Abstract

Optical satellite imagery have been extensively used in marine environmentology, naval surveillance and sea border activities. Ship identification is a vital aspect of optical satellite imagery in harbor dynamic reconnaissance and oceanic administration. Prominent ship detection techniques are bounded by their time complexity and execution accuracy. This paper proposes a strategy of scale invariant feature transform (SIFT) algorithm and a convolution neural network (CNN) which is a class of deep learning. In this paper, a hierarchical method to amalgamate the feature extraction, efficient object masking and an accurate object identification model is proposed where feature extraction is done by applying SIFT algorithm and object detection by CNN. First, in the feature extraction phase SIFT algorithm is applied where scale space extrema detection, keypoint localization, orientation assignment, keypoint descriptor are considered specifically to improve the robustness regarding detection, patented local feature detector and image masking. Finally for the object detection, these sub-models are integrated and output is given to convolution neural network. Ship detection is done by bounding boxes. By implementing CNN-SIFT technique, 97% of accuracy is achieved.