The National Institute of Biological Resources, under the Ministry of Environment, announced on the 21st that it has recently developed a customized AI-based species classification technology that can quickly and accurately identify 13 species of turtles, including snapping turtles.
This technology applies hyperparameter optimization and instance segmentation methods. Hyperparameter optimization is a technique for adjusting settings to maximize model performance according to the dataset, while instance segmentation is a method of precisely distinguishing and recognizing objects in images.
Globally, there are about 378 species of turtles, but it is often difficult to distinguish them solely by appearance due to similarities in shell shape or changes caused by long-term feeding and impacts. This has posed challenges in export and import management.
Since 2021, the National Institute of Biological Resources has promoted a project titled “Human Resources Development Using Biological Information Big Data” in collaboration with Professor Changbae Kim’s team from Sangmyung University, collecting a variety of turtle photos. Based on this, they built a big data-based AI model and developed a turtle-specific species identification technology.
In this study, the performance of the identification was verified on 13 species, including snapping turtles, alligator snapping turtles, and the Chinese stripe-necked turtle, classified as invasive species. By applying hyperparameter optimization technology, the snapping turtle, alligator snapping turtle, and Chinese stripe-necked turtle achieved up to 99% accuracy. The hawksbill sea turtle, green sea turtle, and loggerhead sea turtle, three species of sea turtles, achieved an average accuracy of 92.5% by applying instance segmentation technology.
The National Institute of Biological Resources expects that this technology will be able to derive results faster and more accurately than existing methods such as DNA analysis. They also announced plans to collaborate with relevant agencies to apply it to wildlife export and import management sites.