摘要
针对在高分辨率SAR影像建筑物检测过程中,经典形态学交替滤波算法对影像平滑和抑噪时,存在随着结构元素尺寸增加细节信息丢失以及检测结果出现粘连和粗化的问题,该文提出一种结合NASFs滤波策略与形态学属性剖面的高分辨率SAR影像建筑物检测方法。首先采用变异系数从SAR影像分割出非同质区(建筑区),利用NASFs滤波策略对非同质区进行背景噪声滤除;然后对非同质区影像计算连通分量结构,构建描述建筑物特征的形态学属性剖面和形态学差分属性剖面,对差分剖面进行形态学重建以提取建筑物几何结构;最后通过属性阈值分割得到检测结果。利用北京市TerraSAR-X数据进行试验,结果表明:该方法在平滑和抑制噪声的同时,能较好地保留影像结构细节信息,改善了检测结果出现粘连和粗化的问题,具有较高准确率(T=94.54%)和较低虚警率(FA=4.46%)。
In the process of building detection from high-resolution SAR imagery,use of larger-sized structure elements will remove detailed information and lead to roughening and adhesion problems in suppressing speckle noise using classical alternating sequential filters.A method of combining the NASFs filter strategy with morphological attribute profiles for building detection from high-resolution SAR imagery is proposed in this paper.Firstly,NASFs strategy was used to suppress background noise and strong scattering in the heterogeneous regions(buildings)segmented by coefficient of variation in the SAR image.Secondly,connected components were computed from the filtered heterogeneous regions,and morphological attribute profiles and morphological differential attribute profiles were constructed.Next,differential attribute profiles were reconstructed to extract multiple attribute features.Finally,the detection results were obtained through attribute threshold segmentation.TerraSAR-X satellite image data were applied and results showed that the proposed method was suitable and accurate for building detection from high-resolution SAR imagery and produced higher true positive and lower false alarm rate,with a value of 94.54%and 4.46%,respectively.
作者
胡世明
余洁
谢东海
王彦兵
余莎莎
HU Shi-ming;YU Jie;XIE Dong-hai;WANG Yan-bing;YU Sha-sha(The Key Laboratory of 3-Dimensional Information Acquisition and Application,Ministry of Education,Capital Normal University,Beijing 100048;College of Resource Environment and Tourism of Capital Normal University, Beijing 100048;Beijing Key Laboratory of Resources Environment and Geographic Information System of Capital Normal University,Beijing 100048;The National Key Laboratory Training Institution of Urban Environment Process and Numerical Simulation of Capital Normal University,Beijing 100048,China)
出处
《地理与地理信息科学》
CSCD
北大核心
2018年第6期27-33,共7页
Geography and Geo-Information Science
基金
国家自然科学基金面上项目(41671417/D010702)
科技创新服务能力建设-基本科研业务费(科研类)资助项目(025185305000/191)
