基于l1范数主成分分析的极化SAR图像变化检测

为提高极化合成孔径雷达(synthetic aperture radar,SAR)图像变化检测算法的鲁棒性以及检测精度,提出基于范数主成分分析(linorm principal component analysis,l1-PCA)模型的变化检测算法。首先,采用基于Hotelling-Lawley复矩阵迹变化检测算子构造差异图;其次,采用l1-PCA模型获取差异图的变化信息,使得每个像素以一个特征向量来表示;最后,使用k-means算法对变化信息进行聚类,得到变化检测结果。该方法是一种非监督变化检测方法,相比于基于2范数的PCA检测方法,l1-PCA在特征提取方面具有更高的鲁棒性,并且可以进一步提高变化检测精度。基于RADARSAT-2卫星获取的3幅图像进行的实验结果表明,相较于其他两种典型算法,所提算法更加稳定,精确度更高。

Polarimetric SAR change detection with l1-norm principal component analysis

To improve the robustness and detection accuracy of the polarimetric synthetic aperture radar (SAR) image change detection, we propose a change detection method based on l1-norm principal component analysis (l1-PCA). We use the complex Hotelling-Lawley matrix trace change detection operator to create the difference map. Then, the l1-PCA is used to extract change information from the difference map. Every pixel of the difference map is represented by a feature vector. Finally,the change map is achieved by the k-means classification algorithm.This method is an unsupervised change detection method. Compared with l2-norm principal component analysis (l2-PCA), l1-PCA has higher robustness in feature extraction and can further improve the accuracy of change detection. Experimental results implemented on 3 RADARSAT-2 image datasets illustrate that the proposed method performs better than two typical comparable algorithms in stability and accuracy.