多特征联合的序贯鉴别方法去除SAR ATR中虚假RoIs

Sequential Discrimination with Multi-Features to Remove False ROIs in SAR ATR

在SAR图像机动目标自动识别过程中,因目标预筛选阶段采用次优的异常检测策略而产生大量虚假的感兴趣区域(Region of Interest,ROI),这些虚假ROIs很大程度上降低了目标识别的效率。该文提出一种基于多特征联合的序贯鉴别算法来去除虚假ROIs。该算法首先对ROI切片的目标特征做冗余性、鲁棒性和可分离性的定量分析,以选取互补性强、稳定好的最优特征,并按所选特征鉴别性能的优略进行排序,来构建序贯鉴别的观测矢量,然后利用各鉴别特征的统计模型和设定的虚警概率来计算各特征对应判决阈值,最后联合优选的多个特征进行序贯判决。文中利用X波段的MSTAR数据验证了本文的算法,并与二项式距离鉴别算法做性能比较。

Because the prescreening in the process of mobile target recognition in SAR imagery usually adopts anomaly detection, a suboptimum approach,many false ROIs are produced,which can reduce hardly the efficiency of the ATR. A new method based on se- quential discrimination with multi-features is proposed to remove these false ROIs in this paper. Firstly, the features of mobile target ROIs are quantitatively analyzed for their redundancy, robustness and separability, and the best features are selected to form the observing vec- tor for sequential discrimination and ordered by their discriminating performance. And then ,the threshold for each best feature is calcu- lated based on the probability of false alarm and its statistical model. Finally, these features are incorporated to do a sequential discrimi- nation. The performance of the above algorithm is validated by the X band MSTAR data, and is compared with the quadratic distance discriminating method.