The use of drones and Artificial Intelligence for dugong sighting detection in a limited resource scenario
The use of commercially available drones and artificial intelligence (AI) has grown in popularity in the last decade. Nonetheless, its usage to detect cryptic and high-mobility marine mammals remains constrained by resource-intensive nature, vast coverage areas, hardware limitations, and environment...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
EDP Sciences
2025-01-01
|
| Series: | BIO Web of Conferences |
| Online Access: | https://www.bio-conferences.org/articles/bioconf/pdf/2025/07/bioconf_icfaes24_01004.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The use of commercially available drones and artificial intelligence (AI) has grown in popularity in the last decade. Nonetheless, its usage to detect cryptic and high-mobility marine mammals remains constrained by resource-intensive nature, vast coverage areas, hardware limitations, and environmental variables. This study aims to recount our experience conducting a combination of drone and AI-assisted detection (WISDAM) in a scenario with limited resources to detect dugongs. The operation was conducted in September 2023, April 2024, and May 2024 in North Sulawesi, Indonesia. Prior to flight path design, CMS questionnaires and satellite data were utilized in order to comprehend the spatial and temporal context of the dugongs and their preferred habitat. A DJI Air 2s drone was used in 28 flights, covering 12.09 km2, yielding 8,509 photos. In total, 47 photos comprise dugongs, including seven with multiple individuals. 56 sightings were successfully identified manually by multiple analysts to minimize bias, and seven photos (12.5%) were considered dubious. AI detection is rather limited compared with manual detection’s numbers of positively identified dugong photos. Out of 153 AI detections, only 27 (17.6%) were True Positives. Therefore, more flights are needed to enhance the sample size for machine learning. |
|---|---|
| ISSN: | 2117-4458 |