A carregar...
Publicação
Ship recognition on the sea surface using aerial images taken by Uav : a deep learning approach
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 6.93 MB | Adobe PDF |
Autores
Resumo(s)
Oceans are very important for mankind, because they are a very important source of
food, they have a very large impact on the global environmental equilibrium, and it is
over the oceans that most of the world commerce is done. Thus, maritime surveillance
and monitoring, in particular identifying the ships used, is of great importance to
oversee activities like fishing, marine transportation, navigation in general, illegal
border encroachment, and search and rescue operations. In this thesis, we used images
obtained with Unmanned Aerial Vehicles (UAVs) over the Atlantic Ocean to identify
what type of ship (if any) is present in a given location. Images generated from UAV
cameras suffer from camera motion, scale variability, variability in the sea surface and
sun glares. Extracting information from these images is challenging and is mostly done
by human operators, but advances in computer vision technology and development of
deep learning techniques in recent years have made it possible to do so automatically.
We used four of the state-of-art pretrained deep learning network models, namely
VGG16, Xception, ResNet and InceptionResNet trained on ImageNet dataset, modified
their original structure using transfer learning based fine tuning techniques and then
trained them on our dataset to create new models. We managed to achieve very high
accuracy (99.6 to 99.9% correct classifications) when classifying the ships that appear
on the images of our dataset. With such a high success rate (albeit at the cost of high
computing power), we can proceed to implement these algorithms on maritime patrol
UAVs, and thus improve Maritime Situational Awareness.
Descrição
Dissertation submitted in partial fulfilment of the requirements for the Degree of Master of Science in Geospatial Technologies
Palavras-chave
Ship Recognition Classification UAV Images Deep Learning Deep Convolutional Neural Networks Transfer Learning Maritime Surveillance