Locality Preserving Property Constrained Contrastive Learning for Object Classification in SAR Imagery
Robust unsupervised feature learning is a critical yet tough task for synthetic aperture radar (SAR) automatic target recognition (ATR) with limited labeled data. The developing contrastive self-supervised learning (CSL) method, which learns informative representations by solving an instance discrim...
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2023-07-01
|
| Series: | Remote Sensing |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2072-4292/15/14/3697 |
| _version_ | 1797587625928818688 |
|---|---|
| author | Jing Wang Sirui Tian Xiaolin Feng Bo Zhang Fan Wu Hong Zhang Chao Wang |
| author_facet | Jing Wang Sirui Tian Xiaolin Feng Bo Zhang Fan Wu Hong Zhang Chao Wang |
| author_sort | Jing Wang |
| collection | DOAJ |
| description | Robust unsupervised feature learning is a critical yet tough task for synthetic aperture radar (SAR) automatic target recognition (ATR) with limited labeled data. The developing contrastive self-supervised learning (CSL) method, which learns informative representations by solving an instance discrimination task, provides a novel method for learning discriminative features from unlabeled SAR images. However, the instance-level contrastive loss can magnify the differences between samples belonging to the same class in the latent feature space. Therefore, CSL can dispel these targets from the same class and affect the downstream classification tasks. In order to address this problem, this paper proposes a novel framework called locality preserving property constrained contrastive learning (LPPCL), which not only learns informative representations of data but also preserves the local similarity property in the latent feature space. In LPPCL, the traditional InfoNCE loss of the CSL models is reformulated in a cross-entropy form where the local similarity of the original data is embedded as pseudo labels. Furthermore, the traditional two-branch CSL architecture is extended to a multi-branch structure, improving the robustness of models trained with limited batch sizes and samples. Finally, the self-attentive pooling module is used to replace the global average pooling layer that is commonly used in most of the standard encoders, which provides an adaptive method for retaining information that benefits downstream tasks during the pooling procedure and significantly improves the performance of the model. Validation and ablation experiments using MSTAR datasets found that the proposed framework outperformed the classic CSL method and achieved state-of-the-art (SOTA) results. |
| first_indexed | 2024-03-11T00:41:34Z |
| format | Article |
| id | doaj.art-f7639c6cdb1947a7bc89f2ed0e73bf88 |
| institution | Directory Open Access Journal |
| issn | 2072-4292 |
| language | English |
| last_indexed | 2024-03-11T00:41:34Z |
| publishDate | 2023-07-01 |
| publisher | MDPI AG |
| record_format | Article |
| series | Remote Sensing |
| spelling | doaj.art-f7639c6cdb1947a7bc89f2ed0e73bf882023-11-18T21:14:25ZengMDPI AGRemote Sensing2072-42922023-07-011514369710.3390/rs15143697Locality Preserving Property Constrained Contrastive Learning for Object Classification in SAR ImageryJing Wang0Sirui Tian1Xiaolin Feng2Bo Zhang3Fan Wu4Hong Zhang5Chao Wang6School of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing 210094, ChinaSchool of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing 210094, ChinaSchool of Electronic and Optical Engineering, Nanjing University of Science and Technology, Nanjing 210094, ChinaInternational Research Center of Big Data for Sustainable Development Goals, Beijing 100094, ChinaInternational Research Center of Big Data for Sustainable Development Goals, Beijing 100094, ChinaInternational Research Center of Big Data for Sustainable Development Goals, Beijing 100094, ChinaInternational Research Center of Big Data for Sustainable Development Goals, Beijing 100094, ChinaRobust unsupervised feature learning is a critical yet tough task for synthetic aperture radar (SAR) automatic target recognition (ATR) with limited labeled data. The developing contrastive self-supervised learning (CSL) method, which learns informative representations by solving an instance discrimination task, provides a novel method for learning discriminative features from unlabeled SAR images. However, the instance-level contrastive loss can magnify the differences between samples belonging to the same class in the latent feature space. Therefore, CSL can dispel these targets from the same class and affect the downstream classification tasks. In order to address this problem, this paper proposes a novel framework called locality preserving property constrained contrastive learning (LPPCL), which not only learns informative representations of data but also preserves the local similarity property in the latent feature space. In LPPCL, the traditional InfoNCE loss of the CSL models is reformulated in a cross-entropy form where the local similarity of the original data is embedded as pseudo labels. Furthermore, the traditional two-branch CSL architecture is extended to a multi-branch structure, improving the robustness of models trained with limited batch sizes and samples. Finally, the self-attentive pooling module is used to replace the global average pooling layer that is commonly used in most of the standard encoders, which provides an adaptive method for retaining information that benefits downstream tasks during the pooling procedure and significantly improves the performance of the model. Validation and ablation experiments using MSTAR datasets found that the proposed framework outperformed the classic CSL method and achieved state-of-the-art (SOTA) results.https://www.mdpi.com/2072-4292/15/14/3697synthetic aperture radar (SAR)automatic target recognition (ATR)contrastive self-supervised learning (CSL)instance-level contrastive lossnoise-induced estimation of mutual information (InfoNCE)locality preserving projections (LPP) |
| spellingShingle | Jing Wang Sirui Tian Xiaolin Feng Bo Zhang Fan Wu Hong Zhang Chao Wang Locality Preserving Property Constrained Contrastive Learning for Object Classification in SAR Imagery Remote Sensing synthetic aperture radar (SAR) automatic target recognition (ATR) contrastive self-supervised learning (CSL) instance-level contrastive loss noise-induced estimation of mutual information (InfoNCE) locality preserving projections (LPP) |
| title | Locality Preserving Property Constrained Contrastive Learning for Object Classification in SAR Imagery |
| title_full | Locality Preserving Property Constrained Contrastive Learning for Object Classification in SAR Imagery |
| title_fullStr | Locality Preserving Property Constrained Contrastive Learning for Object Classification in SAR Imagery |
| title_full_unstemmed | Locality Preserving Property Constrained Contrastive Learning for Object Classification in SAR Imagery |
| title_short | Locality Preserving Property Constrained Contrastive Learning for Object Classification in SAR Imagery |
| title_sort | locality preserving property constrained contrastive learning for object classification in sar imagery |
| topic | synthetic aperture radar (SAR) automatic target recognition (ATR) contrastive self-supervised learning (CSL) instance-level contrastive loss noise-induced estimation of mutual information (InfoNCE) locality preserving projections (LPP) |
| url | https://www.mdpi.com/2072-4292/15/14/3697 |
| work_keys_str_mv | AT jingwang localitypreservingpropertyconstrainedcontrastivelearningforobjectclassificationinsarimagery AT siruitian localitypreservingpropertyconstrainedcontrastivelearningforobjectclassificationinsarimagery AT xiaolinfeng localitypreservingpropertyconstrainedcontrastivelearningforobjectclassificationinsarimagery AT bozhang localitypreservingpropertyconstrainedcontrastivelearningforobjectclassificationinsarimagery AT fanwu localitypreservingpropertyconstrainedcontrastivelearningforobjectclassificationinsarimagery AT hongzhang localitypreservingpropertyconstrainedcontrastivelearningforobjectclassificationinsarimagery AT chaowang localitypreservingpropertyconstrainedcontrastivelearningforobjectclassificationinsarimagery |