RFA-Net: Reconstructed Feature Alignment Network for Domain Adaptation Object Detection in Remote Sensing Imagery

With the development of deep learning, great progress has been made in object detection remote sensing (RS) imagery. However, detector is hard to generalize well from one labeled dataset (source domain) another unlabeled (target due discrepancy data distribution (domain shift). Currently, adversarial-based domain adaptation methods align semantic features source and target alleviate shift. But they fail avoid alignment noisy background neglect instance-level features, which are inappropriate for models that focus on instance location classification. To mitigate shift existing detection, we propose a reconstructed feature network (RFA-Net) unsupervised cross-domain The RFA-Net includes sequential augmentation module (SDA) deployed level providing solid gains data, sparse reconstruction (SFR) intensify alignment, pseudo-label generation (PLG) label supervision domain. Extensive experiments illustrate our proposed effective problem RS