Article
Environmental Sciences
Junbo Zhou, Shan Zeng, Zuyin Xiao, Jinbo Zhou, Hao Li, Zhen Kang
Summary: In this study, a novel hyperspectral image classification model called ESFNet is proposed. It addresses the issues of spectral continuity and poor learning of spectral information in traditional models through an optimized multi-scale fused spectral attention module and a 3D convolutional neural network based on the fusion of different spectral strides, resulting in improved classification accuracy.
Article
Remote Sensing
Xinru Fan, Wenhui Guo, Xueqin Wang, Yanjiang Wang
Summary: In hyperspectral image categorization, convolutional neural networks can address the issues of high dimensionality and a lack of labeled samples. This research introduces a multiscale piecewise spectral-spatial attention network (MPSSAN) that improves accuracy with fewer labeled samples. Experimental findings show that MPSSAN enhances classification performance compared to state-of-the-art methods.
INTERNATIONAL JOURNAL OF REMOTE SENSING
(2023)
Article
Computer Science, Information Systems
Wenping Ma, Haoxiang Ma, Hao Zhu, Yating Li, Longwei Li, Licheng Jiao, Biao Hou
Summary: This paper proposes a spatial and spectral kernels generation network (SSKNet) to tackle the classification of hyperspectral images, achieving more efficient feature extraction and fusion through the generation of spatial and spectral convolution kernels. Experimental results demonstrate that this method outperforms existing ones in terms of classification accuracy and generalization performance.
INFORMATION SCIENCES
(2021)
Article
Environmental Sciences
Qiuyue Liu, Min Fu, Xuefeng Liu
Summary: This paper proposes a method for hyperspectral image (HSI) based on 2D DSR shadow enhancement and CNN classification combined with an attention mechanism. It protects the spatial correlation of HSIs and enhances the signals in shadow regions, making full use of the rich information in remote sensing images.
Article
Environmental Sciences
Jun Sun, Junbo Zhang, Xuesong Gao, Mantao Wang, Dinghua Ou, Xiaobo Wu, Dejun Zhang
Summary: In this paper, we propose an encoder-decoder network that fuses spatial attention and spectral-channel attention for hyperspectral image classification. A multi-source attention mechanism is used to encode the spatial and spectral multi-channels contextual information, and three fusion strategies are proposed. The encoder-decoder framework can extract long-range context information and efficiently fuse multi-scale features. Our approach outperforms other state-of-the-art methods in hyperspectral image classification.
Article
Geochemistry & Geophysics
Linzhou Yu, Jiangtao Peng, Na Chen, Weiwei Sun, Qian Du
Summary: This paper proposes a novel two-branch deeper graph convolutional network (TBDGCN) to extract both superpixel and pixel-level features for hyperspectral image classification. The TBDGCN model addresses the oversmoothing and overfitting problems by using the DropEdge technique and residual connection in the GCN branch, and captures attention-based spectral-spatial features through a mixed attention mechanism in the CNN branch. The features from both branches are fused for classification.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Environmental Sciences
Xiaohui Huang, Yunfei Zhou, Xiaofei Yang, Xianhong Zhu, Ke Wang
Summary: This paper proposes a new spatial-spectral Transformer network (SS-TMNet) for HSI classification, which can effectively extract local and global spatial-spectral information. Experimental results on three widely used HSI datasets demonstrate the superiority of SS-TMNet over existing methods.
Article
Environmental Sciences
Haimiao Ge, Liguo Wang, Moqi Liu, Yuexia Zhu, Xiaoyu Zhao, Haizhu Pan, Yanzhong Liu
Summary: In this study, a two-branch convolutional neural network with a polarized full attention mechanism is proposed for hyperspectral image classification. The network efficiently extracts spectral and spatial features through two branches and simplifies the kernel sizes of convolutional layers to reduce complexity. The one-shot connection technique improves feature extraction efficiency, and the polarized full attention mechanism fuses feature maps and provides global contextual information. Experimental results validate the effectiveness of the proposed network.
Article
Environmental Sciences
Haojin Tang, Yanshan Li, Zhiquan Huang, Li Zhang, Weixin Xie
Summary: This paper proposes a fusion framework for small-sample hyperspectral image (HSI) classification. It extracts handcrafted spatial-spectral features using a 3D fuzzy histogram of oriented gradients (3D-FHOG) descriptor and combines CNN-based spatial-spectral features using a multidimensional Siamese network (MDSN). Experimental results show that the proposed fusion framework outperforms representative handcrafted feature-based and CNN-based methods.
Article
Environmental Sciences
Haizhu Pan, Xiaoyu Zhao, Haimiao Ge, Moqi Liu, Cuiping Shi
Summary: Hyperspectral image (HSI) classification is a crucial task in remote sensing processing. Attention mechanisms have been favored over convolutional neural networks (CNNs) due to their superior ability to express information during HSI processing. However, achieving high-accuracy classification with limited labeled samples remains a challenge.
Article
Geochemistry & Geophysics
Qian Liu, Zebin Wu, Yang Xu, Zhihui Wei
Summary: This article proposes a unified attention paradigm (UAP) that defines attention mechanism as a three-stage process of optimizing feature representations, strengthening information interaction, and emphasizing meaningful information. It also introduces a novel efficient spectral-spatial attention module (ESSAM) that adapts feature responses along spectral and spatial dimensions at low parameter cost. Experimental results demonstrate that ESSAM brings higher accuracy improvement compared to advanced attention models.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Environmental Sciences
Erlei Zhang, Jiayi Zhang, Jiaxin Bai, Jiarong Bian, Shaoyi Fang, Tao Zhan, Mingchen Feng
Summary: In this paper, an attention-embedded triple-branch fusion convolutional neural network (AETF-Net) is proposed for hyperspectral image (HSI) classification. The network consists of a spectral attention branch, a spatial attention branch, and a multi-attention fusion branch (MAFB). Experimental results demonstrate the superiority of the proposed AETF-Net on three well-known datasets.
Article
Geochemistry & Geophysics
Haimiao Ge, Liguo Wang, Moqi Liu, Xiaoyu Zhao, Yuexia Zhu, Haizhu Pan, Yanzhong Liu
Summary: This article proposes a pyramidal multiscale spectral-spatial convolutional network with polarized self-attention for pixel-wise HSI classification. The network consists of three stages: channel-wise feature extraction network, spatial-wise feature extraction network, and classification network. Experimental results demonstrate that the proposed network outperforms other related methods on several public HSI datasets.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Geochemistry & Geophysics
Ying Cui, Chao Shao, Li Luo, Liguo Wang, Shan Gao, Liwei Chen
Summary: This paper introduces the basic task of hyperspectral image (HSI) classification and the recent research progress of convolutional neural network (CNN) and graph convolution neural network (GCN) in this field. The authors propose a cooperative network that combines CNN and GraphSAGE, and extract features through superpixels and CNN with center attention. The advantages of this method are validated through experiments.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Computer Science, Artificial Intelligence
Gizem Ortac, Giyasettin Ozcan
Summary: This study utilizes deep learning approaches, specifically Convolutional Neural Networks, to analyze hyperspectral images in multiple dimensions and achieves higher classification accuracy rates, converging to 100% accuracy.
EXPERT SYSTEMS WITH APPLICATIONS
(2021)
Article
Agronomy
Shuo Chen, Kefei Zhang, Yindi Zhao, Yaqin Sun, Wei Ban, Yu Chen, Huifu Zhuang, Xuewei Zhang, Jinxiang Liu, Tao Yang
Summary: Rice bacterial leaf streak is a serious disease that affects the quality and quantity of rice growth, making automatic estimation of disease severity crucial in agricultural production. The proposed BLSNet method, utilizing attention mechanisms and multi-scale extraction in UNet network, demonstrated higher segmentation accuracy compared to benchmark models. Further investigations suggest the strong potential of BLSNet as a reliable automatic estimator of BLS disease severity based on segmentation results.
Article
Environmental Sciences
Xuewei Zhang, Kefei Zhang, Yaqin Sun, Yindi Zhao, Huifu Zhuang, Wei Ban, Yu Chen, Erjiang Fu, Shuo Chen, Jinxiang Liu, Yumeng Hao
Summary: This study investigates the effects of combining spectral and texture features extracted from unmanned aerial systems (UAS) multispectral imagery on maize leaf area index (LAI) estimation. The results show that combining spectral and texture features improves the accuracy of maize LAI estimation.