基于模糊聚类视区划分的SAR目标识别方法

 现有基于模板匹配的SAR目标识别技术,多通过姿态遍历来构建和存储基础模板库.为降低计算消耗和存储开销,借鉴计算机视觉中视区概念,提出了一种基于非均匀视区划分的模板库精简方法.结合关键特征矢量,基于Gustafson-Kessel(GK)算法对视区作模糊聚类,以识别概率最优控制视区划分策略并提炼原型模板.采用典型舰船目标的SAR仿真图像集,验证了方法在精简模板库、实现高效SAR自动目标识别方面具有可行性.

A SAR Target Recognition Method Based on View-Aspects Partitioned by Fuzzy Clustering

The template-based classification,which is considered as a conventional synthetic-aperture radar(SAR) automatic target recognition(ATR) approach,needs a stored set of templates of the targets at a reasonable number of different orientations.However,it results in the great consumption of computation and storage.Depending on the term,View-aspect,in Computer Vision and thus an approach is proposed to construct simplified prototype templates to replace the conventional library.The prototype templates are extracted by the Gustafson-Kessel(GK) algorithm with target key features according to the rules of optimizing recognition rate.Experiment was runned on simulating SAR data sets of typical ship targets and the results demonstrate that the approach can not only simplify the templates,but also improve the recognition performance effectively.