Label-Decoupled Style Augmentation for Domain Generalization in Multi-Label Remote Sensing Scene Classification

2026-07-14

AI Summary

This work addresses the limited generalization of multi-label remote sensing scene classification in cross-domain deployment, where existing global style augmentation methods often induce inter-class interference due to their neglect of label-specific discrepancies. To mitigate this, we propose a label-decoupled style augmentation framework that leverages label-aware spatial attention to extract localized feature statistics, enabling independent coefficient mixing for cross-domain samples sharing the same labels, followed by attention-weighted normalization for feature reconstruction. Our approach is the first to integrate label-specific attention into multi-label remote sensing domain generalization, achieving label-wise decoupled style perturbations that effectively prevent category contamination while introducing negligible parameters and preserving the original inference pipeline. On leave-one-domain-out benchmarks built from UCM, AID, and DFC15, our best variant achieves an average mAP of 71.5%, outperforming empirical risk minimization by 5.0 points, the strongest global baseline by 1.3 points, and yielding up to a 7.7-point gain in the most challenging transfer scenario.

Abstract

Multi-label classification assigns several co-occurring labels to each aerial scene, yet deployed models often encounter data distributions different from their training. Feature-statistics augmentation such as MixStyle, EFDMix, and correlated style uncertainty improves generalization at low cost but perturbs channel statistics globally, treating each image as a single style; one class can then contaminate the augmentation of another. Domain generalization is understudied for multi-label remote sensing; no prior method or multi-source benchmark targets it. A label-decoupled augmentation framework is therefore proposed, confining style perturbation to label-specific regions. Per-label attention, obtained from a learnable module or from gradient class-activation maps, yields per-label feature statistics; these statistics are mixed with cross-domain samples that share present labels, under independent per-label coefficients, and features are recomposed by attention-weighted normalization. Three operators combined with two attention sources produce six variants, evaluated on a leave-one-domain-out benchmark from multi-label UCM, AID, and DFC15 over six shared labels. Averaged over three splits and five seeds, the best variant attains 71.5% mean average precision, exceeding empirical risk minimization by 5.0 points and the strongest global-statistics baseline by 1.3 points, with the largest gain on the hardest transfer (up to 7.7 points). Ablations indicate that spatial attention and refreshed localization maps are most influential. The framework adds at most 0.35% parameters, leaves inference unchanged, and appears to offer a generic, inexpensive upgrade path for multi-label statistics-based domain generalization. Code is available upon acceptance at https://github.com/Alaa-Almouradi/Style-Augmentation-Upgrade.

Problem

• domain generalization
• multi-label classification
• remote sensing
• style augmentation
• label decoupling

Innovation

• label-decoupled augmentation
• domain generalization
• multi-label remote sensing
• style transfer
• attention-weighted normalization