Abstract:

Because the most existing ship detection methods do not perform well for various high sea clutter situations, according to the characteristic difference between man?made object and natural background, we propose a novel saliency detection model by computing the eigenvalues of region covariances. To the best of our knowledge, there is no existing reference in the literature about exploring the relationship between the different eigenvalues of region covariances and saliency so far. We further integrate the feature maps linearly by assigning adaptive weight value based on information entropy. The proposed approach has a remarkable ability to pop out targets and suppress distractors against clutter introduced by heavy clouds, islands as well as ship wakes. Furthermore, our model is fast and efficient which has great potential in engineering applications. Extensive experiments have been carried out on optical satellite images and experimental results demonstrate that our approach outperforms 5 existing salient object detection methods in the Area Under the Receiver Operating Characteristics.

Key words: information processing technology, ship detection, visual attention model, region covariance, information entropy