Terrain Classification with Airborne Multi-band PolSAR Data Based on Polarization Feature Registration

PolSAR data contains rich terrain scattering information, which can distinguish different scattering types and facilitate the differentiation and discrimination of terrain types. Fusion of SAR images of different bands can obtain richer information on scattering features. However, there are still difficulties with multi-band SAR image fusion at present due to the resolution differences, especially in the registration of polarization complex matrices involving the preservation of amplitude and phase information. In response to the above issues, this article proposes to extract and register key polarization features, instead of the complex scattering matrix of multi-band SAR data, using image registration algorithms, and finally achieves terrain classification through a high-performance Transformer network on registered features of multi-band. The experimental results show that the classification accuracy after the polarimetric feature registration and fusion of S, C, and Ka bands reaches 94.2%, which improves the accuracy by about 8% compared to the individual band.