Deep Geospatio-Semantic Guided Network With Pseudo-Label Consistency for Domain-Adaptive Remote Sensing Segmentation

Domain-adaptive remote sensing image (RSI) semantic segmentation mitigates the overfitting problem that affects the effectiveness of segmentation, which results from the scarcity of high-quality labels and the cross-domain styles of ground objects. The effectiveness of domain-adaptive segmentation remains suboptimal in complex scenarios due to inadequate exploitation of latent geographic knowledge. Consequently, interclass ambiguity and boundary agnostic are further exacerbated under cross-domain transfer scenarios. To address this issue, we first devise a deep geospatio-semantic guided network named deep spatio-semantic awareness learning (DSSAL), which comprehensively investigates the potential spatial relationship and semantic correlation between classes of RSIs by geospatial-aware interaction and geosemantic-aware interaction, respectively. To mitigate class-wise cognitive deviation in the unlabeled domain, DSSAL domain adaptation (DSSAL-DA) is developed to further enhance the segmentation effect with the spatio-semantic domain alignment (SSDA) module in manifold cross-domain tasks. Furthermore, a pseudo-label consistency filter (PCF) is developed for DSSAL-DA to ensure reliability in self-training through cross-view consistency verification. Extensive experiments on two public datasets and a private dataset demonstrate the superiority of DSSAL and DSSAL-DA over the state-of-the-art (SOTA) methods for UDA semantic segmentation of RSIs.