Enhancing Few-Shot SAR Ship Recognition: Pseudospectrum Information Generation and Fusion
The limited number of samples in synthetic aperture radar (SAR) ship datasets hampers the advancement of target recognition performance using deep learning. Given the complex-valued nature of SAR data, incorporating spectrum information is beneficial for few-shot target recognition methods. However,...
| Published in: | IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing |
|---|---|
| Main Authors: | , , |
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10966213/ |
| Summary: | The limited number of samples in synthetic aperture radar (SAR) ship datasets hampers the advancement of target recognition performance using deep learning. Given the complex-valued nature of SAR data, incorporating spectrum information is beneficial for few-shot target recognition methods. However, the SAR ship domain faces two significant issues: a scarcity of datasets that include spectrum information and a lack of target recognition networks specifically designed to leverage this spectrum information. In order to solve the above problems, first, a SpecGenGANwith generating pseudospectrum information is proposed to solve the problem of missing spectrum information. Second, a SpecAmpFusionNet is designed to fully exploit the deep features of spectrum and amplitude information. Finally, a few-shot target recognition method based on pseudospectrum information generation and fusion network is presented, allowing flexibility and integration with various popular recognition methods. Experimental results demonstrate that under 3way-10shots and 5way-10shots conditions, our method improves average accuracies by 12.04% and 10.83%, respectively, compared to methods using only amplitude information, validating the effectiveness of our approach in enhancing few-shot SAR ship recognition. |
|---|---|
| ISSN: | 1939-1404 2151-1535 |