Thursday, January 29, 2015

Pattern-recognition software usage in identifying individual Eastern Box Turtles

 Introduction

     In studies of the behavior and ecology of wild animals, being able to identify and monitor individuals is essential. After all, without marking, how do you know that the turtle you just picked up is the one you're studying or not? Traditional methods of tagging, like bands and transponder tags are all good and well, but they might cause altered behavior, stress, or opportunities for infection, and usually, when studying behavior and ecology, you want fairly healthy, happy animals whose behavior is the way it normally would be. Now, with modern technology, pattern recognition via photographs is possible, but with a large number of images, that's time-consuming and frankly, quite boring. But, what if you could use pattern recognition software to identify turtles? It's cost-effective, doesn't affect the behavior of the animals, and could be much better.

Goals
     The goals of the project were to see if pattern recognition was a decent method of identifying box turtles, to determine which parts of the turtles it's best to take photos of for pattern recognition, and to see if the pattern recognition program could identify individuals from different populations.


Methods
     Photographs of turtles were collected from the Oak Openings region of northwest Ohio, the Ft. Custer Recreation Area in Michigan's southwestern area, and the Manistee National Forest from 2009 to 2013. The reason images from different states were collected was to see if images from different places would end up with false matches. Upon capture, pictures were taken of the carapace and plastron of the turtles, and there were also images taken from any angle. The pictures were cropped to contain the least possible background.Very young turtles with a plastron length of less than 7 centimeters weren't used because their shell patterns weren't fully developed. The program used was Wild-ID, which in its user interface, compared each focal image with the top 20 images and lets the user discern visually which is a match. If it was a recapture, in no case was the top-ranked image a different turtle. Wild-ID didn't erroneously match images between sampling locations or states, as confirmed by notches on shells and PIT tags.

Results
     Wild-ID worked fairly well for identifying Eastern Box Turtle recaptures, but didn't work quite as well with off-center images. There was no real statistical difference between carapace and plastral images, even though it was expected that plastron images wouldn't have quite as much in the way of identifying patterns. There were no mismatches between states, so therefore Wild-ID is in fact suitable for identifying individual Eastern Box Turtles!

Wild-ID's user interface, image from original work.


Cross, Matthew, Lipps, Gregory, Sapak, Janice, Tobin, Eric, Root, Karen. "Pattern-Recognition Software as a Supplemental Method of Identifying Individual Eastern Box Turtles (Terrapene c. carolina)" Herpetological Review, 45(4), 584-586. 2014.

No comments:

Post a Comment