DILRS: Domain-Incremental Learning for Semantic Segmentation in Multi-Source Remote Sensing Data

With the exponential growth in speed and volume of remote sensing data, deep learning models are expected to adapt continually learn over time. Unfortunately, domain shift between multi-source data from various sensors regions poses a significant challenge. Segmentation face difficulty adapting incremental domains due catastrophic forgetting, which can be addressed via methods. However, current methods mainly focus on class-incremental learning, wherein classes belong same domain, neglect investigations into sensing. To solve this problem, we propose domain-incremental method for semantic segmentation data. Specifically, our model aims incrementally new while preserving its performance previous without accessing achieve this, has unique parameter structure that reparametrizes domain-agnostic domain-specific parameters. We use different optimization strategies learning. Additionally, adopt multi-level knowledge distillation loss mitigate impact label space among domains. The experiments demonstrate achieves excellent settings, outperforming existing with only few