Enhancing training set through multi-temporal attention analysis in transformers for multi-year land cover mapping

Rocco Sedona, Jan Ebert, C. Paris, Morris Riedel, Gabriele Cavallaro

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

Abstract

The continuous stream of high spatial resolution satellite data offers the opportunity to regularly produce land cover (LC) maps. To this end, Transformer deep learning (DL) models have recently proven their effectiveness in accurately classifying long time series (TS) of satellite images. The continual generation of regularly updated LC maps can be used to analyze dynamic phenomena and extract multi-temporal information. However, several challenges need to be addressed. Our paper aims to study how the performance of a Transformer model changes when classifying TS of satellite images acquired in years later than those in the training set. In particular, the behavior of the attention in the Transformer model is analyzed to determine when the information provided by the initial training set needs to be updated to keep generating accurate LC products. Preliminary results show that: (i) the selection of the positional encoding strategy used in the Transformer has a significant impact on the classification accuracy obtained with multi-year TS, and (ii) the most affected classes are the seasonal ones.
Original languageEnglish
Title of host publicationIGARSS 2023
Subtitle of host publication 2023 IEEE International Geoscience and Remote Sensing Symposium
PublisherIEEE
Pages5411-5414
Number of pages4
ISBN (Electronic)9798350320107
ISBN (Print)979-8-3503-3174-5
DOIs
Publication statusPublished - 20 Oct 2023
Event43rd IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2023 - Pasadena Convention Center, Pasadena, United States
Duration: 16 Jul 202321 Jul 2023
Conference number: 43
https://2023.ieeeigarss.org/index.php

Conference

Conference43rd IEEE International Geoscience and Remote Sensing Symposium, IGARSS 2023
Abbreviated titleIGARSS 2023
Country/TerritoryUnited States
CityPasadena
Period16/07/2321/07/23
Internet address

Keywords

  • Training
  • Analytical models
  • Satellites
  • Time series analysis
  • Streaming media
  • Transformers
  • Satellite images
  • 2023 OA procedure

Fingerprint

Dive into the research topics of 'Enhancing training set through multi-temporal attention analysis in transformers for multi-year land cover mapping'. Together they form a unique fingerprint.

Cite this