Applied Anomaly Detection on unsupervised detection of marine mammals
Tutor / directorRomero Merino, Enrique
Document typeMaster thesis
Rights accessOpen Access
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder
In this report, techniques for Anomaly Detection are explored with the aim of automating the process of detecting marine mammals in satellite imagery. This can be later be applied to extract and prepare data for an object detection system. The project is mostly comprised by an exploration of two different approaches, Generative Adversarial Networks and Auto-encoders. On top of that, a brand new synthetic dataset was created for this Computer Vision task and contribute to the research of similar oceanic challenges. Three new metrics will be explored for the task of making an assessment and comparison of the models. The mentioned exploration of the models for Anomaly Detection will also report results and analyses on experiments that test configurations and methodologies to use such models in the task of Anomaly Detection. While other projects might focus their efforts into devising brand new approaches or documenting the state of the art, most of the work in this project will revolve around the applicability and experimentation process.
DegreeMÀSTER UNIVERSITARI EN INTEL·LIGÈNCIA ARTIFICIAL (Pla 2017)