Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification
Recently, state-of-the-art classification performance of natural images has been obtained by self-supervised learning (S2L) as it can generate latent features through learning between different views of the same images. However, the latent semantic information of similar images has hardly been explo...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-03-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/15/6/1713 |
| _version_ | 1797609119509643264 |
|---|---|
| author | Yao Qin Yuanxin Ye Yue Zhao Junzheng Wu Han Zhang Kenan Cheng Kun Li |
| author_facet | Yao Qin Yuanxin Ye Yue Zhao Junzheng Wu Han Zhang Kenan Cheng Kun Li |
| author_sort | Yao Qin |
| collection | DOAJ |
| description | Recently, state-of-the-art classification performance of natural images has been obtained by self-supervised learning (S2L) as it can generate latent features through learning between different views of the same images. However, the latent semantic information of similar images has hardly been exploited by these S2L-based methods. Consequently, to explore the potential of S2L between similar samples in hyperspectral image classification (HSIC), we propose the nearest neighboring self-supervised learning (N2SSL) method, by interacting between different augmentations of reliable nearest neighboring pairs (RN2Ps) of HSI samples in the framework of bootstrap your own latent (BYOL). Specifically, there are four main steps: pretraining of spectral spatial residual network (SSRN)-based BYOL, generation of nearest neighboring pairs (N2Ps), training of BYOL based on RN2P, final classification. Experimental results of three benchmark HSIs validated that S2L on similar samples can facilitate subsequent classification. Moreover, we found that BYOL trained on an un-related HSI can be fine-tuned for classification of other HSIs with less computational cost and higher accuracy than training from scratch. Beyond the methodology, we present a comprehensive review of HSI-related data augmentation (DA), which is meaningful to future research of S2L on HSIs. |
| first_indexed | 2024-03-11T05:57:03Z |
| format | Article |
| id | doaj.art-2f4a1aea454649f8aef9cdc40e8a7e1c |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-11T05:57:03Z |
| publishDate | 2023-03-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-2f4a1aea454649f8aef9cdc40e8a7e1c2023-11-17T13:40:56ZengMDPI AGRemote Sensing2072-42922023-03-01156171310.3390/rs15061713Nearest Neighboring Self-Supervised Learning for Hyperspectral Image ClassificationYao Qin0Yuanxin Ye1Yue Zhao2Junzheng Wu3Han Zhang4Kenan Cheng5Kun Li6Northwest Institute of Nuclear Technology, Xi’an 710024, ChinaFaculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, Chengdu 610031, ChinaSchool of Computer Science and Information Engineering, Hubei University, Wuhan 430062, ChinaNorthwest Institute of Nuclear Technology, Xi’an 710024, ChinaNorthwest Institute of Nuclear Technology, Xi’an 710024, ChinaNorthwest Institute of Nuclear Technology, Xi’an 710024, ChinaCollege of Electronic Science, National University of Defense Technology, Changsha 410073, ChinaRecently, state-of-the-art classification performance of natural images has been obtained by self-supervised learning (S2L) as it can generate latent features through learning between different views of the same images. However, the latent semantic information of similar images has hardly been exploited by these S2L-based methods. Consequently, to explore the potential of S2L between similar samples in hyperspectral image classification (HSIC), we propose the nearest neighboring self-supervised learning (N2SSL) method, by interacting between different augmentations of reliable nearest neighboring pairs (RN2Ps) of HSI samples in the framework of bootstrap your own latent (BYOL). Specifically, there are four main steps: pretraining of spectral spatial residual network (SSRN)-based BYOL, generation of nearest neighboring pairs (N2Ps), training of BYOL based on RN2P, final classification. Experimental results of three benchmark HSIs validated that S2L on similar samples can facilitate subsequent classification. Moreover, we found that BYOL trained on an un-related HSI can be fine-tuned for classification of other HSIs with less computational cost and higher accuracy than training from scratch. Beyond the methodology, we present a comprehensive review of HSI-related data augmentation (DA), which is meaningful to future research of S2L on HSIs.https://www.mdpi.com/2072-4292/15/6/1713self-supervised learningnearest neighboringhyperspectral image classification (HSIC)data augmentation (DA)bootstrap your own latent (BYOL)spectral spatial residual network (SSRN) |
| spellingShingle | Yao Qin Yuanxin Ye Yue Zhao Junzheng Wu Han Zhang Kenan Cheng Kun Li Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification Remote Sensing self-supervised learning nearest neighboring hyperspectral image classification (HSIC) data augmentation (DA) bootstrap your own latent (BYOL) spectral spatial residual network (SSRN) |
| title | Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification |
| title_full | Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification |
| title_fullStr | Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification |
| title_full_unstemmed | Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification |
| title_short | Nearest Neighboring Self-Supervised Learning for Hyperspectral Image Classification |
| title_sort | nearest neighboring self supervised learning for hyperspectral image classification |
| topic | self-supervised learning nearest neighboring hyperspectral image classification (HSIC) data augmentation (DA) bootstrap your own latent (BYOL) spectral spatial residual network (SSRN) |
| url | https://www.mdpi.com/2072-4292/15/6/1713 |
| work_keys_str_mv | AT yaoqin nearestneighboringselfsupervisedlearningforhyperspectralimageclassification AT yuanxinye nearestneighboringselfsupervisedlearningforhyperspectralimageclassification AT yuezhao nearestneighboringselfsupervisedlearningforhyperspectralimageclassification AT junzhengwu nearestneighboringselfsupervisedlearningforhyperspectralimageclassification AT hanzhang nearestneighboringselfsupervisedlearningforhyperspectralimageclassification AT kenancheng nearestneighboringselfsupervisedlearningforhyperspectralimageclassification AT kunli nearestneighboringselfsupervisedlearningforhyperspectralimageclassification |