Aug 2026· Italian National Conference on Sensors· Vol 26, pp. 5197· 0 citations· 30 references
Medicine
TL;DR
MAIG-Net combines a target-label-free ground-sampling-distance rule, an intermediate domain constructed by Fourier domain adaptation (FDA) that transfers only low-frequency target appearance onto labeled source images while preserving the complete source phase and road labels, and Domain-Invariant Feature Alignment modules that perform reliability-weighted, topology-conditioned adversarial alignment at three encoder depths.
Abstract
In unsupervised cross-domain remote sensing road extraction, severe data distribution shifts and heterogeneous background interference significantly constrain model performance. In addition to appearance shift, source and target images acquired by different sensors often present roads at unequal physical scales. To address this, MAIG-Net combines a target-label-free ground-sampling-distance (GSD) rule that matches the physical field of view of the two domains, an intermediate domain constructed by Fourier domain adaptation (FDA) that transfers only low-frequency target appearance onto labeled source images while preserving the complete source phase and road labels, and Domain-Invariant Feature Alignment (DIFA) modules that perform reliability-weighted, topology-conditioned adversarial alignment at three encoder depths. An exponential-moving-average (EMA) teacher supplies detached reliability and topology conditions for the domain discriminators; no target prediction is used as a direct segmentation label. Under a matched protocol with a fixed 30-epoch budget and three seeds, MAIG-Net improves the mean road IoU from 0.327 to 0.376 on SpaceNet→DeepGlobe and from 0.450 to 0.458 on SpaceNet→Massachusetts relative to source-only training, with consistent per-seed gains in both directions; the smaller Massachusetts effect is directionally reproduced on a previously untouched validation holdout. Reverse adaptation, reliability perturbations, computational cost, and failure cases are further reported to delimit the applicability of the method.
Deep learning has boosted remote sensing (RS) scene classification, but adversarial examples can still cause high-confidence misclassification with imperceptible perturbations. Adversarial purification (AP) offers a practical test-time defense without retraining the classifier. However, most existing methods are confined to pixel-space restoration, which may leave residual adversarial effects that persist and amplify through feature extraction, ultimately biasing the prediction. To address these issues, a dual-domain AP (DDAP) framework is proposed to mitigate adversarial effects at both the pixel and feature levels in a unified pipeline. In the pixel domain, a pixel-domain frequency-aware diffusion purification (PFDP) module performs diffusion-based restoration through a frequency-aware dual-stream U-Net (FD-UNet). By integrating adaptive spectral filtering with multidomain consistency constraints, PFDP reduces adversarial-perturbation-dominated high-frequency responses while preserving structural details and semantic information in RS imagery. In the feature domain, an adversarial vulnerable channel dropout (AVCD) strategy models unshifted shallow-feature statistics with a Gaussian mixture model (GMM) and adaptively assigns channelwise dropout probabilities based on a samplewise shift score and channel vulnerability, thereby suppressing residual adversarial influence before downstream classification. Extensive experiments on UC Merced (UCM) and aerial image dataset (AID) across multiple backbones and attack types demonstrate that DDAP consistently improves robustness while maintaining a favorable clean–robust balance compared with representative baselines.
Yuru Su, Shaohui Mei, Mingyang Ma et al.· IEEE Transactions on Geoscie...· 0 citations
Cross-domain remote sensing scene classification suffers from severe domain shifts caused by sensor and geographic variations. Existing adversarial domain adaptation methods mainly align global feature distributions but often neglect category-level structure, leading to ambiguous decision boundaries in the target domain. In this paper, we propose a neighborhood geometry–guided prototype contrastive adaptation (NGPCA) framework built upon the domain-adversarial neural network. The proposed method leverages local neighborhood geometry in the target feature space to construct neighborhood-consensus pseudo labels and further introduces prototype contrastive learning to align target features with source-domain class prototypes. By combining neighborhood consensus supervision with prototype-level contrastive adaptation, NGPCA improves pseudo-label robustness and promotes more discriminative feature learning. Extensive experiments on University of California, Merced land-use data set, the Aerial Image Data Set, and the NWPU-RESISC45 data set with six cross-domain transfer tasks demonstrate the effectiveness of the proposed framework. NGPCA achieves the best average accuracy of 98.26% among the compared remote sensing–oriented and general domain adaptation methods, showing robust performance for cross-domain remote sensing scene classification
Domain shifts caused by varying remote sensing sensors severely degrade the model performance of object detection when deployed in target domains. While unsupervised domain adaptation provides a promising solution, existing methods tend to suffer from pseudolabel noise accumulation, leading to unstable and unreliable detection in complex remote sensing scenarios. To address both domain discrepancy and pseudolabel degradation, we propose ICPNet, an Iterative Consistency Pseudo-Labeling Network that integrates self-supervised learning with pseudolabel refinement to enhance domain robustness and label reliability. Specifically, we design an adversarial teacher-student framework enhanced with an Auxiliary Masked Autoencoder to learn domain-invariant representations by randomly masking feature regions and enforcing cross-domain reconstruction. In addition, we introduce an Iterative Consistency Pseudolabel Filter (ICPF) that leverages a memory of historical predictions. By filtering labels based on their stability across multiple training iterations, the ICPF generates a high-quality and stable supervisory information. A confidence decay mechanism further enhances stability by reducing the influence of outdated predictions. Extensive experiments on diverse remote sensing datasets demonstrate that our ICPNet outperforms state-of-the-art unsupervised domain adaptation (UDA) approaches in cross-domain object detection.
Shu-ming Zhuang, Yongxing Hou, Xinlei Hu et al.· IEEE Journal of Selected Top...· 0 citations
Remote sensing change detection (RSCD) aims to identify and localize changes in the same geographical region using bi-temporal or multitemporal images. However, significant feature distribution shifts commonly exist not only between training and real-world data but also between paired images acquired at different times within the same dataset, severely limiting model generalization. Domain-generalized RSCD seeks to learn domain-invariant representations from source domains (seen), enabling direct deployment to target domains (unseen) without requiring target-domain data during training. Existing methods typically attribute performance degradation to style discrepancies and attempt to suppress style variations through feature regularization. However, because style and content information are highly coupled, such strategies often discard critical content representations. In addition, style discrepancies across datasets can further introduce change-domain shifts, which are largely overlooked by existing methods. To address these issues, this article proposes a domain generalization network for RSCD that enables models trained solely on a source domain to generalize effectively to target domains. Specifically, a feature constraint (FC) mechanism is introduced at the encoder stage to mitigate style interference through covariance alignment while preserving essential content information. At the decoder stage, a cross-domain learning (CDL) module is designed to construct a more discriminative embedding space and separate features prone to misclassification. Extensive experiments demonstrate that the proposed method achieves strong robustness and superior detection accuracy across both source and target domains, significantly outperforming existing methods.
Jiahang Liu, Zitong Qi, Mao-yin Guo et al.· IEEE Transactions on Geoscie...· 0 citations
Remote sensing object detection suffers from severe performance degradation under cross-domain transfer, where domain gaps arise from differences in spectral response, spatial resolution, and viewing geometry. Most existing unsupervised domain-adaptive object detection (DAOD) methods pursue cross-domain invariance through feature distribution alignment but encounter two limitations specific to remote sensing: cross-domain appearance variation, where differences in imaging conditions produce divergent visual appearances, and foreground-background imbalance, where targets are sparsely distributed across vast backgrounds and alignment is dominated by background statistics. To address these two limitations, frequency-spatial dual-level selective alignment (FS2A), a framework with two complementary modules, is proposed. At the frequency level, scale-class-conditioned frequency modulation (SCFM) decomposes multiscale features via FFT and selectively modulates the low-frequency amplitude conditioned on scale level and categorical composition, restricting adversarial alignment to domain-variant spectral components while preserving the domain-invariant phase spectrum. At the spatial level, kernel relational distillation (KRD) distills pairwise relational structure from a frozen satellite-pretrained vision foundation model (VFM) in polynomial kernel space, where polynomial kernel functions preferentially concentrate alignment on foreground feature pairs over weakly correlated background pairs. Both modules are decoupled from the detection forward pass, introducing no additional inference cost. Experiments on two cross-domain remote sensing benchmarks demonstrate that FS2A achieves 67.6% mAP50 on xView $\rightarrow $ DOTA, surpassing the state-of-the-art by 2.7%, and competitive results on satellite-to-UAV benchmarks. The code will be available at https://github.com/sparklejojo/FSSA-DAOD
Tingting Qiao, He Chen, Jue Wang et al.· IEEE Transactions on Geoscie...· 0 citations