Please use this identifier to cite or link to this item:
Title: Identification of alien objects underwater
Authors: Chetcuti, Stephanie (2019)
Keywords: Neural networks (Computer science)
Transfer learning (Machine learning)
Autonomous underwater vehicles
Computer vision
Issue Date: 2019
Citation: Chetcuti, S. (2019). Identification of alien objects underwater (Bachelor's dissertation).
Abstract: From the earliest sailors centuries ago to the present day, the human fascination with the deep blue seas has never waned. The technology available to humans today is far different from what was available to our ancestors thousands of years ago, but understanding the oceans still contains challenges. The task of underwater object detection is one such area which is fraught with difficulties. Underwater environments vary greatly from one another, with each environment holding its own unique inherent qualities and features. Neural Networks and Deep Learning approaches have proven their capabilities in in-air imagery and, as a result, this has sparked an interest to train these same models and approaches for use on underwater images. However, collecting a large enough dataset is a tedious task which is often deemed infeasible. Furthermore, attempting to train a model on a small sample size will lead to over- fitting. Overcoming these challenges would prove useful for a variety of different fields ranging from the environmental, through ocean cleanups, the economical, through pipeline inspections, and the historical, through underwater archaeology, along with various other fields. To overcome the problem of over- fitting, the approach taken in this project was to use a transfer learning technique, with the argument that Convolutional Neural Networks are not only classifiers but are also feature extractors. Hence, a CNN trained on a large dataset of in-air images will be sufficient enough to classify objects in underwater scenes after some fine-tuning using images taken underwater since the pre-trained model will already be sensitive to information such as colours, textures and edges. Mask R-CNN is the chosen model used for this project and achieved a Mean Average Precision of 0:509.
Appears in Collections:Dissertations - FacICT - 2019
Dissertations - FacICTAI - 2019

Files in This Item:
File Description SizeFormat 
Chetcuti Stephanie.pdf
  Restricted Access
5.57 MBAdobe PDFView/Open Request a copy

Items in OAR@UM are protected by copyright, with all rights reserved, unless otherwise indicated.