Finding solutions to improve turtle re-identification and supporting machine learning projects across Africa
Protecting the ecosystems around us is crucial to securing the future of our planet and all its living citizens. Fortunately, novel artificial intelligence (AI) systems are advancing conservation efforts around the world, helping to solve sophisticated problems on a gigantic scale – from studying the behavior of animal communities in the Serengeti to lend a hand protect a vanishing ecosystem, to looking out for poachers and their injured victims to prevent the extinction of species.
As part of our mission to lend a hand humanity with the technologies we develop, it’s essential that we ensure that diverse groups of people build the AI systems of the future so that they are fair and equitable. This includes expanding the machine learning (ML) community and engaging a broader audience in solving essential problems with AI.
As a result of the investigation we came across Others – a dedicated partner with complementary goals – that builds the largest community of African data scientists and organizes competitions aimed at solving Africa’s most pressing challenges.
Our Scientific teamThe Diversity, Equity, and Inclusion (DE&I) team worked with Zindi to identify a science challenge that could lend a hand drive environmental action and escalate engagement with AI. Inspired by Zindi box bounding turtle challengewe came across a project that could have a real impact on reality: recognizing turtle faces.
Biologists consider turtles an indicator species. They are a class of organisms whose behavior helps scientists understand the basic well-being of their ecosystem. The presence of otters in rivers, for example, is considered a sign of a pristine, robust river, since a ban on chlorine pesticides in the 1970s brought the species back from the brink of extinction.
Turtles are another such species. By grazing on seagrass cover, they maintain the ecosystem, providing habitat for numerous fish and crustaceans. Traditionally, individual turtles have been identified and tracked by biologists using physical tags, but the constant loss or erosion of these tags in seawater has made this method unreliable. To lend a hand address some of these challenges, we launched an ML challenge called Turtle withdrawal.
Given the added challenge of keeping the turtle still long enough to find its tag, the Turtle Recall challenge was designed to circumvent these problems with turtle facial recognition. This is possible because the pattern of scales on a turtle’s face is unique to the individual and remains the same throughout its multi-decade life.
The goal of the challenge was to improve the reliability and speed of turtle re-identification, and potentially offer a way to completely replace the cumbersome physical tags. To make this possible, we needed a dataset that we could work with. Fortunately, after a previous turtle-based Zindi challenge with a Kenyan charity Local Ocean ProtectionThe teams kindly provided a dataset containing annotated images of turtle faces.
The competition started in November 2021 and lasted for five months. To encourage participants to participate, the team implemented Collaboration Notebooka browser-based development environment that introduced two popular development tools: JAX and Haiku.
Participants were tasked with downloading challenge data and training models to predict the identity of a turtle as accurately as possible from a photo taken from a specific angle. After submitting their predictions based on the data omitted from the model, they could visit a public leaderboard that tracks each participant’s progress.
The community engagement was overwhelmingly positive, as were the technical innovations presented by the teams during the challenge. During the competition, we received submissions from a diverse range of AI enthusiasts from 13 different African countries – including countries that are traditionally underrepresented at major ML conferences, such as Ghana and Benin.
Our turtle conservation partners have indicated that the accuracy of participants’ predictions will be immediately useful in identifying turtles in the field, meaning these models can have a real and immediate impact on wildlife conservation.
As part of Zindi’s ongoing efforts to support positive climate challenges, they are also working on Swahili Audio Classification in Kenya to lend a hand with translation and emergency services, and air quality forecast in Uganda to improve social well-being.
We are grateful to Zindi for their partnership and to all those who have volunteered their time to the Turtle Recall Challenge and the emerging field of AI for conservation. And we look forward to seeing how people around the world continue to find ways to apply AI technology to build a robust, sustainable future for the planet.
Read more about Turtle’s product recall at Zindi’s Blog and find out more about Zindi at https://zindi.africa/