e-TransUNet: TransUNet provides a strong spatial transformation for precise deforestation mapping

Jamali, A; Roy, SK; Pradhan, B

e-TransUNet: TransUNet provides a strong spatial transformation for precise deforestation mapping

Jamali, A Roy, SK Pradhan, B

Permalink

Publisher:: Elsevier
Publication Type:: Journal Article
Citation:: Remote Sensing Applications: Society and Environment, 2024, 35, pp. 101221
Issue Date:: 2024-08-01

In Progress

	Filename	Description	Size
	1-s2.0-S2352938524000855-main.pdf	Published version	6.12 MB	Adobe PDF	View/Open

Copyright Clearance Process

Recently Added
In Progress
Open Access

This item is being processed and is not currently available.

Full metadata record

Field	Value	Language
dc.contributor.author	Jamali, A
dc.contributor.author	Roy, SK
dc.contributor.author	Pradhan, B https://orcid.org/0000-0001-9863-2054
dc.date.accessioned	2024-09-24T05:20:09Z
dc.date.available	2024-09-24T05:20:09Z
dc.date.issued	2024-08-01
dc.identifier.citation	Remote Sensing Applications: Society and Environment, 2024, 35, pp. 101221
dc.identifier.issn	2352-9385
dc.identifier.issn	2352-9385
dc.identifier.uri	http://hdl.handle.net/10453/180950
dc.description.abstract	Forest loss significantly contributes to climate change. As a result, scientists aim to understand its causes and evaluate forest loss classification maps. While Vision Transformers (ViTs) have demonstrated superior performance compared to convolutional neural networks (CNNs) in computer vision applications, they face challenges in recognizing and analyzing remote sensing images. These challenges include computational complexity, which increases processing costs, and the requirement for more labeled input data compared to CNNs. To address these issues, this paper introduces enhanced transformer UNet (e-TransUNet), which integrates spatial transformation architecture to enhance feature description in the skip connection path of the attention TransUNet for precise deforestation mapping. The model is validated in two prominent South American forest biomes—the Amazon and Atlantic forests. Experimental results highlight the significant improvement achieved by e-TransUNet, particularly in the Atlantic rainforests, where it boosts recall, overall accuracy, F1-score, dice coefficient, and precision by approximately 2, 2, 3, 4, and 4 percentage points, respectively, compared to the base TransUNet segmentation architecture. The code will be made publicly available at https://github.com/aj1365/e-TransUNet.
dc.language	en
dc.publisher	Elsevier
dc.relation.ispartof	Remote Sensing Applications: Society and Environment
dc.relation.isbasedon	10.1016/j.rsase.2024.101221
dc.rights	info:eu-repo/semantics/restrictedAccess
dc.subject	0406 Physical Geography and Environmental Geoscience
dc.subject.classification	3709 Physical geography and environmental geoscience
dc.subject.classification	4406 Human geography
dc.title	e-TransUNet: TransUNet provides a strong spatial transformation for precise deforestation mapping
dc.type	Journal Article
utslib.citation.volume	35
utslib.for	0406 Physical Geography and Environmental Geoscience
pubs.organisational-group	University of Technology Sydney
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology
pubs.organisational-group	University of Technology Sydney/Faculty of Engineering and Information Technology/School of Civil and Environmental Engineering
pubs.organisational-group	University of Technology Sydney/Strength - CAMGIS - Centre for Advanced Modelling and Geospatial lnformation Systems
pubs.organisational-group	University of Technology Sydney/UTS Groups
pubs.organisational-group	University of Technology Sydney/UTS Groups/Transport Research Centre (TRC)
pubs.organisational-group	University of Technology Sydney/UTS Groups/Centre for Advanced Modelling and Geospatial lnformation Systems (CAMGIS)
pubs.organisational-group	University of Technology Sydney/UTS Groups/Transport Research Centre (TRC)/Associate Member
utslib.copyright.status	in_progress	*
dc.date.updated	2024-09-24T05:20:03Z
pubs.publication-status	Published
pubs.volume	35

Abstract:

Forest loss significantly contributes to climate change. As a result, scientists aim to understand its causes and evaluate forest loss classification maps. While Vision Transformers (ViTs) have demonstrated superior performance compared to convolutional neural networks (CNNs) in computer vision applications, they face challenges in recognizing and analyzing remote sensing images. These challenges include computational complexity, which increases processing costs, and the requirement for more labeled input data compared to CNNs. To address these issues, this paper introduces enhanced transformer UNet (e-TransUNet), which integrates spatial transformation architecture to enhance feature description in the skip connection path of the attention TransUNet for precise deforestation mapping. The model is validated in two prominent South American forest biomes—the Amazon and Atlantic forests. Experimental results highlight the significant improvement achieved by e-TransUNet, particularly in the Atlantic rainforests, where it boosts recall, overall accuracy, F1-score, dice coefficient, and precision by approximately 2, 2, 3, 4, and 4 percentage points, respectively, compared to the base TransUNet segmentation architecture. The code will be made publicly available at https://github.com/aj1365/e-TransUNet.

Please use this identifier to cite or link to this item:

http://hdl.handle.net/10453/180950