New Approaches and Tools for Ship Detection in Optical Satellite Imagery

Aaron Walter Avila Cordova, William Condori Quispe, Remy Jorge Cuba Inca, Wilder Nina Choquehuayta, Eveling Castro Gutierrez

Research output: Contribution to journalConference articlepeer-review

Abstract

Ship detection using optical satellite images is a very important task for the field of maritime security, either in search of lost ships or in maritime control of a commercial or military type. Added to this are the advances in the field of Computer Vision, especially in the use of models based on Artificial Intelligence, which allow the construction of robust and more precise detection systems. However, geographic scenarios, typical of a satellite image, limit the development of this type of system since they require the availability of a large number of images in different scenarios. In this paper, a new approach to Ship Detection is proposed using two new data sets labeled with horizontal bounding boxes (HBB). Likewise, a new labeling tool (DATATOOL) is presented that allows better organization and distribution of data. The new data sets, Peruvian Ship Dataset (PSDS) and Mini Ship Dataset (MSDS), have been generated from optical satellite images obtained from different sources. PSDS is created from 22 satellite images of PERUSAT-1 with 0.7m spatial resolution, giving a total of 1310 images. MSDS has been generated using Google Earth satellite images, generating 2993 images of 900x900 pixels. Ships are found both at sea or inshore. Finally, results of the tests using Deep Learning Algorithms such as YOLT and YOLOv4 are presented, following the approach and the proposed tools. Resource and source code available at https://gitlab.com/williamccondori/datatool

Original languageEnglish
Article number012003
JournalJournal of Physics: Conference Series
Volume1642
Issue number1
DOIs
StatePublished - 25 Sep 2020
Event4th International Conference on Artificial Intelligence Applications and Technologies, AIAAT 2020 - Virtual, Online, China
Duration: 8 Aug 2020 → …

Bibliographical note

Publisher Copyright:
© Published under licence by IOP Publishing Ltd.

Fingerprint

Dive into the research topics of 'New Approaches and Tools for Ship Detection in Optical Satellite Imagery'. Together they form a unique fingerprint.

Cite this