New Methods for Waterfowl and Habitat Survey Using AI and Drone Imagery
Monitoring waterfowl populations is essential for informing habitat management, conservation strategies, and sustainable harvest regulations. Many target species such as mallards and northern pintails are keystone components of wetland ecosystems, serving as ecological indicators due to their sensit...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-06-01
|
| Series: | Drones |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2504-446X/9/7/451 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Monitoring waterfowl populations is essential for informing habitat management, conservation strategies, and sustainable harvest regulations. Many target species such as mallards and northern pintails are keystone components of wetland ecosystems, serving as ecological indicators due to their sensitivity to environmental changes. The integration of drone technology and artificial intelligence (AI) is significantly transforming the field of wildlife conservation and habitat monitoring. Existing methods for waterfowl monitoring face critical challenges such as low accuracy in identifying overlapping image regions and limited segmentation accuracy in complex habitats. To address these issues, this paper presents an end-to-end system and several new methods for efficiently and accurately identifying waterfowl populations in their natural habitats using AI and drone imagery. We applied advanced deep learning models to drone imagery for detecting and counting waterfowl. To handle overlapping regions in consecutive images, we developed a bird-location-based method that quickly and accurately identifies overlaps. For habitat segmentation, we proposed an effective approach combining Meta’s Segment Anything Model (SAM) with a ResNet50 classifier. Additionally, we used ChatGPT to generate clear, easy-to-read reports summarizing detection results. Experimental results show that our bird detection model (Faster R-CNN) achieved 86.57% mAP, our habitat segmentation method reached 85.1% accuracy (average F1 score: 81.8%), and our overlap detection method maintained an error rate below 5% with faster performance compared to traditional techniques. These outcomes highlight the practical effectiveness of our integrated pipeline for wildlife conservation and habitat monitoring. |
|---|---|
| ISSN: | 2504-446X |