A novel scheme of unsupervised target detection for high-resolution SAR image
Song Tu, Li Yu, Yi Su · 2014
How to detect the interested targets efficiently and accurately from a large-scale and high-resolution synthetic aperture radar (SAR) image is still a research challenge. This paper presents a novel scheme based on saliency detection approach and active contour model (ACM) for SAR image detection. Due to the high efficiency of Spectral Residual (SR) approach, the scheme can find the potential interested regions rapidly. Then a modified local and global intensity fitting (MLGIF) ACM based on ratio and distribution metric is proposed in this paper, which overcomes the defect of some well-known ACMs tending to fall into local minimums in SAR image detection. Due to the robustness of the MLGIF model to multiplicative speckle, the detection scheme can locate the targets more accurately and is more suitable to SAR image processing. Experiments of large-scale and high resolution SAR image detection show that the proposed scheme outperforms classical Constant False Alarm Rate (CFAR) Detector in terms of the efficiency and false alarm.