Identifying and Tracking Individual Coconut Crabs with AI

The worldwide coconut crab population is in decline, and to get an accurate count of how many remain, researchers in Japan developed a deep learning system to identify and track specific coconut crabs individually.
Coconut crabs are among the biggest crabs in the world weighing up to four pounds and measuring nearly three feet long.  The automated method identifies individual crabs using their shell patterns, making the process far more efficient and precise than doing it manually.
“The coconut crab is an endangered species,” said Chonho Lee, a professor, and lead researcher on this project at Osaka University.  “One of the goals is to track and analyze their behaviors.”
Using NVIDIA Tesla P100 GPUs and the cuDNN-accelerated TensorFlow and Keras deep learning frameworks, the team trained their neural network on hundreds of shell pattern images taken from a top view of the crabs. The group also performed a data augmentation process to create additional samples for training.
Shell patterns of coconut crabs differ from creature to creature, making this process similar to identifying an individual fingerprint.
“Identifying crab individuals from images has a few benefits, one is we can avoid injecting sensors or chips into crabs,” Lee included.

So far the team has counted over 400 individual crabs at the Okinawa Churashima Foundation Research Center’s Ocean Expo Park in Motobu, Okinawa.
According to The Mainichi, a newspaper in Japan, the center has photographed around 1,500 crabs and cross-checked them with photographs of about 300 to 500 crabs.
The coconut crab identification project is a collaboration among the Okinawa Churashima Foundation, Cogito Inc, and the Mainichi Shinbun and Cybermedia Center at Osaka University.
The team is currently working on an automated robotic solution that uses an NVIDIA Jetson embedded platform to automatically identify and track the crabs.
Read more >