Answers are grounded in this paper and cite their sources.
High-resolution remote sensing image (HRRSI) semantic segmentation plays a crucial role in urban analysis and environmental monitoring. Although multimodal approaches that combine RGB imagery and digital surface models (DSM) have shown promising performance, they still face specific challenges, including insufficient interaction and alignment between RGB spectral–textural cues and DSM geometric and elevation cues across hierarchical encoder stages, as well as the progressive degradation of small objects and boundary-sensitive regions during decoding. To address these issues, this article proposes MMGC-AttnNet, a multilevel multimodal cross-attention network for HRRSI semantic segmentation. Rather than introducing isolated feature enhancement modules, the proposed framework is designed to establish a collaborative mechanism that couples hierarchical cross-modal interaction at the encoder side with stage-aware adaptive feature redistribution at the decoder side. Specifically, a multiscale cross-fusion module is designed to enhance multimodal representation learning by jointly modeling intramodal contextual consistency and intermodal complementary dependency, thereby improving the alignment of spectral–textural and geometric features. In addition, a decoder-aware adaptive feature fusion module is introduced to explicitly model the dependency between adjacent decoder-stage features and perform adaptive channel recalibration, which alleviates semantic dominance during progressive decoding and enhances the preservation of fine structures and small objects. Extensive experiments on the ISPRS Vaihingen and Potsdam datasets demonstrate that the proposed method achieves competitive performance compared with state-of-the-art approaches, particularly in boundary-sensitive and structurally complex urban scenarios.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: