SEALNET: Facial recognition software for ecological studies of harbor seals.

Area of Science:

• Marine Biology
• Conservation Technology
• Computer Vision

Background:

• Long-term monitoring of coastal species like harbor seals is challenging due to high costs, time investment, and invasive methods.
• There is a critical need for advanced, non-invasive techniques for collecting and analyzing data on marine mammal populations.

Purpose of the Study:

• To introduce and evaluate SealNet, an automated facial recognition software designed for identifying individual harbor seals.
• To demonstrate the utility of SealNet in ecological and population studies, offering an improved alternative to traditional monitoring methods.

Main Methods:

• Developed SealNet, incorporating a graphical user interface (GUI) for seal face detection, alignment, and feature extraction.
• Utilized a deep convolutional neural network (CNN) for individual seal classification, optimized for small datasets.
• Piloted SealNet using 1752 images of 408 harbor seals from Casco Bay, Maine, collected over two years (2019-2020).

Main Results:

• SealNet achieved 88% Rank-1 and 96% Rank-5 accuracy in closed-set identification of individual harbor seals.
• The SealNet software demonstrated superior performance compared to a re-trained primate facial recognition system (PrimNet).
• The system efficiently processed a large volume of image data, facilitating ecological and behavioral research.

Conclusions:

• SealNet provides a valuable, automated tool for the non-invasive identification of individual seals, significantly enhancing marine mammal research.
• This technology streamlines data collection and analysis, supporting conservation efforts and behavioral studies in marine ecosystems.
• The development of SealNet represents a significant advancement in conservation technology for wildlife monitoring.