Proposal based saliency model for generic target detection in remote sensing image

Object detection in remote sensing images have become prominent due to their importance in remote sensing image analysis. This paper presents a novel automatic generic objects detection method that is based on coarse-to-fine saliency. First, we proposed a background based sparse reconstruction algorithm to construct a coarse saliency map which can precisely highlight the salient foreground while suppress the background. Then, we collect training samples from coarse saliency map for second step. Second, a strong classifier based on the training samples is constructed to detect salient pixels. By introducing the object proposals method to enhance the results from the strong classifier, we construct the fine saliency map which can highlight the target completely. In order to further improve the detection performance, multi-scale saliency maps are integrated to generate the final saliency map. Quantitative analyses of experiment results on a real remote sensing image data set containing 200 images of airport, residence and oil-tank verify that proposed algorithm outperforms 10 state-of-art saliency models.