Drones, automatic counting tools, and artificial neural networks in wildlife population censusing

Area of Science:

• Ornithology
• Wildlife Biology
• Remote Sensing

Background:

• Accurate waterbird population counts are crucial for conservation and ecological monitoring.
• Traditional survey methods can be labor-intensive and may miss remote or inaccessible populations.
• Unmanned Aerial Vehicles (UAVs) offer a potential solution for efficient wildlife surveys.

Purpose of the Study:

• To investigate the efficacy of drone technology for counting waterbird flock sizes across 33 species.
• To compare automated counting methods, including ImageJ/Fiji software and machine learning algorithms.
• To assess potential adverse reactions of waterbirds to drone presence during breeding and non-breeding seasons.

Main Methods:

• Drone surveys were conducted to count flock sizes of 33 waterbird species during breeding and non-breeding periods.
• Image analysis was performed using ImageJ/Fiji software for automated bird counting.
• Machine learning algorithms (neural networks) were employed for automated bird counting and compared with ImageJ/Fiji.

Main Results:

• Drone counting achieved success in 96% of 343 surveyed cases.
• Adverse reactions were observed: 18.8% of non-breeding birds were flushed, and 3.6% of breeding birds exhibited aggressive behavior towards the drone.
• Machine learning provided faster counting (7s per 100 birds) than ImageJ/Fiji (64s per 100 birds), though initial setup was time-consuming.

Conclusions:

• Drones are a highly successful tool for waterbird population surveys, demonstrating high accuracy and efficiency.
• Machine learning is recommended for large datasets and assemblages due to its speed, despite initial preparation time.
• Responsible drone use in wildlife research requires experienced personnel to mitigate disturbance and ensure ethical practices.