Skip to content

Text-Anchored Deep Optimal Transport Clustering for Remote Sensing Cross-Scene Classification

2026 · IEEE Transactions on Geoscience and Remote Sensing · Vol 64, pp. 5648718-5648718 · 0 citations · 64 references

Abstract

Remote sensing scene classification (RSSC) under cross-scene settings aims to transfer semantic knowledge learned from a labeled source domain to an unlabeled target domain that exhibits substantial distribution shifts caused by variations in geography, sensors, and imaging conditions. Existing adaptation methods mainly rely on image-level feature alignment, which leads to weak semantic understanding and consistency across domains and poor generalization in complex remote sensing (RS) scenarios. In this article, we propose a text-anchored deep optimal transport clustering network (TA-DOTCNet) to explicitly introduce category-level textual semantics as external guidance for cross-scene representation learning. The core idea is to leverage domain-invariant textual descriptions to regulate visual adaptation and enforce semantically consistent clustering across heterogeneous domains. First, a text-anchored external guidance strategy (TaGS) integrates textual semantics into visual representation learning through extended prompt construction, cross-scene alignment, and prototype-driven counterpart fusion, enhancing discriminability while preserving semantic coherence across domains. Second, a deep optimal transport clustering (DOTC) module formulates image–text alignment as an entropy-regularized optimal transport (EOT) problem, transforming similarity-based assignments into compact target distributions and enforcing domain-invariant clustering structures. Our proposal effectively mitigates domain bias and improves cross-scene generalization. Extensive experiments on multiple RS benchmarks demonstrate the superiority and robustness of the proposed method over state-of-the-art approaches. The source code is available at https://github.com/bling2beyond/TA-DOTCNet

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.