Detection, Localization and Classification of Fish and Fish Species in Poor Conditions using Convolutional Neural Networks

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Abstract

In this work the initial steps towards a system capable of parametrising fish schools in underwater images are presented. For this purpose a deep convolutional neural network called Optical Fish Detection Network (OFDNet) is introduced. This is based on state-of-the-art deep learning object detection architectures and carries out the task of fish detection, localization and species classification using visual data obtained by underwater cameras. This work is focused towards applications in the poorly conditioned North and Baltic Sea and is initially developed for the purpose of recognizing herring and mackerel. Based on experiments on a dataset obtained at sea, OFDNet is shown to successfully detect 66.7% of the fish included and furthermore classify 89.7% of these correctly.
Original languageEnglish
Title of host publicationProceedings of 2018 IEEE/OES Autonomous Underwater Vehicle Workshop (AUV)
Number of pages6
PublisherIEEE
Publication date2019
Pages1-6
ISBN (Electronic)978-1-7281-0253-5
DOIs
Publication statusPublished - 2019
Event2018 IEEE OES Autonomous Underwater Vehicle Symposium - Rectory Building, University of Porto, Porto, Portugal
Duration: 6 Nov 20199 Nov 2019
Conference number: 13
https://auv2018.lsts.pt/

Workshop

Workshop2018 IEEE OES Autonomous Underwater Vehicle Symposium
Number13
LocationRectory Building, University of Porto
CountryPortugal
CityPorto
Period06/11/201909/11/2019
Internet address
Series2018 Ieee/oes Autonomous Underwater Vehicle Workshop (auv)
ISSN2377-6536

Keywords

  • Artificial intelligence
  • Deep learning
  • Convolutional neural networks
  • Object detection
  • Fish detection

Fingerprint Dive into the research topics of 'Detection, Localization and Classification of Fish and Fish Species in Poor Conditions using Convolutional Neural Networks'. Together they form a unique fingerprint.

Cite this