一种新的SAR图像目标检测像素聚类算法  

A New Clustering Method for Target Pixels Detected from SAR Images

作　　者：宋文青 徐玮[1] 相飞[1] SONG Wenqing;XU Wei;XIANG Fei(Xi'an Electronic Engineering Research Institute, Xi'an 710100)

机构地区：[1]西安电子工程研究所,西安710100

出　　处：《火控雷达技术》2020年第2期1-5,11,共6页Fire Control Radar Technology

基　　金：国家自然科学基金青年项目(61802335);河北省自然科学基金青年科学基金(F201803096)。

摘　　要：针对SAR图像目标检测任务中多目标区域目标像素聚类问题,本文提出一种结合密度聚类和区域合并的目标像素聚类算法。算法首先采用基于密度的聚类方法对图像中目标像素进行初步聚类,生成目标局部结构层的聚类区域;然后再利用基于最远距离约束的最近距离区域合并方法对初步聚类区域进行合并,进而获得目标层的聚类区域。最后根据聚类区域中像素个数特征对虚警目标进行一定地剔除,生成最终的SAR图像目标层的检测结果。实测SAR图像数据实验表明,本文算法能够有效地解决多目标场景中的目标像素聚类问题。In order to solve the target pixels clustering in multiple targets regions for SAR target detection tasks,a new method was proposed in this paper,which combines a density-based clustering method and a region merging method.Firstly,the density-based clustering method is used to cluster the pixels into regions which are parts of targets.Secondly,the region merging method based on nearest distance constrained by farthest distance is adopted to merge the local part regions into integrated targets.Finally,the false alarm targets are eliminated based on the number of pixels in the cluster area.Experiments with real SAR images show that the proposed method can effectively solve the pixel clustering problem in multiple targets regions.

关 键 词：SAR图像 目标检测 目标像素聚类 密度聚类 区域合并 

分 类 号：TN958.94[电子电信—信号与信息处理]