Texture-based vessel classifier for electro-optical satellite imagery

Satellite imagery provides a valuable source of information for maritime surveillance. The vast majority of the research regarding satellite imagery for maritime surveillance focuses on vessel detection and image enhancement, whilst vessel classification remains a largely unexplored research topic. This paper presents a vessel classifier for spaceborne electro-optical imagery based on a feature representative across all satellite imagery, texture. Local Binary Patterns were selected to represent vessels for their high distinctivity and low computational complexity. Considering vessels characteristic super-structure, the extracted vessel signatures are sub-divided in three sections bow, middle and stern. A hierarchical decision-level classification is proposed, analysing first each vessel section individually and then combining the results in the second stage. The proposed approach is evaluated with the electro-optical satellite image dataset presented in [1]. Experimental results reveal an accuracy of 85.64% across four vessel categories.