Visual Drone Terrain Classification: A Manual Classification Approach

Joon Du Randt, Kevin Meehan

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

Abstract

This research investigates a method for performing manual classification of the terrain in imagery from the on-board camera of an unmanned aerial vehicle, to develop classifiers for systematic terrain classification. Drone images were captured across rural County Donegal in Ireland, and software was developed to manually label the terrain in these images, labelled in a lattice of 30 x30-pixel tiles. This dataset was used to train both a classic computer vision model and a Convolutional Neural Net model to classify the type of terrain under the UAV. The accuracy of the computer vision approach to the classification was compared to that of a Convolutional Neural Network trained using the Semantic Segmentation approach. The Convolutional Neural Network classifier was found to be the most accurate approach, achieving an fl score of 0.95.

Original languageEnglish
Title of host publication2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies, 3ICT 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781728196732
DOIs
Publication statusPublished - 20 Dec 2020
Event2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies, 3ICT 2020 - Sakheer, Bahrain
Duration: 20 Dec 202021 Dec 2020

Publication series

Name2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies, 3ICT 2020

Conference

Conference2020 International Conference on Innovation and Intelligence for Informatics, Computing and Technologies, 3ICT 2020
Country/TerritoryBahrain
CitySakheer
Period20/12/2021/12/20

Keywords

  • Computer Vision
  • Drone
  • Manual Classification
  • Terrain
  • Training Dataset

Fingerprint

Dive into the research topics of 'Visual Drone Terrain Classification: A Manual Classification Approach'. Together they form a unique fingerprint.

Cite this