PSGRNet: Multi-scale spectral–frequency representation learning with gated residual fusion for coastal wetland classification using UAV hyperspectral imagery

Coastal wetlands are characterized by diverse vegetation, high spectral similarity, fragmented spatial distributions, and dynamic environmental conditions, posing significant challenges for accurate hyperspectral image classification. Although UAV hyperspectral imagery provides fine-grained spatial and spectral observations that can help capture such complex and fragmented features, effectively leveraging this information to distinguish spectrally similar and spatially intermixed vegetation types remains an open problem. Therefore, we propose a Progressive Spectral Gated Residual Network (PSGRNet). Specifically, a spectral enhancement and multi-scale sequence modeling strategy is first introduced to strengthen discriminative spectral representations and capture multi-scale spectral dependencies. Building on these enhanced representations, a frequency-aware module further exploits global frequency-domain characteristics to aggregate informative spectral features adaptively. To effectively integrate both spectral and frequency-domain representations, a GRF strategy is developed to improve the robustness of the network in heterogeneous and fragmented coastal wetland scenes. We have collected three UAV-based hyperspectral datasets covering representative coastal wetland scenarios, including reed-dominated natural wetlands (NT-YK), highly fragmented landscapes with dynamic land-cover transitions (YC-YS), and human-influenced estuarine environments (LYG-LH). Experimental results on these three datasets demonstrate that PSGRNet outperforms 9 representative methods, achieving macro F1 scores of 89.25%, 81.99%, and 81.29%, respectively. These results verify the effectiveness and robustness of the proposed method for accurate coastal wetland classification and demonstrate the value of the released datasets. The datasets will be publicly available at https://github.com/HAPPYFST/Wetland-HSI-Data .

Authors

Institutions

Publication Details

Journal
International Journal of Applied Earth Observation and Geoinformation
Published
2026-11-16
DOI
https://doi.org/10.1016/j.jag.2026.105674
Primary Topic
Remote-Sensing Image Classification
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

PSGRNet: Multi-scale spectral–frequency representation learning with gated residual fusion for coastal wetland classification using UAV hyperspectral imagery

Hongjun Su, Qian Du, Jun Zhao, Hui Yang et al.
5 links
International Journal of Applied Earth Observation and Geoinformation
Remote-Sensing Image Classification
article

PSGRNet: Multi-scale spectral–frequency representation learning with gated residual fusion for coastal wetland classification using UAV hyperspectral imagery

Hongjun Su, Qian Du, Jun Zhao, Hui Yang, Yang Tao
article en

Abstract

Coastal wetlands are characterized by diverse vegetation, high spectral similarity, fragmented spatial distributions, and dynamic environmental conditions, posing significant challenges for accurate hyperspectral image classification. Although UAV hyperspectral imagery provides fine-grained spatial and spectral observations that can help capture such complex and fragmented features, effectively leveraging this information to distinguish spectrally similar and spatially intermixed vegetation types remains an open problem. Therefore, we propose a Progressive Spectral Gated Residual Network (PSGRNet). Specifically, a spectral enhancement and multi-scale sequence modeling strategy is first introduced to strengthen discriminative spectral representations and capture multi-scale spectral dependencies. Building on these enhanced representations, a frequency-aware module further exploits global frequency-domain characteristics to aggregate informative spectral features adaptively. To effectively integrate both spectral and frequency-domain representations, a GRF strategy is developed to improve the robustness of the network in heterogeneous and fragmented coastal wetland scenes. We have collected three UAV-based hyperspectral datasets covering representative coastal wetland scenarios, including reed-dominated natural wetlands (NT-YK), highly fragmented landscapes with dynamic land-cover transitions (YC-YS), and human-influenced estuarine environments (LYG-LH). Experimental results on these three datasets demonstrate that PSGRNet outperforms 9 representative methods, achieving macro F1 scores of 89.25%, 81.99%, and 81.29%, respectively. These results verify the effectiveness and robustness of the proposed method for accurate coastal wetland classification and demonstrate the value of the released datasets. The datasets will be publicly available at https://github.com/HAPPYFST/Wetland-HSI-Data .

International Journal of Applied Earth Observation and GeoinformationVol. 155
Hohai University (CN), China University of Mining and Technology (CN), Mississippi State University (US)
National Natural Science Foundation of China
Remote-Sensing Image Classification
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon