测绘学报 ›› 2018, Vol. 47 ›› Issue (12): 1609-1620.doi: 10.11947/j.AGCS.2018.20170551

• 摄影测量与遥感 • 上一篇    下一篇

利用目标分解特征的全极化SAR海冰分类

赵泉华, 郭世波, 李晓丽, 李玉   

  1. 辽宁工程技术大学测绘与地理科学学院遥感科学与应用研究所, 辽宁 阜新 123000
  • 收稿日期:2017-09-26 修回日期:2018-01-19 出版日期:2018-12-20 发布日期:2018-12-24
  • 作者简介:赵泉华(1978-),女,博士,教授,研究方向为遥感图像建模与分析、解析几何在遥感图像处理中的应用。E-mail:zqhlby@163.com
  • 基金资助:
    国家自然科学基金(41271435;41301479);辽宁省自然科学基金(2015020090)

Polarimetric SAR Sea Ice Classification Based on Target Decompositional Features

ZHAO Quanhua, GUO Shibo, LI Xiaoli, LI Yu   

  1. Institute for Remote Sensing Science and Application, School of Geomatics, Liaoning Technical University, Fuxin 123000, China
  • Received:2017-09-26 Revised:2018-01-19 Online:2018-12-20 Published:2018-12-24
  • Supported by:
    The National Natural Science Foundation of China (Nos. 41271435;41301479);The Science and Technology Program of Liaoning Province (No. 2015020090)

摘要: 特征提取及其选择是SAR海冰分类的重要步骤之一。在众多特征中选取有效特征,进而构建表达地物类型的特征空间是提高分类精度的关键。为此,本文提出一种基于目标分解特征的全极化SAR海冰分类算法。首先,对全极化SAR数据进行多视化处理及滤波操作,生成相干矩阵;其次,对相干矩阵进行目标分解,并针对分解结果提取散射特征参数,进而构建特征空间;再次,通过对所提取的特征进行统计相关性分析,并对高相关特征采用PCA降维,以优化特征组合;最后,设计BP神经网络分类器,并将所得的优化特征矢量作为输入,海冰类别为输出,实现海冰分类。本文以格陵兰中部海域作为研究试验区域,采用L波段ALOS PALSAR全极化数据。通过对本文算法与对比算法的分类结果进行定性定量分析,可以得出本文所选取的特征对海冰识别较好。此外,通过对利用各个不同特征海冰分类结果的性能分析,可以得出基于散射模型的目标分解比基于特征值的H/α/A分解更有助于海冰分类。

关键词: 海冰分类, 目标分解, 特征提取, 极化SAR

Abstract: Feature extraction and its selection are one of the most important steps in the SAR sea ice classification. The key to improve the classification accuracy is to select effective features and to construct the feature space that effectively expresses the type of ground objects. For this purpose, a full polarimetric SAR sea ice classification algorithm based on target decomposition features is proposed in this paper. First of all, multilook process and filter operation are preformed to full-pol SAR data and result in coherency matrix. Secondly, in order to construct the feature space, target decomposition on coherency matrix is employed to extract related scattering feature parameters. Thirdly, after analysis of statistical correlation about extracting features, PCA feature reduction operation is carried out on those higher relevant features for the purpose of optimizing the combination of features. Finally, a BP neural network-based classification algorithm is designed to classify sea ice, and the optimization of the feature vector as input layer, the class of sea ice as output layer. In experiment, the central Greenland area is regard as the research area and L-band ALOS PALSAR full polarimetric data are utilized as experimental data. Through the qualitative and quantitative analysis for the proposed and comparing algorithms, it can be found that the feature space built up can efficiently distinguish various sea ices. Furthermore, by analyzing the performance of sea ice classification results with different feature combination, we can conclude that the features of the target decomposition based on scattering model can provide a better capability to identify water and sea ices compared to H/α/A decomposition based on eigenvalue.

Key words: sea ice classification, target decomposition, feature extraction, PolSAR

中图分类号: