Multi-Scale Salient-Region Detection for Optical Remote Sensing Images

Ship detection is an important part of marine monitoring system. In order to detect ship targets in optical satellite images under complex sea background, a salient ship detection algorithm based on multi-scale in frequency domain is proposed in this paper. In this method, we first use hypercomplex Fourier transform to get the amplitude matrix and phase matrix of the image and then use wavelet transform to calculate six saliency images of different scales in the frequency domain. And we use the evaluation function to select two best multi-scale saliency maps to generate the final saliency map by adaptive weight combination. Finally, the threshold is used to process the enhanced saliency map to locate the ship target. Experiments on relevant ship data show that the proposed method can improve the detection accuracy, detect ship targets of different sizes, effectively filter out wave clutter and noise, and is robust to sea clutter spots and heterogeneous sea areas.