基于流形学习的高分SAR图像建筑区提取方法

发布时间：2018-05-11 04:06

本文选题：高分SAR图像 + 流形学习　；参考：《国土资源遥感》2017年04期

【摘要】：高空间分辨率(简称"高分")SAR图像具有高维非线性特点,以高维空间蕴含的低维流形描述SAR图像,会更有利于目标识别。将流形学习应用到高维SAR目标识别的特征表达中,提出一种新的高分SAR图像建筑区提取方法。首先,对高分SAR图像进行预处理;然后,采用灰度共生矩阵(gray level co-occurrence matrix,GLCM)提取8种纹理特征,与灰度图像共同构建SAR图像的高维特征集;利用自适应邻域选择的邻域保持嵌入(adaptive neighborhood selection neighborhoods preserving embedding,ANSNPE)算法对高维特征集进行特征提取,提取出新的特征;最后,通过阈值分割及后处理提取建筑区,并进行精度评价。选择Terra SAR-X数据进行实验研究,结果表明,ANSNPE算法能够从高分SAR图像中有效提取建筑区,并具有较强的泛化能力;通过训练数据获得的投影矩阵可直接应用到新样本中,建筑区提取精度达85%以上。
[Abstract]:High spatial resolution (abbreviated as "high fraction") SAR images have the characteristics of high dimensional nonlinearity. It is more advantageous for target recognition to describe SAR images with low-dimensional manifolds contained in high-dimensional space. In this paper, manifold learning is applied to feature expression of high-dimensional SAR target recognition, and a new method of building area extraction for high-score SAR images is proposed. Firstly, the high score SAR image is preprocessed, then the gray level co-occurrence matrix GLCM) is used to extract eight texture features, and then the high Vitert solicitation of the SAR image is constructed together with the gray level image. The adaptive neighborhood selection algorithm is used to extract new features from high Viterbi solicitation by using adaptive neighborhood selection neighborhoods preserving embedding ANSNPE algorithm. Finally, the building area is extracted by threshold segmentation and post-processing, and the accuracy is evaluated. Terra SAR-X data are selected for experimental study. The results show that ANSNPE algorithm can effectively extract the building area from high score SAR images, and has a strong generalization ability, and the projection matrix obtained from the training data can be directly applied to new samples. The precision of construction area is more than 85%.
【作者单位】：中国科学院遥感与数字地球研究所;中国科学院大学;
【基金】：国家自然科学基金项目“高分辨率SAR图像典型地物目标样本特征提取和识别研究”(编号:61372189)资助
【分类号】：TN957.52

【相似文献】