Transactions of Nanjing University of Aeronautics & Astronautics
Recognition of Oscillatory Ships in Missile-Borne SAR
- Article
- Figures
- Metrics
- Preview PDF
- Reference
- Related
- Cited by
- Materials
Abstract:
An end-to-end recognition strategy is proposed for oscillatory ships in missile-borne synthetic aperture radar （SAR）， eliminating the need for image refocusing. Unlike conventional “focus-then-recognize” paradigm， the approach directly exploits oscillation-degraded SAR images for training and recognition， avoiding the unreliability of refocusing under complex imaging conditions. A multi-azimuth ship dataset under the “sea state five” condition is simulated， where ResNet-18 achieves a baseline accuracy of 66.66%， validating the feasibility of the end-to-end framework. By further incorporating a domain-adversarial neural network （DANN） to extract cross-azimuth invariant features， the recognition rate increases to 76.22%， demonstrating the potential of this strategy. The results indicate that， even with a non-optimal backbone， the end-to-end approach shows clear applicability in challenging scenarios， while offering a foundation for future performance gains with more advanced architectures.
Keywords:
Project Supported:
Clc Number:
TN925
Mobile website
