Unsupervised Domain Adaptation Architecture Search with Self-Training for Land Cover Mapping

TL;DR

This work tackles the challenge of domain-shifted land cover mapping under resource constraints by marrying Markov random field neural architecture search (MRF-NAS) with a self-training unsupervised domain adaptation framework. It automatically discovers lightweight networks ($<2 ext{M}$ parameters, $ ext{FLOPs}\nolinebreak[4]=30.16$) that transfer from a labeled source domain to an unlabeled target domain, using a teacher-student scheme and two pseudo-labeling strategies (confidence- and energy-based) within a ClassMix-augmented training regime. The approach achieves competitive or state-of-the-art mIoU on two remote sensing benchmarks (OpenEarthMap target: $59.38 ext{ exttt{%}}$; FLAIR #1 target: $51.19 ext{ exttt{%}}$) while maintaining strong transferability across datasets, demonstrating the practicality of NAS-driven, self-training UDA for land cover mapping on resource-limited platforms. By enabling automatic architecture search under budget constraints, the method broadens the feasibility of deploying domain-adaptive RS models on CubeSats, drones, and mobile devices, and provides a blueprint for future RS-NAS-Domain Adaptation research.

Abstract

Unsupervised domain adaptation (UDA) is a challenging open problem in land cover mapping. Previous studies show encouraging progress in addressing cross-domain distribution shifts on remote sensing benchmarks for land cover mapping. The existing works are mainly built on large neural network architectures, which makes them resource-hungry systems, limiting their practical impact for many real-world applications in resource-constrained environments. Thus, we proposed a simple yet effective framework to search for lightweight neural networks automatically for land cover mapping tasks under domain shifts. This is achieved by integrating Markov random field neural architecture search (MRF-NAS) into a self-training UDA framework to search for efficient and effective networks under a limited computation budget. This is the first attempt to combine NAS with self-training UDA as a single framework for land cover mapping. We also investigate two different pseudo-labelling approaches (confidence-based and energy-based) in self-training scheme. Experimental results on two recent datasets (OpenEarthMap & FLAIR #1) for remote sensing UDA demonstrate a satisfactory performance. With only less than 2M parameters and 30.16 GFLOPs, the best-discovered lightweight network reaches state-of-the-art performance on the regional target domain of OpenEarthMap (59.38% mIoU) and the considered target domain of FLAIR #1 (51.19% mIoU). The code is at https://github.com/cliffbb/UDA-NAS}{https://github.com/cliffbb/UDA-NAS.

Figures (3)

• Figure 1: Self-training UDA with MRF-NAS process. The supernet can be any neural network. After the search, i.e., learning pairwise factors in MRF of the student supernet $\mathcal{G}^{\theta}_{stud}$, $m$ optimal subnets (lightweight networks) are inference over the learned factors via diverse M-best loopy inference payman2012. Then, we retrained the discovered lightweight networks. $\mathcal{L}_s$ and $\mathcal{L}_t$ are cross-entropy loss functions for the source labelled data and target pseudo-labelled data, respectively. The teacher supernet $\mathcal{G}^{\theta}_{teach}$ is only used to generate pseudo-labels for the target images to train $\mathcal{G}^{\theta}_{stud}$, in addition to the source labelled data. See Section \ref{['sec:3']} for more details.
• Figure 2: Visual comparison of land cover mapping results of the best-discovered network and some representative baselines in Table \ref{['tab:oem_search_results']}.
• Figure 3: Visual comparison of land cover mapping results of the best-discovered network and some representative baselines in Table \ref{['tab:flair_transfer_results']}. The land cover maps of the baselines, GeoMTNet and UDA_for_RS, were obtained from GeoMultiTaskNet marcocci2023.