Deep Prototypical Networks With Hybrid Residual Attention for Hyperspectral Image Classification
Recently, convolutional neural networks (CNNs) have attracted enormous attention in pattern recognition and demonstrated excellent performance in hyperspectral image (HSI) classification. However, high-dimensional HSI dataset versus limited training samples is easy to cause the overfitting phenomeno...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2020-01-01
|
| Series: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/9126161/ |
| _version_ | 1818890093645004800 |
|---|---|
| author | Bobo Xi Jiaojiao Li Yunsong Li Rui Song Yanzi Shi Songlin Liu Qian Du |
| author_facet | Bobo Xi Jiaojiao Li Yunsong Li Rui Song Yanzi Shi Songlin Liu Qian Du |
| author_sort | Bobo Xi |
| collection | DOAJ |
| description | Recently, convolutional neural networks (CNNs) have attracted enormous attention in pattern recognition and demonstrated excellent performance in hyperspectral image (HSI) classification. However, high-dimensional HSI dataset versus limited training samples is easy to cause the overfitting phenomenon in deep neural networks. Additionally, the intraclass distance of the embedding features extracted through the softmax-based CNNs may be greater than that of the interclass, which makes it difficult to further improve the classification accuracy. To address these issues, this article proposes a deep prototypical network with hybrid residual attention, which can effectively investigate the spectral-spatial information in the HSI. Specifically, in order to improve the generalization capability of the model, feature extraction with a hybrid residual attention module is presented to enhance the critical spectral-spatial features and suppress the useless ones in the classification task. Furthermore, a novel discriminant distance-based cross-entropy loss is proposed to increase the intraclass compactness, to obtain more superior results. Extensive experiments on three benchmark datasets are carried out to convincingly evaluate the proposed framework. With the generation of optimal prototypes representing each class and more discriminative embedding features, encouraging classification results are achieved compared with state-of-the-art methods. |
| first_indexed | 2024-12-19T17:19:26Z |
| format | Article |
| id | doaj.art-e0034a40f63d42269bed5a5bb65cd4cd |
| institution | Directory Open Access Journal |
| issn | 2151-1535 |
| language | English |
| last_indexed | 2024-12-19T17:19:26Z |
| publishDate | 2020-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
| spelling | doaj.art-e0034a40f63d42269bed5a5bb65cd4cd2022-12-21T20:12:44ZengIEEEIEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing2151-15352020-01-01133683370010.1109/JSTARS.2020.30049739126161Deep Prototypical Networks With Hybrid Residual Attention for Hyperspectral Image ClassificationBobo Xi0https://orcid.org/0000-0002-0309-1556Jiaojiao Li1https://orcid.org/0000-0002-0470-9469Yunsong Li2https://orcid.org/0000-0002-0234-6270Rui Song3Yanzi Shi4https://orcid.org/0000-0002-7717-985XSonglin Liu5Qian Du6https://orcid.org/0000-0001-8354-7500School of Telecommunications Engineering, State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an, ChinaSchool of Telecommunications Engineering, State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an, ChinaSchool of Telecommunications Engineering, State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an, ChinaSchool of Telecommunications Engineering, State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an, ChinaSchool of Telecommunications Engineering, State Key Laboratory of Integrated Service Networks, Xidian University, Xi’an, ChinaState Key Laboratory of Geo-Information Engineering, Xi’an Research Institute of Surveying and Mapping, Xi’an, ChinaDepartment of Electronic and Computer Engineering, Mississippi State University, Starkville, MS, USARecently, convolutional neural networks (CNNs) have attracted enormous attention in pattern recognition and demonstrated excellent performance in hyperspectral image (HSI) classification. However, high-dimensional HSI dataset versus limited training samples is easy to cause the overfitting phenomenon in deep neural networks. Additionally, the intraclass distance of the embedding features extracted through the softmax-based CNNs may be greater than that of the interclass, which makes it difficult to further improve the classification accuracy. To address these issues, this article proposes a deep prototypical network with hybrid residual attention, which can effectively investigate the spectral-spatial information in the HSI. Specifically, in order to improve the generalization capability of the model, feature extraction with a hybrid residual attention module is presented to enhance the critical spectral-spatial features and suppress the useless ones in the classification task. Furthermore, a novel discriminant distance-based cross-entropy loss is proposed to increase the intraclass compactness, to obtain more superior results. Extensive experiments on three benchmark datasets are carried out to convincingly evaluate the proposed framework. With the generation of optimal prototypes representing each class and more discriminative embedding features, encouraging classification results are achieved compared with state-of-the-art methods.https://ieeexplore.ieee.org/document/9126161/HSI classificationhybrid residual attentionprototypical networksspectral–spatial information |
| spellingShingle | Bobo Xi Jiaojiao Li Yunsong Li Rui Song Yanzi Shi Songlin Liu Qian Du Deep Prototypical Networks With Hybrid Residual Attention for Hyperspectral Image Classification IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing HSI classification hybrid residual attention prototypical networks spectral–spatial information |
| title | Deep Prototypical Networks With Hybrid Residual Attention for Hyperspectral Image Classification |
| title_full | Deep Prototypical Networks With Hybrid Residual Attention for Hyperspectral Image Classification |
| title_fullStr | Deep Prototypical Networks With Hybrid Residual Attention for Hyperspectral Image Classification |
| title_full_unstemmed | Deep Prototypical Networks With Hybrid Residual Attention for Hyperspectral Image Classification |
| title_short | Deep Prototypical Networks With Hybrid Residual Attention for Hyperspectral Image Classification |
| title_sort | deep prototypical networks with hybrid residual attention for hyperspectral image classification |
| topic | HSI classification hybrid residual attention prototypical networks spectral–spatial information |
| url | https://ieeexplore.ieee.org/document/9126161/ |
| work_keys_str_mv | AT boboxi deepprototypicalnetworkswithhybridresidualattentionforhyperspectralimageclassification AT jiaojiaoli deepprototypicalnetworkswithhybridresidualattentionforhyperspectralimageclassification AT yunsongli deepprototypicalnetworkswithhybridresidualattentionforhyperspectralimageclassification AT ruisong deepprototypicalnetworkswithhybridresidualattentionforhyperspectralimageclassification AT yanzishi deepprototypicalnetworkswithhybridresidualattentionforhyperspectralimageclassification AT songlinliu deepprototypicalnetworkswithhybridresidualattentionforhyperspectralimageclassification AT qiandu deepprototypicalnetworkswithhybridresidualattentionforhyperspectralimageclassification |