Learning general representations for semantic segmentation and height estimation from remote sensing images

Wufan Zhao; C. Persello; Hu Ding; A. Stein

doi:10.1109/JURSE57346.2023.10144138

Learning general representations for semantic segmentation and height estimation from remote sensing images

Wufan Zhao^*, C. Persello, Hu Ding, A. Stein

^*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

38 Downloads (Pure)

Abstract

Deep-learning methods have been applied to remote-sensing imagery to achieve height estimation and semantic segmentation. Recent research has demonstrated that multi-task learning methods can act as a welcome addition to task-specific features by improving prediction accuracy across a range of tasks. How to effectively learn representations that achieve good performance for multiple tasks still remains challenging. We propose to adopt a unified network jointly learning multiple vision tasks' general representations by aligning them via small task-specific adapters. The experimental results on the Vaihingen dataset demonstrate that general representation learning can improve the performance of state-of-the-art methods in height estimation and semantic segmentation tasks.

Original language	English
Title of host publication	2023 Joint Urban Remote Sensing Event
Publisher	IEEE
Number of pages	4
ISBN (Electronic)	9781665493734
DOIs	https://doi.org/10.1109/JURSE57346.2023.10144138
Publication status	Published - 8 Jun 2023
Event	Joint Urban Remote Sensing Event, JURSE 2023 - Heraklion, Greece Duration: 17 May 2023 → 19 May 2023 http://jurse2023.org/

Publication series

Name	2023 Joint Urban Remote Sensing Event, JURSE 2023

Conference

Conference	Joint Urban Remote Sensing Event, JURSE 2023
Abbreviated title	JURSE 2023
Country/Territory	Greece
City	Heraklion
Period	17/05/23 → 19/05/23
Internet address	http://jurse2023.org/

Keywords

height estimation
knowledge distillation
Multi-task learning
remote sensing image
semantic segmentation
2023 OA procedure

Access to Document

10.1109/JURSE57346.2023.10144138

Learning_General_Representations_for_Semantic_Segmentation_and_Height_Estimation_from_Remote_Sensing_ImagesFinal published version, 5.18 MBLicence: Taverne

Cite this

@inproceedings{7c58cae0b77e4ad1a5988cbd1b685837,

title = "Learning general representations for semantic segmentation and height estimation from remote sensing images",

abstract = "Deep-learning methods have been applied to remote-sensing imagery to achieve height estimation and semantic segmentation. Recent research has demonstrated that multi-task learning methods can act as a welcome addition to task-specific features by improving prediction accuracy across a range of tasks. How to effectively learn representations that achieve good performance for multiple tasks still remains challenging. We propose to adopt a unified network jointly learning multiple vision tasks' general representations by aligning them via small task-specific adapters. The experimental results on the Vaihingen dataset demonstrate that general representation learning can improve the performance of state-of-the-art methods in height estimation and semantic segmentation tasks.",

keywords = "height estimation, knowledge distillation, Multi-task learning, remote sensing image, semantic segmentation, 2023 OA procedure",

author = "Wufan Zhao and C. Persello and Hu Ding and A. Stein",

note = "Publisher Copyright: {\textcopyright} 2023 IEEE.; Joint Urban Remote Sensing Event, JURSE 2023, JURSE 2023 ; Conference date: 17-05-2023 Through 19-05-2023",

year = "2023",

month = jun,

day = "8",

doi = "10.1109/JURSE57346.2023.10144138",

language = "English",

series = "2023 Joint Urban Remote Sensing Event, JURSE 2023",

publisher = "IEEE",

booktitle = "2023 Joint Urban Remote Sensing Event",

address = "United States",

url = "http://jurse2023.org/",

}

Zhao, W, Persello, C, Ding, H & Stein, A 2023, Learning general representations for semantic segmentation and height estimation from remote sensing images. in 2023 Joint Urban Remote Sensing Event. 2023 Joint Urban Remote Sensing Event, JURSE 2023, IEEE, Joint Urban Remote Sensing Event, JURSE 2023, Heraklion, Greece, 17/05/23. https://doi.org/10.1109/JURSE57346.2023.10144138

Learning general representations for semantic segmentation and height estimation from remote sensing images. / Zhao, Wufan; Persello, C.; Ding, Hu et al.
2023 Joint Urban Remote Sensing Event. IEEE, 2023. (2023 Joint Urban Remote Sensing Event, JURSE 2023).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › Academic › peer-review

TY - GEN

T1 - Learning general representations for semantic segmentation and height estimation from remote sensing images

AU - Zhao, Wufan

AU - Persello, C.

AU - Ding, Hu

AU - Stein, A.

PY - 2023/6/8

Y1 - 2023/6/8

N2 - Deep-learning methods have been applied to remote-sensing imagery to achieve height estimation and semantic segmentation. Recent research has demonstrated that multi-task learning methods can act as a welcome addition to task-specific features by improving prediction accuracy across a range of tasks. How to effectively learn representations that achieve good performance for multiple tasks still remains challenging. We propose to adopt a unified network jointly learning multiple vision tasks' general representations by aligning them via small task-specific adapters. The experimental results on the Vaihingen dataset demonstrate that general representation learning can improve the performance of state-of-the-art methods in height estimation and semantic segmentation tasks.

AB - Deep-learning methods have been applied to remote-sensing imagery to achieve height estimation and semantic segmentation. Recent research has demonstrated that multi-task learning methods can act as a welcome addition to task-specific features by improving prediction accuracy across a range of tasks. How to effectively learn representations that achieve good performance for multiple tasks still remains challenging. We propose to adopt a unified network jointly learning multiple vision tasks' general representations by aligning them via small task-specific adapters. The experimental results on the Vaihingen dataset demonstrate that general representation learning can improve the performance of state-of-the-art methods in height estimation and semantic segmentation tasks.

KW - height estimation

KW - knowledge distillation

KW - Multi-task learning

KW - remote sensing image

KW - semantic segmentation

KW - 2023 OA procedure

U2 - 10.1109/JURSE57346.2023.10144138

DO - 10.1109/JURSE57346.2023.10144138

M3 - Conference contribution

AN - SCOPUS:85163770962

T3 - 2023 Joint Urban Remote Sensing Event, JURSE 2023

BT - 2023 Joint Urban Remote Sensing Event

PB - IEEE

T2 - Joint Urban Remote Sensing Event, JURSE 2023

Y2 - 17 May 2023 through 19 May 2023

ER -

Learning general representations for semantic segmentation and height estimation from remote sensing images

Abstract

Publication series

Conference

Keywords

Access to Document

Fingerprint

Cite this