TS2GNet: A temporal–spatial–spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imagery

Satellite hyperspectral imagery (SHSI) provides detailed biochemical and physical characteristics of forest canopies, making it crucial for mapping large-scale tree species. However, two main challenges limit its effectiveness: (1) the spatial resolution of SHSI restricts its applicability in subtro...

Full description

Saved in:
Bibliographic Details
Main Authors: Kaijian Xu, Henghui Han, Shuzhou Wang, Ping Zhao, Jun Geng, Hailan Jiang, Anxin Ding
Format: Article
Language:English
Published: Elsevier 2025-08-01
Series:International Journal of Applied Earth Observations and Geoinformation
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S1569843225003620
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850035117409435648
author Kaijian Xu
Henghui Han
Shuzhou Wang
Ping Zhao
Jun Geng
Hailan Jiang
Anxin Ding
author_facet Kaijian Xu
Henghui Han
Shuzhou Wang
Ping Zhao
Jun Geng
Hailan Jiang
Anxin Ding
author_sort Kaijian Xu
collection DOAJ
description Satellite hyperspectral imagery (SHSI) provides detailed biochemical and physical characteristics of forest canopies, making it crucial for mapping large-scale tree species. However, two main challenges limit its effectiveness: (1) the spatial resolution of SHSI restricts its applicability in subtropical regions, where tree species diversity is high, and (2) the complexity of spectral data and structural variability complicates seasonal feature extraction. Most existing methods are focused on drone data and struggle to effectively integrate the multiscale temporal, spatial, and spectral information inherent in satellite-based hyperspectral data. To address these challenges, we enhance ZY1-02E hyperspectral data via a superresolution technique that incorporates Sentinel-2 data. We then propose TS2GNet, a novel hyperspectral classification network that integrates temporal-, spatial-, and spectral-domain guidance. TS2GNet employs a dual-stream architecture that focus on dynamic interactions between spatial and spectral domains, while also incorporating temporal modeling and SHSI feature-domain guidance. We evaluate our method on eight dominant tree species in the Ta-pieh Mountains and compare its performance with six state-of-the-art (SOTA) deep learning-based hyperspectral classification methods. The results demonstrate that (1) the enhanced ZY1-02E data lead to 1.93 %-to-3.47 % seasonal overall accuracy (OA) improvements; (2) TS2GNet consistently outperforms the other methods, achieving OA improvements ranging from 6.02 % to 6.36 % across different seasons; and (3) by incorporating multiseasonal SHSI features, TS2GNet increases OA from 86.81 ± 0.46 % to 90.72 ± 0.60 %, surpassing the other models by 2.77 % to 6.29 %. Additionally, TS2GNet demonstrates strong generalizability when applied to Sentinel-2 data. These findings advance satellite-based hyperspectral remote sensing methodologies and offer new insights into tree species distribution mapping in complex forest ecosystems.The codes associated with this paper are publicly available at https://github.com/FRS-HFUT/TS2GNet.
format Article
id doaj-art-e2ab89a0b8df488e915beadb3bee17ba
institution DOAJ
issn 1569-8432
language English
publishDate 2025-08-01
publisher Elsevier
record_format Article
series International Journal of Applied Earth Observations and Geoinformation
spelling doaj-art-e2ab89a0b8df488e915beadb3bee17ba2025-08-20T02:57:35ZengElsevierInternational Journal of Applied Earth Observations and Geoinformation1569-84322025-08-0114210471510.1016/j.jag.2025.104715TS2GNet: A temporal–spatial–spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imageryKaijian Xu0Henghui Han1Shuzhou Wang2Ping Zhao3Jun Geng4Hailan Jiang5Anxin Ding6School of Resources and Environmental Engineering, Hefei University of Technology, Hefei 230009, China; Institute for Spatial Information Intelligence Analysis and Application, Hefei University of Technology, Hefei 230009, China; Corresponding author.School of Resources and Environmental Engineering, Hefei University of Technology, Hefei 230009, ChinaSchool of Resources and Environmental Engineering, Hefei University of Technology, Hefei 230009, ChinaSchool of Resources and Environmental Engineering, Hefei University of Technology, Hefei 230009, China; Institute for Spatial Information Intelligence Analysis and Application, Hefei University of Technology, Hefei 230009, ChinaSchool of Civil Engineering, Hefei University of Technology, Hefei 230009, ChinaSchool of Resources and Environmental Engineering, Hefei University of Technology, Hefei 230009, ChinaSchool of Resources and Environmental Engineering, Hefei University of Technology, Hefei 230009, ChinaSatellite hyperspectral imagery (SHSI) provides detailed biochemical and physical characteristics of forest canopies, making it crucial for mapping large-scale tree species. However, two main challenges limit its effectiveness: (1) the spatial resolution of SHSI restricts its applicability in subtropical regions, where tree species diversity is high, and (2) the complexity of spectral data and structural variability complicates seasonal feature extraction. Most existing methods are focused on drone data and struggle to effectively integrate the multiscale temporal, spatial, and spectral information inherent in satellite-based hyperspectral data. To address these challenges, we enhance ZY1-02E hyperspectral data via a superresolution technique that incorporates Sentinel-2 data. We then propose TS2GNet, a novel hyperspectral classification network that integrates temporal-, spatial-, and spectral-domain guidance. TS2GNet employs a dual-stream architecture that focus on dynamic interactions between spatial and spectral domains, while also incorporating temporal modeling and SHSI feature-domain guidance. We evaluate our method on eight dominant tree species in the Ta-pieh Mountains and compare its performance with six state-of-the-art (SOTA) deep learning-based hyperspectral classification methods. The results demonstrate that (1) the enhanced ZY1-02E data lead to 1.93 %-to-3.47 % seasonal overall accuracy (OA) improvements; (2) TS2GNet consistently outperforms the other methods, achieving OA improvements ranging from 6.02 % to 6.36 % across different seasons; and (3) by incorporating multiseasonal SHSI features, TS2GNet increases OA from 86.81 ± 0.46 % to 90.72 ± 0.60 %, surpassing the other models by 2.77 % to 6.29 %. Additionally, TS2GNet demonstrates strong generalizability when applied to Sentinel-2 data. These findings advance satellite-based hyperspectral remote sensing methodologies and offer new insights into tree species distribution mapping in complex forest ecosystems.The codes associated with this paper are publicly available at https://github.com/FRS-HFUT/TS2GNet.http://www.sciencedirect.com/science/article/pii/S1569843225003620Tree species mappingSatellite hyperspectral imagerySuperresolutionSentinel-2Transformer
spellingShingle Kaijian Xu
Henghui Han
Shuzhou Wang
Ping Zhao
Jun Geng
Hailan Jiang
Anxin Ding
TS2GNet: A temporal–spatial–spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imagery
International Journal of Applied Earth Observations and Geoinformation
Tree species mapping
Satellite hyperspectral imagery
Superresolution
Sentinel-2
Transformer
title TS2GNet: A temporal–spatial–spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imagery
title_full TS2GNet: A temporal–spatial–spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imagery
title_fullStr TS2GNet: A temporal–spatial–spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imagery
title_full_unstemmed TS2GNet: A temporal–spatial–spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imagery
title_short TS2GNet: A temporal–spatial–spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imagery
title_sort ts2gnet a temporal spatial spectral multidomain guided network for classifying hyperspectral tree species using multiseason satellite imagery
topic Tree species mapping
Satellite hyperspectral imagery
Superresolution
Sentinel-2
Transformer
url http://www.sciencedirect.com/science/article/pii/S1569843225003620
work_keys_str_mv AT kaijianxu ts2gnetatemporalspatialspectralmultidomainguidednetworkforclassifyinghyperspectraltreespeciesusingmultiseasonsatelliteimagery
AT henghuihan ts2gnetatemporalspatialspectralmultidomainguidednetworkforclassifyinghyperspectraltreespeciesusingmultiseasonsatelliteimagery
AT shuzhouwang ts2gnetatemporalspatialspectralmultidomainguidednetworkforclassifyinghyperspectraltreespeciesusingmultiseasonsatelliteimagery
AT pingzhao ts2gnetatemporalspatialspectralmultidomainguidednetworkforclassifyinghyperspectraltreespeciesusingmultiseasonsatelliteimagery
AT jungeng ts2gnetatemporalspatialspectralmultidomainguidednetworkforclassifyinghyperspectraltreespeciesusingmultiseasonsatelliteimagery
AT hailanjiang ts2gnetatemporalspatialspectralmultidomainguidednetworkforclassifyinghyperspectraltreespeciesusingmultiseasonsatelliteimagery
AT anxinding ts2gnetatemporalspatialspectralmultidomainguidednetworkforclassifyinghyperspectraltreespeciesusingmultiseasonsatelliteimagery