首页 >  , Vol. , Issue () : 1993-2002

摘要

引用本文:

DOI:

10.11834/jrs.20209366

收稿日期:

2019-10-10

修改日期:

2020-01-20

PDF Free   EndNote   BibTeX
基于栈式稀疏自编码网络的多时相全极化SAR散射特征降维
摘要:

利用极化合成孔径雷达(PolSAR)能够实现地物的识别和分类,而多时相全极化SAR可以获取地物的更多散射特征,提升地物识别精度,但高维散射特征的引入会带来严重的维数灾难问题。为了实现对高维散射特征的有效降维,本文提出了一种基于栈式稀疏自编码网络S-SAE(Stacked Sparse Autoencoder)的多时相PolSAR散射特征降维方法。该方法首先对PolSAR数据进行极化目标分解以获取高维散射特征,再使用S-SAE对获取的多维特征进行降维处理,其中S-SAE降维方法首先采用无监督训练方式进行逐层贪婪训练,再结合Sigmod分类器,利用监督训练的方式对S-SAE进行参数优化,实现高维特征的有效降维,最后以降维后的特征作为支持向量机(SVM)和卷积神经网络(CNN)分类器的输入,实现地物分类。通过仿真和实测的两组多时相Sentinel-1数据处理结果表明,双隐层的S-SAE降维方法在各分类器上均取得了最优的降维效果;对比各降维方法在SVM分类器上的分类精度,S-SAE较于局部线性嵌入(LLE)与主成分分析(PCA)降维方法,总体分类精度分别至少提升了9%和14%;在CNN分类器上,S-SAE较于LLE与PCA降维方法,总体分类精度分别至少提升了7%和9%。

Scattering Feature Dimension Reduction of Multi-temporal Fully PolSAR Image Based on Stack Sparse Autoencoder
Abstract:

Polarimetric Synthetic Aperture Radar (PolSAR) has been proved to recognize and classify various ground objects, and multi-temporal fully PolSAR can acquire more scattering features to improve the accuracy of recognition and classification. However, the decomposed scattering features with high dimensionality can cause serious problems of ‘curse of dimensionality’. In order to effectively reduce the dimensionality of high-dimensional scattering features, this paper proposes a multi-temporal PolSAR scattering feature dimension reduction method based on Stacked Sparse Autoencoder (S-SAE). The method firstly decomposes the PolarSAR data to obtain high-dimensional scattering features, and then adopts S-SAE to reduce the dimensionality of the acquired multi-dimensional features. For S-SAE construction, unsupervised layer-by-layer greedy training is performed to optimize main parameters. Combined with the Sigmod classifier, the parameters of S-SAE are finely tuned by supervised training to achieve effective dimension reduction of high-dimensional features. Finally, in order to evaluate the performances of feature dimension reduction, the reduced low-dimensional features are taken as the input of SVM and CNN classifiers. The performances of the proposed method are validated with two data sets of multi temporal simulated and real Sentinel-1 data. The experimental results show that the S-SAE method with two hidden layers achieves the best performance of feature dimension reduction. Compared with traditional LLE and PCA dimension reduction methods, for SVM classifier, the overall classification accuracy with S-SAE is raised by at least 9% and 14%. For CNN classifier, the overall classification accuracy with S-SAE is at least 7% and 9% higher than that of LLE and PCA, respectively.

本文暂时没有被引用!

欢迎关注学报微信

遥感学报交流群 分享按钮