Ship Extraction using Post CNN from High Resolution Optical Remotely Sensed Images

A post convolutional neural networks (CNN) method is proposed to extract ships from high resolution optical remotely sensed images. It consist of two parts: ship proposal detection and ship extraction based on CNN. The first part aims to locate possible ships through classification of water and no-water, seawater area extraction using mathematical morphology, and ship proposal extraction. Ships are extracted in the second part by implementing a trained CNN on the ship proposals. Experimental results on a high resolution optical image of San Francisco Bay show the efficiency and robustness of the proposed post CNN method.