Short Talk (7,5 mins) - Edits Required Australian Marine Sciences Association 2022

Drone Flight to FAIR Dataset (#124)

Milica Stankovic 1 , Nick Mortimer 2 , Mat Vanderklift 3
  1. Excellence Center for Biodiversity of Peninsular Thailand, Prince of Songkla University, Hat Yai, Songkhla, Thailand
  2. CSIRO Oceans and Atmosphere, Indian Ocean Marine Research Centre, Crawley, Western Australia, Australia
  3. IORA Blue Carbon Hub, CSIRO Oceans and Atmosphere, Indian Ocean Marine Research Centre, Crawley, Western Australia, Australia

Drones have become an important tool in monitoring coastal and marine ecosystems, for researchers, managers and local communities. Currently there is a lack of standardization and automation in the workflows. Building workflows based on open-source tools to deliver standard products that are ready for use in machine learning can enable researchers and communities to deliver FAIR datasets. We present a workflow that has been developed using open-source Python tools that accepts images from a range of drones. The workflow applies corrections to the image position and stores them with meaning full names in a regular directory structure while providing quality data management metrics and feedback. For habitat assessment the system builds a series of machine learning ready tiles using direct geo referencing. This approach is fast, efficient and allows for multiple looks at the same area. For object identification we have adapted a popular opensource tool labelImg to allow for efficient human classification of marine objects such as turtles and sharks. The workflow then produces a training set ready to be shared for use in machine learning.

 

  1. Lin (2015) LabelImg (v1.8.1). https://github.com/tzutalin/labelImg