finFindR

finFindR automates photographic identification of common bottlenose dolphins (Tursiops truncatus) by detecting and tracing dorsal fin markings using neural network algorithms to enable individual-level monitoring.

Key Features:

• R package implementation: Provided as an R package that performs automated photographic identification workflows.
• Automated cropping and tracing: Detects and crops dorsal fins from raw field photographs and traces the trailing edge of each fin.
• Neural network algorithms: Employs advanced neural network algorithms for fin detection and feature matching.
• Efficient matching system: Compares traced fin characteristics against an existing catalog and returns a ranked list of candidate identities, reporting the top 50 matches.
• Performance metrics: In test sets the correct identity was ranked first in 88% of cases (238/272 and 354/400), within the top 10 in 94% of cases, and within the top 50 in 97% of cases.
• Processing time reduction: Comparative tests reduced the average number of images reviewed per identification to about 10 versus 124 for manual methods.

Scientific Applications:

• Photographic mark–recapture: Supports individual-based mark–recapture studies for estimating population size of Tursiops truncatus using dorsal fin markings.
• Long-term monitoring: Enables large-scale and longitudinal monitoring of population dynamics and site fidelity in bottlenose dolphins.
• Conservation and management: Provides identification data to inform conservation, ecological research, and wildlife management decisions for bottlenose dolphins.

Methodology:

Implemented as an R package that uses neural network algorithms to detect and crop dorsal fins from raw photographs, trace the trailing edge of each fin, and compare traced fin characteristics to an existing catalog to generate a ranked list of up to 50 candidate matches.