Graduate Exam Abstract

Abstract: An effective automated algorithm is proposed to solve the real-world animal identification problem: identifying K unknown individual animals in N unlabeled camera trap images of a given species, African leopards, with most animals only represented by a single image. The proposed algorithm is different from other methods that assume all images in a dataset are from known individuals, i.e., regarding the animal ID problem as a retrieval identification task. This approach consists of a leopard segmentation technique, an image similarity scoring mechanism, a new adaptive k-medoids++ clustering algorithm, and a novel post-clustering verification procedure. To determine leopards' IDs, a modified ternary search is employed to determine the best clustering that is estimated by an expanded definition of the silhouette score. The accuracy of this proposed algorithm was demonstrated on a real-world image dataset of African leopards, a small dataset with a relatively large ratio of K/N, courtesy of Panthera.

Adviser: Anthony Maciejewski
Co-Adviser: Agnieszka Miguel
Non-ECE Member: Asa Ben-Hur, CSc
Member 3: Ali Pezeshki, ECE
Addional Members: N/A

Publications:
Cheng Guo, Agnieszka Miguel, and Anthony A. Maciejewski. Automatic identification of individual african leopards in unlabeled camera trap images. IEEE Transactions on Automation Science and Engineering, pages 1–12, 2024

Program of Study:
ECE455 - Introduction to Robot Programming/Simulation
ECE555 - Advanced Robotics: Redundancy and Optimization
ECE512 - Digital Signal Processing
ECE513 - Digital Image Processing
ECE520 - Optimization Methods-Control and Communication
CS445: Introduction to Machine Learning
CS545: Machine Learning
GRAD 550 - STEM Communication