Research On Detection Methods Of Vehicle Targets From Sar Image Based On Statistical Model

SAR, a sensor with many advantages, has tremendous application potentials in the field of military reconnaissance and has already been widely studied by many countries all over the world. According to SAR interpretion on moving target, an Automatic Target Recogintion(ATR) system can be built, which has been a major subject in SAR applications. In fact, because of variabilities in the target filed scene, it can not meet the demands for robust target recoginition with the help of current technology. Thus, in view of practicability, it is a hotspot subject on how to detect targets quickly.Detection, is the first step of an ATR system, its performance has great influence on the successive processing. According to current literatures, the target detection algorithm based on noise statistical characteristic analysis is the most robust and practical one with the fastest developing speed among all algorithms. Therefore, the clutter statistical model of SAR image is thoroughly discussed at first in this paper. Then the CFAR target detection algorithm is studied systematically. In order to improve the performance of target detection, an algorithm based on the statistical model using features is brought forward. The major contents in this paper include: Firstly, as the theory foundation, the SAR clutter statistical model is studied in depth. Combined with the different clutter scatter mechanism, the clutter is divided into three classes: the homogeneous clutter regions, the heterogeneous clutter regions and the extremely heterogeneous clutter regions. The statistical models of each class are presented respectively. Both the theoretical deduction and the parameter estimation of each statistical model are given in the paper. Besides, in order to test the degree of fitness between the different models and the clutter data, different goodness-of–fit tests are presented and compared.Secondly, we systematically study the CFAR target detection algorithm with the guide of the conclusion gained from the research of the clutter statistical models. The concrete forms of CFAR algorithm for different clutter distributions are deduced particularly. It concludes the relationship between the detection threshold and the false alarm rate and the estimation of the clutter statistic measure, and so on. Several typical CFAR detectors are given, such as, CA-CFAR detector, OS-CFAR detector, etc. At last, the contrasted experiments are given in order to provide the basis of choosing the appropriate algorithm to meet different needs for successive process.Finally, in order to solve the problem that the CFAR detection has higher false alarm rate in the strong clutter environment and to improve the detection performance, a detection scheme based on the scatistical model using features is brought forward. We introduce and improve two detection methods: the extended fractal (EF) detection and the quadratic gamma detection. The EF detection can overcome the limitation of CFAR detection and can detect the targets in strong clutter environments. However, the EF feature is symmetric in the sense that a deep target-sized shadow can initiate a high response as well as bright target-sized objects. So, a new method by fusing the EF feature detection with the CFAR detection is provided. The experiments proved...