Home AI News Improving Turtle Reidentification and Supporting ML Projects for Conservation in Africa

Improving Turtle Reidentification and Supporting ML Projects for Conservation in Africa

Improving Turtle Reidentification and Supporting ML Projects for Conservation in Africa

Finding Solutions to Improve Turtle Reidentification and Support AI in Conservation

Protecting our planet’s ecosystems and its inhabitants is crucial for the future of our world. Fortunately, advancements in artificial intelligence (AI) are playing a vital role in global conservation efforts. From studying animal behavior in the Serengeti to identifying poachers and their prey, AI systems are making a significant impact.

In our commitment to benefiting humanity through technology, it is essential to ensure diverse groups of people participate in building the AI systems of the future. This includes expanding the machine learning (ML) community and engaging wider audiences in solving important problems using AI.

During our exploration, we discovered Zindi – a valuable partner with similar goals. Zindi is the largest community of African data scientists and hosts competitions focused on solving Africa’s most pressing issues.

Working hand in hand with Zindi’s Diversity, Equity, and Inclusion (DE&I) team, we identified a scientific challenge that would advance conservation efforts and promote AI involvement. Inspired by Zindi’s bounding box turtle challenge, we decided on a project with significant potential – turtle facial recognition.

The Importance of Turtle Reidentification in Conservation

Turtles are considered indicator species because their behavior provides insights into the health of their ecosystem. They play a vital role in maintaining the balance of marine ecosystems by grazing on seagrass, which creates habitats for numerous marine life. Traditionally, individual turtles have been identified and tracked using physical tags, but this method is unreliable due to tag loss or erosion in seawater.

To address these challenges, we launched an ML challenge called Turtle Recall, aimed at developing more reliable and speedy turtle reidentification methods. The goal was to explore the possibility of eliminating physical tags altogether by using turtle facial recognition technology.

The Turtle Recall Challenge and Dataset

The Turtle Recall challenge faced the additional hurdle of keeping turtles still enough to locate and tag them accurately. However, turtle facial recognition became a viable solution because the pattern of scales on a turtle’s face remains unchanged throughout its multi-decade lifespan.

In collaboration with the Kenyan-based charity Local Ocean Conservation, we were able to obtain a dataset of labeled images of turtle faces from a previous turtle-based challenge organized by Zindi. This dataset became the foundation for the Turtle Recall challenge.

Community Engagement and Innovation

The competition took place from November 2021 and lasted for five months, attracting participants from all over Africa. To encourage participation, we provided a colab notebook – an in-browser programming environment – that introduced common programming tools like JAX and Haiku.

Participants were tasked with downloading the challenge data and training models to accurately predict a turtle’s identity based on a photograph taken from a specific angle. Their predictions were evaluated using withheld data, and a public leaderboard tracked their progress.

The Turtle Recall challenge garnered significant community engagement and showcased impressive technical innovation. We received submissions from AI enthusiasts across 13 different African countries, including Ghana and Benin, which are not typically well-represented in major ML conferences.

Real Impact on Wildlife Conservation

The accuracy of participants’ predictions in the Turtle Recall challenge will immediately benefit turtle identification in the field. These models can have a tangible and immediate impact on wildlife conservation efforts.

Zindi continues to lead climate-positive initiatives, such as Swahili audio classification in Kenya and air quality prediction in Uganda, leveraging AI to enhance translation services and social welfare.

We are thankful for our partnership with Zindi and appreciate the valuable contributions made by all those involved in the Turtle Recall challenge. We look forward to witnessing how AI technologies will continue to contribute to building a healthy and sustainable future for our planet.

Find more information about Turtle Recall on Zindi’s blog and explore Zindi’s initiatives at https://zindi.africa/.

Source link


Please enter your comment!
Please enter your name here