Unsupervised Cross-Scene Aerial Image Segmentation via Spectral Space Transferring and Pseudo-Label Revising

Unsupervised domain adaptation (UDA) is essential since manually labeling pixel-level annotations consuming and expensive. Since the discrepancies have not been well solved, existing UDA approaches yield poor performance compared with supervised learning approaches. In this paper, we propose a novel sequential network (SLNet) for unsupervised cross-scene aerial image segmentation. The whole system decoupled into two parts—the translation model segmentation model. Specifically, introduce spectral space transferring (SST) approach to narrow visual discrepancy. high-frequency components between source images translated can be transferred in Fourier better preserving important identity fine-grained details. To further alleviate distribution discrepancy, an efficient pseudo-label revising (PLR) was developed guide via entropy minimization. Without additional parameters, map works as adaptive threshold, constantly pseudo labels target domain. Furthermore, numerous experiments single-category multi-category demonstrate that our SLNet state-of-the-art.