Research On Target Detection Techniques Of Synthetic Aperture Radar Images

Synthetic Aperture Radar (SAR) is the kind of initiative radar system which works on microwave bandwidth with the coherent way. SAR has the power of imaging for larger area under all weather condition and without daylight. This made it extensively used in many fields, such as military, in which target detection and recognition technique has been researched and developed. Target detection techniques based on SAR images is one of the most important techniques in the whole Auto Target Recognition (ATR) system. This dissertation focuses on the research of target detection techniques based on SAR images.The analysis of the statistic characteristics of SAR image is essential for the SAR image analysis and processing. In generally, based on the principle of SAR imaging, the traditional methods analyze the statistic characteristics of SAR image through the theory and certain conditions. However, based on the performance criterions of statistic models, such as Kolmogorov-Smirnov test, Anderson-Darling test, this dissertation analyzed deeply the statistic characteristics of SAR image through the experiments. Target detection techniques based on Const False Alarm Ration (CFAR) include estimator for the intensity of clutter, selector of the distribution models and estimator for the parameters of the models. The algorithms of estimating for the intensity of clutter are presented, such as cell-average, Smallest-of, Greatest-of and Order Statistic. We can obtain the detection threshold through estimating the parameters of the certain distribution model Then, this dissertation analyzes the principle of CFAR detection algorithms detailedly, realizes all the algorithms and analyzes the detection results of real SAR images.On the basis of traditional CFAR, a novel intelligent CFAR detection algorithm based on region classification is proposed in this dissertation. Firstly, the clutter region of moving windows model is divided into four parts. Secondly, some appropriate sub-blocks are selected as real background region, which is used to estimate the parameters and calculate the threshold of detection. The experimental results show that the new method has a good detection performance not only in homogeneous clutter background but also in non-homogenous clutter background including multi-target environment and clutter edge environment. In pan of the analysis and evaluation of detection results, the evaluation criterions of target detection are presented. Based on the evaluation criterions, the new algorithm is shown to be effective; Moreover, the platform of SAR images processing is introduced by explaining the function and framework of this software and showing the interface demo.The research of this dissertation is supported by the national nature science fund, national hi-tech development plan (863) projects. Some research productions have been applied in the radar data processing module of the development of the universal remote sensing data processing software, which is the key project of the tenth-five years plan.