Machine Learning Algorithms for Automatic Identification of Limnonectes Species Using Image Data

• Activity Number: 135 - Applications of Machine Learning Methods to Imaging Data Analysis
• Type: Contributed
• Date/Time: Monday, July 29, 2019 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Imaging
• Author(s): Li Xu* and Eric Smith and Yili Hong and David McLeod
• Companies: Virginia Tech and Virginia Tech and Virginia Tech and James Madison University
• Keywords: deep learning; classification; computer vision

Abstract

Limnonectes cf. kuhlii, is a species complex that has more than 22 unique evolutionary lineages (candidate species). Morphology is commonly used to delineate species boundaries, but the characters that we have historically employed may not be adequate for resolving fine scale differences between species — especially those belonging to a species complex. Among the L. kuhlii complex lineages, one characters that seems to have great potential for differentiating species, but has been poorly explored, is the texture of the skin on the hind legs (tuberculation). This tuberculation can be described qualitatively on the spectrum of smooth to rough or dense to sparse. In this paper, we investigate machine learning methods such as deep neural networks, classification trees and logistic regression models in the classification of L. kuhlii complex lineages. With a large number of images from frogs collected across Southeast Asia, we develop a classifier that takes a photograph of an individual frog as the input and then automatically classifies it into one of the existing species for which we have an existing set of images.