Author: Kiyrah Keith (kiyrahkeith@gmail.com)
Mentor: Dr. Szilárd Vajda
Central Washington University
This research was funded by the Ronald E. McNair Post-Baccalaureate Achievement Program through a grant from the U.S. Department of Education
TensorFlow, Python, Keras, TensorBoard
Animal Face Dataset can be located at https://data.mendeley.com/datasets/z3x59pv4bz/7
CZoo and CTai Datasets can be located at https://github.com/cvjena/chimpanzee_faces/
Recent advancements in machine learning, coupled with the availability of affordable and powerful hardware, have led to a surge in studies within the interdisciplinary field of animal recognition using computer vision. These studies have enabled significant achievements such as monitoring invasive species, protecting against poaching, improving public road safety, estimating population sizes, and enhancing animal welfare both in captivity and in the wild. Despite these successes, a concise set of guidelines to help small-scale research projects achieve high accuracy in animal classification models remains lacking. This research provides a comprehensive review of the current state of the field, highlighting common challenges and solutions emerging from collaborations between animal experts and computer scientists. A subset of these solutions has been tested on publicly available non-human primate face datasets, successfully identifying individuals and broader species with high accuracy. The findings demonstrate that small, deliberate adjustments to convolutional neural network models during training can yield accuracy above 95%, even on small and unbalanced datasets. These conclusions are significant for the broader goal of animal monitoring, as they underscore the feasibility of integrating machine learning as a cost-effective and time-saving tool for processing large volumes of visual data.
Analysis of each training cycle results can be found in Model Training Results.xlsx
Please see Primate_Recognition_Report.pdf for the full research report.