SDA-Net: A Spatially Optimized Dual-Stream Network with Adaptive Global Attention for Building Extraction in Multi-Modal Remote Sensing Images DOI Creative Commons

Xuran Pan,

Kexing Xu,

Shuhao Yang

et al.

Sensors, Journal Year: 2025, Volume and Issue: 25(7), P. 2112 - 2112

Published: March 27, 2025

Building extraction plays a pivotal role in enabling rapid and accurate construction of urban maps, thereby supporting planning, smart city development, management. Buildings remote sensing imagery exhibit diverse morphological attributes spectral signatures, yet their reliable interpretation through single-modal data remains constrained by heterogeneous terrain conditions, occlusions, spatially variable illumination effects inherent to complex geographical landscapes. The integration multi-modal for building offers significant advantages leveraging complementary features from sources. However, the heterogeneity complicates effective feature extraction, while multi-scale cross-modal fusion encounters semantic gap issue. To address these challenges, novel network based on called SDA-les (AGAFMs) was designed decoding stage fuse at various scales, which dynamically adjust importance global perspective better balance information. superior performance proposed method is demonstrated comprehensive evaluations ISPRS Potsdam dataset with 97.66% F1 score 95.42% IoU, Vaihingen 96.56% 93.35% DFC23 Track2 91.35% 84.08% IoU.

Language: Английский

SDA-Net: A Spatially Optimized Dual-Stream Network with Adaptive Global Attention for Building Extraction in Multi-Modal Remote Sensing Images DOI Creative Commons

Xuran Pan,

Kexing Xu,

Shuhao Yang

et al.

Sensors, Journal Year: 2025, Volume and Issue: 25(7), P. 2112 - 2112

Published: March 27, 2025

Building extraction plays a pivotal role in enabling rapid and accurate construction of urban maps, thereby supporting planning, smart city development, management. Buildings remote sensing imagery exhibit diverse morphological attributes spectral signatures, yet their reliable interpretation through single-modal data remains constrained by heterogeneous terrain conditions, occlusions, spatially variable illumination effects inherent to complex geographical landscapes. The integration multi-modal for building offers significant advantages leveraging complementary features from sources. However, the heterogeneity complicates effective feature extraction, while multi-scale cross-modal fusion encounters semantic gap issue. To address these challenges, novel network based on called SDA-les (AGAFMs) was designed decoding stage fuse at various scales, which dynamically adjust importance global perspective better balance information. superior performance proposed method is demonstrated comprehensive evaluations ISPRS Potsdam dataset with 97.66% F1 score 95.42% IoU, Vaihingen 96.56% 93.35% DFC23 Track2 91.35% 84.08% IoU.

Language: Английский

Citations

0