Ensemble Transfer Learning Framework for Vessel Size Estimation from 2D Images (CROSBI ID 677192)
Prilog sa skupa u zborniku | izvorni znanstveni rad | međunarodna recenzija
Podaci o odgovornosti
Miličević, Mario ; Žubrinić, Krunoslav ; Grbavac, Ivan ; Kešelj, Ana
engleski
Ensemble Transfer Learning Framework for Vessel Size Estimation from 2D Images
The term gross tonnage refers to the internal volume of a vessel and it has several legal, administrative and safety uses. Therefore, there is significant value in developing a mechanism for the automatic estimation of vessel size based on 2D images taken in uncontrolled conditions. However, this is a demanding task as vessels can be photographed from various angles and distances, a part of a vessel can be obstructed, or a vessel can blend with the background. We proposed an ensemble of fine-tuned transfer learning models, which we trained on 20, 000 images in a training dataset consisting of randomly downloaded images from the Shipspotting website. Multiple deep learning methods were applied and modified for regression problems, together with two classical machine learning algorithms. A detailed analysis of model performances was given, based on which it can be concluded that such an approach results in a vessel size evaluation of the same quality as with the best human experts from the corresponding field.
deep learning ; convolutional neural networks ; transfer learning ; regression ; ensemble methods ; computer vision ; vessel size estimation
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
nije evidentirano
Podaci o prilogu
258-269.
2019.
objavljeno
10.1007/978-3-030-20518-8_22
Podaci o matičnoj publikaciji
Advances in Computational Intelligence. IWANN 2019. : Conference proceedings
Rojas, Ignacio ; Joya, Gonzalo ; Catala, Andreu
Springer
978-3-030-20521-8
Podaci o skupu
15th International Work-Conference on Artificial Neural Networks
predavanje
12.06.2019-14.06.2019
Las Palmas de Gran Canaria, Španjolska