Home // GEOProcessing 2022, The Fourteenth International Conference on Advanced Geographic Information Systems, Applications, and Services // View article

A Study of Zero-shot Learning for Visual Search on Satellite and Aerial Images

Authors:
A. Chuong Dang
Ion-George Todoran
Srushti Rashmi Shirish

Keywords: visual search; zero-shot learning; data indexing; deep learning

Abstract:
In this article, we present a visual search system based on latest Deep Learning techniques, which enables users to find images containing similar content as a query image. As many applications battle the dual challenge of limited labelled data that does not cover all the possible classes, we propose to mitigate this issue with an approach we call zero shot learning. We prove the potential of this approach by extensively experimenting on 3 of all time popular aerial imagery datasets. In addition, we show that pre-training the model on top-down imagery improves the final performance of the visual search system.

Pages: 13 to 18

Copyright: Copyright (c) IARIA, 2022

Publication date: June 26, 2022

Published in: conference

ISSN: 2308-393X

ISBN: 978-1-61208-983-6

Location: Porto, Portugal

Dates: from June 26, 2022 to June 30, 2022