Ship Detection Using Machine Learning and Optical Imagery in the Maritime Environment

Abstract

Machine Learning (ML) is increasingly being used to enable machines to aid our understanding of digital imagery. Convolution Neural Network (CNN) models have become a common method of implementing ML. In this paper, five award winning CNN models have been assessed for their ability to identify ship types in images from remote sensors. The CNNs are trained with a collection of known ship images using supervised learning techniques, and, due to the low number of images available for training, Transfer Learning (TL) is adopted to take advantage of pre-trained low-level features.