An AI environment now includes marine life, with the introduction of an innovative facial recognition tool, SealNet, offering a non-invasive approach to monitoring seal populations.
- Students from Colgate University have developed a seal face-finding app.
- The app, SealNet, is tailored to identify the harbour seal, the most widely dispersed seal.
- This is a non-invasive tool to study seals and could be used to allow citizen scientists to contribute to logging seal faces.
Smithsonian Magazine reports on the development of a pioneering facial recognition tool for seals, a noteworthy step forward in artificial intelligence’s (AI) application to environmental research and conservation. The AI environment was created by a team of undergraduate students from Colgate University in New York. The system, known as SealNet, uses a convolutional neural network trained with thousands of seal photos and it serves as a powerful tool for identifying individual harbor seals.
The idea for SealNet originated from previous facial recognition systems for primates and bears. Biologist Krista Ingram led the student team in creating this AI system tailored to harbor seals, an intriguing species known for congregating in haul outs along the coast.
Creating the system required substantial training to recognize seal faces from a given photo and adjust it to a standard size. Notably, the team manually identified facial features such as the nose, mouth, and the center of the eyes to enhance the precision of the AI system. The process was lengthy and meticulous but essential in the development of an effective tool.
For this ambitious project, the Colgate University team collected more than 2,000 seal photos from around Casco Bay, Maine, over two years. The software was then put to the test with images of 406 distinct seals, with impressive results: SealNet correctly identified the seals’ faces 85 percent of the time. It was then further refined, and the database was expanded to include about 1,500 seal faces. As the database grows, so does the system’s identification accuracy.
However, like all technology, SealNet is not without its flaws. In a variation of the old saying “if all you have is a hammer, everything looks like a nail”, sometimes SealNet mistakenly identified other objects, like rocks or vegetation, as seal faces. In such cases, the team believes it’s best to manually check the results to ensure they do indeed belong to a real seal.
Understanding the necessity of such technology brings us to the crux of this innovation. The AI environment is increasingly extending into wildlife and environmental research, and SealNet provides a valuable, non-invasive resource for researchers. Harbor seals, despite being the most widely dispersed among pinnipeds (the group that includes seals, walruses, and sea lions), still have much unknown about them. Traditional tracking methods, such as tagging and aerial monitoring, often prove invasive or prohibitively costly.
SealNet presents an opportunity to gain more insight into seal behavior, specifically site fidelity. The system revealed that some harbor seals return to the same haul out sites annually. Increased understanding of seal movements could bolster arguments for protecting specific areas, thus significantly aiding conservation efforts.
The project is even receiving praise from those not directly involved with it, ecologists at Aarhus University in Denmark responsible for monitoring Denmark’s seal populations, commended the software, highlighting its “promise.” They suggests combining it with another photo identification method that recognizes seals by their distinct markings, subject to improvement in identification rates.
Looking to the future, the team anticipates developing an app based on SealNet. This app could engage citizen scientists in logging seal faces, potentially expanding the scope of this AI environment to other pinnipeds and possibly even cetaceans. SealNet’s transformative potential shows the vast untapped potential of the AI environment in environmental research and conservation, with the promise of a future where technology and nature collaborate for mutual benefit.
