Accurate and robust animal species classification in the wild
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Thesis
Subject
Abstract
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI--COLUMBIA AT REQUEST OF AUTHOR.] Wildlife monitoring with camera-traps allows us to collect data at large scales in space and time to study the impact of climate changes, land-use, and human actions on wildlife population dynamics, and biodiversity. With the increase of camera trap used per study and the large number of images generated by camera traps, processing and managing the generated images has become a big challenge. Unlike many other image processing and vision analysis tasks, detecting, segmenting, and classifying animal species from the camera-trap images is very challenging since natural scenes in the wild are often highly cluttered due to heavy vegetation and dynamic background. In this dissertation, we focus on animal object detection and species classification in camera-trap images collected in highly cluttered natural scenes. Using a deep neural network (DNN) model trained for animal- background image classification, we analyze the input camera-trap images to generate its multi-level visual representation. We detect semantic regions of interest for animals from this representation using k-mean clustering and graph cut in the DNN feature domain. These animal regions are then classified into animal species using multi-class deep neural network model. According the experimental results, our method achieves 99.75% accuracy for classifying animals and background and 90.89 percent accuracy for classifying 26 animal species on the Snapshot Serengeti dataset, outperforming existing image classification methods. We develop a robust learning method for animal classification from camera-trap images collected in highly cluttered natural scenes and annotated with noisy labels. We proposed two different network structures with and without clean samples to handle noisy labels. We use k-means clustering to divide the training samples into groups with different characteristics, which are then used to train different networks. These networks with enhanced diversity are then used to jointly predict or correct sample labels using max voting. We evaluate the performance of the proposed method on two public available camera-trap image datasets: Snapshot Serengeti and Panama-Netherlands datasets. Our experimental results demonstrate that our method outperforms the state-of-the-art methods from the literature and achieved improved accuracy on animal species classification from camera-trap images with high levels of label noise. We also develop a new approach for learning a deep neural network for image classification with noise labels using ensemble diversified learning. We first partition the training set into multiple subsets with diversified image characteristics. For each subset, we train a deep neural network image classifier. These networks are then used to encode the input image into different feature vectors, providing diversified observations of the input image. The encoded features are then fused together and further analyzed by a decision network to produce the final classification output. We also study image classification on noise labels with and without the access to clean samples. Our extensive experimental results on the CIFAR-10 and MNIST datasets demonstrate that our proposed method outperforms existing methods by a large margin.
Table of Contents
DOI
PubMed ID
Degree
Ph. D.
Thesis Department
Rights
Access to files is limited to the University of Missouri--Columbia.
License
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 License. Copyright held by author.