利用多尺度SVM-CRF模型的极化SAR图像建筑物提取
发布时间:2018-03-14 23:37
本文选题:极化合成孔径雷达 切入点:建筑物提取 出处:《遥感技术与应用》2017年03期 论文类型:期刊论文
【摘要】:极化SAR图像中建筑物相关特征的不充分利用将影响建筑物提取的有效性或引发错误。为解决该问题,提出了一种利用多尺度SVM-CRF模型的极化SAR图像建筑物提取方法。在图像最优分割的基础上,将基于像素的SVM-CRF模型扩展到面向对象的多尺度SVM-CRF模型,使之能同时有效地描述建筑物突出的"面状"特征及其层次、空间上下文相关性。同时,考虑对建筑物描述特征利用不充分所引起的类别模糊问题,使用随机森林算法实现多特征的选择,形成更有效的特征组合以优化SVM-CRF模型中的特征向量。采用Oberpfaffenhofen地区E-SAR数据进行了实验,定性和定量的结果验证了该方法的有效性和准确性。
[Abstract]:The inadequate use of building-related features in polarized SAR images will affect the effectiveness of building extraction or cause errors. In this paper, a building extraction method based on multi-scale SVM-CRF model for polarimetric SAR image is proposed. Based on the optimal segmentation of the image, the pixel based SVM-CRF model is extended to the object-oriented multi-scale SVM-CRF model. It can also effectively describe the prominent "surface" features of buildings and their hierarchy, spatial contextual relevance. At the same time, it considers the problem of category ambiguity caused by inadequate use of building description features. The stochastic forest algorithm is used to select multiple features, and a more effective feature combination is formed to optimize the feature vectors in the SVM-CRF model. The experiments are carried out using E-SAR data in the Oberpfaffenhofen region. The qualitative and quantitative results show that the method is effective and accurate.
【作者单位】: 新疆气象局气象服务中心;中国地质大学(武汉)信息工程学院;
【基金】:国家自然科学基金项目(41301477、41471355) 中国博士后科学基金面上项目(2012M521497)资助
【分类号】:P237;TN957.52
,
本文编号:1613489
本文链接:https://www.wllwen.com/kejilunwen/dizhicehuilunwen/1613489.html