A Study On Target Detection On Multi-polarimetric Synthetic Aperture Radar Images

With the increasing of available data from Poiarimetric Synthetic Aperture Radar, it is possible for the exploitation and application of poiarimetric information on poiarimetric SAR images. Nowadays, the study of target detection on multi-polarimetric SAR images has not been deeply investigated. The purpose of this thesis is to study on the target detection algorithms used in poiarimetric SAR images, and to evaluate the detection performance of different algorithms and different multi-polarimetric combinations. Based on the basic theory of poiarimetric SAR and its measured model, the SIR-C and ENVISAT data is used to analysis the detection performance of different detectors with regards to different background models and real background data, and ship detection in SIR-C and ENVISAT multi-polarimetric images is experimented as well.After brief introduction of poiarimetric SAR theory and different clutter and target models, the Optimal Poiarimetric Detector (OPD), the Poiarimetric Whitening Filter (PWF) detector and other detectors are introduced, among which the PWF detector provides the best trade-off between detection performance and pre-known statistical knowledge.The detection performance is evaluated based on different kinds of background and different poiarimetric combinations. Based on the ComplexGaussian target and clutter model, the calculation of detection probability and false alarm probability is theoretically deduced, the application on land and ocean targets using full polarimetric data is compared, also the dual polarimetric detectors is also discussed. Based on the Non-Gaussian clutter model, channel independent methods are used to simulate polarimetric SAR images and the detection rate and false alarm rate are calculated using land and ocean data. Based on the real SAR background and target, the Receiver Operating Characteristic (ROC) curves are constructed to describe the performance and analysis the errors. Also the amplitude PWF method using SIR-C and ENVISAT data is included.The ship detection experiments in SIR-C and ENVISAT images are carried out using both centralized and decentralized methods, yielding similar conclusion with theoretical analysis. The improved PWF method is proposed by adding the parameter estimation procedure and its improved performance has been demonstrated as well. For ENVISAT dual polarimetric data, both PWF based centralized and "AND" "OR" based decentralized methods are used to detect ships, and the performance of different methods is compared.