Multiple ship targets association method of remote sensing images based on SIFT and bags of visual words model

Ships with unique moving and group activity characteristics can be considered as a type of very important military targets in marine situation awareness. We propose a novel method for multiple ship targets association with utilizing multi-temporal optical satellite images. First, a large number of local invariant features are extracted from the region of the optical satellite imagery containing the ship targets. Second, we present a weighted Bag of Visual Words model to perform transforming the 128-dim features to high-order semantic features. Finally, the optimization model based on the Associated Cost Matrix is constructed to solve the target optimal correlation matching. The experimental results clearly demonstrate that the proposed method is robust to multi-target association ambiguity and produces good matching accuracy with low computational effort.