Study Of Special Targets Detection Techniques In SAR Images

Synthetic Aperture Radar (SAR) has been widely used in both military reconnaissance and civil activity based on its unique advantages, so it's meaningful and has application prospect to study target detection method of SAR images. In this dissertation, we detailedly analyze the special targets detection methods of SAR images based on directional filter, Nonsubsampled Contourlet Transform (NSCT) and extended fractal feature, and validate them by experiments.A new algorithm for SAR images speckle reduction is proposed based on adaptive shrinkage in NSCT Domain. The Nonsubsampled Contourlet coefficients of SAR images at high frequency subband are modified by the corresponding Pizurica adaptive shrinkage factors. The shrinkage factor takes into account not only the local noise measure, but also prior directional spatial consistency, and combines the shift-invariance, direction selectivity and multiresolution of the Nonsubsampled Contourlet transform. Experiments show that the filter algorithm can reduce speckle noise more effectively while preserving the edges of the SAR images.Based on directional selectivity of directional filter bank, an algorithm for straight line targets detection in SAR images is put forward. The straight line target can be detected by searching two parallel straight lines along its edges. Many real SAR images are used to illustrate our method, and the performance is satisfactory.A new method of detecting target and estimating target azimuth in SAR images is proposed based on NSCT energy feature. The technique mainly focuses on the different NSCT energy distribution feature of the background and the targets. Considering the cross-scale distribution character and manifest directional character of target pixels, we can detect targets and estimate their azimuth in energy feature image and directional feature image. The results of experiments prove that this method has good performance on target detection and meanwhile can get accurate targets azimuth range.The false alarm causation of the algorithm for target-sized objects detection in SAR images based on extended fractal feature is analyzed. The original algorithm is improved by combining consistency of target pixels gray and contrast of targets and background. Experiments show that this improved algorithm provides a lower false alarm rate and higher figures of merit than the original one.