Learning to Understand Earth Observation Images with Weak and Unreliable Ground Truth - Archive ouverte HAL Access content directly
Conference Papers Year :

Learning to Understand Earth Observation Images with Weak and Unreliable Ground Truth

Apprendre à comprendre les images d'observation de la Terre avec des annotations pauvres et non fiables

Abstract

In this paper we discuss the issues of using inexact and inaccurate ground truth in the context of supervised learning. To leverage large amounts of Earth observation data for training algorithms, one often has to use ground truth which was not been carefully assessed. We address both the problems of training and evaluation. We first propose a weakly supervised approach for training change classifiers which is able to detect pixel-level changes in aerial images. We then propose a data poisoning approach to get a reliable estimate of the accuracy that can be expected from a classifier, even when the only ground-truth available does not match the reality. Both are assessed on practical land use and land cover applications.
Fichier principal
Vignette du fichier
DTIS19090.1571822827_preprint.pdf (8.2 Mo) Télécharger le fichier
Origin : Files produced by the author(s)
Loading...

Dates and versions

hal-02355360 , version 1 (08-11-2019)

Identifiers

  • HAL Id : hal-02355360 , version 1

Cite

Rodrigo Caye Daudt, Adrien Chan-Hon-Tong, Bertrand Le Saux, Alexandre Boulch. Learning to Understand Earth Observation Images with Weak and Unreliable Ground Truth. IGARSS 2019, Jul 2019, YOKOHAMA, Japan. ⟨hal-02355360⟩
42 View
66 Download

Share

Gmail Facebook Twitter LinkedIn More