Feature Extraction Of Target Chips In SAR Images

In military field, Synthetic Aperture Radar (SAR) has been an important method for intelligence requiring. People get required intelligences by the interpretation of SAR images. With the advancement of the capacity of acquiring SAR images, Automatic Target Recognition (ATR) of SAR images has become an important aspect in the interpretation of SAR images. The feature extraction of the SAR targets can reduce the dimension of the data to be processed and recognize the target rapidly, so it plays a key role in a SAR ATR system. According to the data we used in the process of feature extraction, we classify the features of SAR targets into geometrical features and electromagnetic features, which will be analysed and extracted subsequently.From the aspect of the extracting of geometrical features, two image segmentation methods are investigated firstly to acquire target area in SAR images, the performance of the two methods is also analysed and compared. We analyse the leading contour method and the encapsulating box method which are both commonly used in azimuth estimation. A combined method is proposed which improves the accuracy of the two estimation methods mentioned above. And then we analyse the mathematic characteristics and corresponding extraction methods of several point features by the using of the concept in topography. A series of area features are extracted in the last of the chapter.From the aspect of the extracting of electromagnetic features, we use the attributed scattering center model to describe target's electromagnetic characteristics in high-frequency region, by which we investigate the attributed scattering center features in stead of electromagnetic features. We introduce the concept of the attributed scattering center feature, the parameteric model, the theory of the SAR image formation and the normalization of the parameteric model. The method of parameter estimation is analysised and approved. We also simplify the analysis of the model and the parameter estimation, and approve the effectivity of the feature extraction method with simulated SAR data. At last the Cramer-Raw Low Bound (CRLB) of the estimation variance is derived, and the effect of the resolution and SNR versus the performance are analysed respectively.