comprehensive technical assessment of LRTA for HSI restoration, covering topics such as denoising, fusion, destriping, inpainting, deblurring, and super-resolution.
SCIENCE CHINA-INFORMATION SCIENCES
(2023)
Article
Geochemistry & Geophysics
Jiaojiao Li, Yihong Leng, Rui Song, Wei Liu, Yunsong Li, Qian Du
Summary: Spectral reconstruction (SR) aims to directly recover hyperspectral images (HSIs) from corresponding RGB images. Most existing supervised SR methods require large amounts of annotated data, which is limited by complex imaging techniques and laborious annotation calibration. Unsupervised strategies have attracted attention, but suffer from low accuracy. This paper proposes an unsupervised SR architecture with strong constraints, including a novel Masked Transformer (MFormer) to restore realistic HSIs by excavating latent hyperspectral characteristics.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Mingyang Ma, Shaohui Mei, Fan Li, Yaoyang Ge, Qian Du
Summary: This article proposes a spectral correlation-based diverse band selection method for hyperspectral images, which improves the representativeness and diversity of the selected bands by utilizing spectral correlations. The method uses a correlation-derived weight for weighted sparse reconstruction to select bands that are more correlated with the whole HSI, and a correlation minimization term to remove highly correlated bands. Additionally, an adjustable sparse constraint is imposed by using an l(2,0<p<=1) norm. Experimental results demonstrate that the proposed method outperforms state-of-the-art methods in HSI classification.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Yuxin Meng, Feng Gao, Eric Rigall, Ran Dong, Junyu Dong, Qian Du
Summary: Traditionally, numerical models have been used in oceanography studies to simulate ocean dynamics. However, many factors of ocean dynamics are ill-defined. We argue that transferring physical knowledge from observed data could improve the accuracy of numerical models in predicting sea surface temperature (SST).
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Jiaqing Zhang, Jie Lei, Weiying Xie, Zhenman Fang, Yunsong Li, Qian Du
Summary: In this article, the authors propose SuperYOLO, an accurate and fast object detection method for remote sensing images. By fusing multimodal data and utilizing assisted super resolution learning, SuperYOLO achieves high-resolution object detection on multiscale objects while considering the computation cost. Experimental results show that SuperYOLO outperforms state-of-the-art models in terms of accuracy and computational efficiency.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Jie Lei, Simin Xu, Weiying Xie, Jiaqing Zhang, Yunsong Li, Qian Du
Summary: This article introduces a novel method for detecting hyperspectral image (HSI) targets with certain spatial information, referred to as the semantic transferred priori for hyperspectral target detection with spatial-spectral association (SSAD). By using transfer learning, a semantic segmentation network adapted for HSIs is designed to discriminate the spatial areas of targets, and a customized target spectrum is aggregated with those spectral pixels localized. Experiments demonstrate that our proposed method achieves higher detection accuracy and superior visual performance compared to the other benchmark methods.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Hong Fang, Shanchuan Guo, Peng Zhang, Wei Zhang, Xin Wang, Sicong Liu, Peijun Du
Summary: Scene change detection provides a higher level understanding of changes on the Earth's surface, but there are issues with insufficient temporal change feature extraction and contradictory prediction results. To address these problems, a novel framework that integrates a differential aggregation network and class probability-based fusion strategy was proposed.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Bin Cui, Yao Peng, Hao Zhang, Wenmei Li, Peijun Du
Summary: This paper proposes a spectral-spatial convolutional network based on MRF and CoV for HSI classification. By combining MRF models and CoVs with spectral information, the fused features are used as input to produce reliable classification results.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)