Answers are grounded in this paper and cite their sources.
Semantic segmentation of very-high-resolution remote sensing (RS) data is foundational for many RS applications in urban environments. Images from sources like Worldview-3/4 and Pleiades-Neo offer resolutions as high as 0.3 m, enabling a fine-grained understanding of urban structures. Recent deep learning‐based methods have significantly outperformed traditional approaches in RS semantic segmentation/classification tasks. However, they require large training data sets and often lack transferability due to highly disparate RS image content across different geographical regions. However, no comprehensive analysis exists on their transferability—i.e., to what extent a model trained on a source domain can be applied to a target domain in urban areas. This paper investigates the raw transferability of traditional and deep-learning models and the effectiveness of domain adaptation approaches in enhancing deep-learning model transferability (adapted transferability). Using five highly diverse RS data sets from different cities (6792 patches of 1024 × 1024 pixels each), we trained six models with and without three domain adaptation approaches to quantitatively analyze transferability between data sets. To facilitate easy assessment of model transferability, we developed a simple method to quantify transferability using spectral indices as a medium, demonstrating its effectiveness in evaluating model transferability at the target domain when labels are unavailable. Our experiments yield several important but underreported observations on raw and adapted transferability. Moreover, our proposed label-free transferability assessment method outperforms posterior model confidence and can guide model selection for urban studies globally. The models and datasets are publicly available on GitHub at: https://github.com/GDAOSU/Transferability-Remote-Sensing.
No takes yet. Share an insight, caveat, or question.
Qin et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: